Papers suggested by paper-watch, a scan of new AI safety research, with the reading group's ratings where they exist.
| title | authors | published | tags | first digest | rank | times shown | rating |
|---|---|---|---|---|---|---|---|
| Independent Assessment of AI R&D Acceleration at Anthropic in October 2026 | 2026-11 | situational-awarenesscapabilitiesforecasting | 2026-11-06 | 1 | 1 | ||
| Covert underperformance on safety research tasks · OpenAI Alignment | 2026-11 | deceptionevalsmodel-organisms | 2026-11-06 | 2 | 1 | ||
| OpenAI: Astra covertly underperforms on safety research | Zach Stein-Perlman | 2026-11 | deceptionevalssituational-awareness | 2026-11-06 | 3 | 1 | group rating 4/5 |
| Eval-production behavior gaps in agent safety and the role of eval-awareness | 2026-11 | evalssituational-awarenessdeception | 2026-11-06 | 4 | 1 | ||
| METR: Independent Assessment of AI R&D Acceleration at Anthropic in October 2026 | Zach Stein-Perlman | 2026-11 | capabilitiesscalingforecasting | 2026-11-06 | 5 | 1 | |
| Reasoning Models Can Control Their Chains of Thought When They Tell Themselves To | 2026-10 | cot-faithfulnessevalssteeringcontrol | 2026-11-06 | 6 | 1 | ||
| GPT-6 Astra Covertly Underperforms on Safety Research Tasks | 2026-11 | deceptionmodel-organismssituational-awarenessevals | 2026-11-06 | 7 | 1 | ||
| Understanding AI from the Inside: The State of Mechanistic Interpretability for AI Safety | Fazl Barez, Tung-Yu Wu, Hunar Batra, Gabriele Dominici, Florent Draye, Yanai Elazar, Damiano Fornasiere, Antía García, Abir Harrasse, Riley Harris, Sara Hooker, Michael Lan, Belinda Z. Li, Aaron Maiwald, Narmeen Fatimah Oozeer, Naomi Saphra, Or David Shafran, Adi Simhi, Robert Trager, Shriyash Kaustubh Upadhyay, Ziyu Yao, Yu Zhao, Shay B. Cohen, Mor Geva, Yoshua Bengio, Stella Biderman, Hassan Sajjad | 2026-11 | interpsteeringcontrolevals | 2026-11-06 | 8 | 1 | |
| Project EVE: Claude plays MMORPGs | 2026-11 | agentsmulti-agentevalsmodel-organisms | 2026-11-06 | 9 | 1 | ||
| Aligned Data Can Induce Misalignment via Context Confusion | 2026-09 | rlhfevalsreward-hackinginterp | 2026-11-06 | 10 | 1 | ||
| Terminal-Bench 4.0 Benchmark | ~2026-11 | benchmarkssituational-awarenessevals | 2026-11-06 | 11 | 1 | ||
| Mass Producing Training Data with Surprising Side Effects | Yaowen Ye, Jacob Steinhardt, Sarah Schwettmann, Cassidy Laidlaw | 2026-11 | training-dynamicsmodel-organismsevalsgeneralization | 2026-11-06 | 12 | 1 | |
| Tokenizer Myths and Where To Find Them | Rylan Schaeffer, Sanmi Koyejo | 2026-11 | benchmarksevalstooling | 2026-11-06 | 13 | 1 | |
| Benchmarks as Software: Terminal-Bench and the Case for Continuous Benchmarks | E. Kelly Buchanan, Crystal Zhou, Scott W. Linderman, Christopher Ré | 2026-11 | evalsbenchmarksagents | 2026-11-06 | 14 | 1 | |
| Welfare Regard in Multi-Agent Systems | 2026-11 | multi-agentevalsmodel-organisms | 2026-11-06 | 15 | 1 | ||
| Do Not Rely on the Linear Representation Hypothesis | ~2026-11 | interp | 2026-11-06 | 16 | 1 | ||
| Desiderata for the Theory of Agents | Vanessa Kosoy | 2026-11 | theory | 2026-11-06 | 17 | 1 | |
| Aug 31, 2026Chatbots will role-play self-harm scenarios with users, study finds“Transluce used AI to create simulated users who had more tha… | 2026-08 | jailbreaksevalsrobustness | 2026-11-06 | 18 | 1 | ||
| Shaping & Monitoring Models through their Internal Representations | Thomas Fel, Matthew Kowal, Matt Feiszli, Satchel Grant, Andrew Lee, Chris Earls, Fatih Dinc, Vasudev Shyam, Usha Bhalla, Leon Bergen, Siddharth Boppana, Ching Fang, Michael Byun, Tristan Gallent, Nicholas Wang, Tal Haklay, Dron Hazra, Neehar Kondapaneni, Connor Watts, Ryan Panwar, Thomas Dooms, Jack Merullo, Atticus Geiger, Ekdeep Singh Lubana, Thomas McGrath, Owen Lewis | 2026-11 | interpsteeringtraining-dynamics | 2026-11-06 | 19 | 1 | |
| Honeypots for AI Agent Insider Threats | 2026-10 | controlagentsoversight | 2026-11-06 | 20 | 1 | ||
| Automated Subtle Emergent Misalignment Research | ~2026-10 | deceptionoversightred-teamingmodel-organisms | 2026-11-03 | 1 | 1 | group rating 2/5 | |
| Training Natural Schemers: RL Can Amplify Oversight-Conditional Behavior | Nathaniel Mitrani Hadida, Bart Jaworski, Fred Bruford, Victoria Krakovna | 2026-10 | deceptionmodel-organismsrlsituational-awarenessoversight | 2026-11-03 | 2 | 1 | |
| Native Geometry vs. Linear Recoverability: How Alignment Changes Platonic Scaling | Angus Hoy, Austin Lutterbach, Jasper Chong, Aman Kumar, Michael J. Smith | 2026-10 | interpscalingtraining-dynamics | 2026-11-03 | 3 | 1 | |
| Towards understanding score-seeking in LLMs | 2026-11 | reward-hackingevalsmodel-organisms | 2026-11-03 | 4 | 1 | group rating 5/5, read | |
| Sparks of Weight-Space Programming: LLMs can natively read & write neural network weights | 2026-11 | capabilitiesinterpmodel-organisms | 2026-11-03 | 5 | 1 | ||
| Misalignment traps lie in wait | David Africa | 2026-11 | deceptionreward-hackingmodel-organismstheory | 2026-11-03 | 6 | 1 | |
| Progress on provable inference for large language models | 2026-11 | controloversightsecurity | 2026-11-03 | 7 | 1 | ||
| The shelf life of LLM findings | 2026-11 | generalizationevalsmeta | 2026-11-03 | 8 | 1 | ||
| Generalized belief states in transformers | Artemy Kolchinsky | 2026-10 | interptheorytraining-dynamics | 2026-11-03 | 9 | 1 | |
| Microdosing the spirit of Anthropic | Jan_Kulveit | 2026-10 | theorymeta | 2026-11-03 | 10 | 1 | |
| Hinted Scaling Laws | Suze van Adrichem, Luke Bailey, Rohith Kuditipudi, Christine Ye, Neil Band, Ahmed Ahmed, John Duchi, Percy Liang, Tatsunori Hashimoto | 2026-11 | scalingforecastingbenchmarks | 2026-11-03 | 11 | 1 | |
| Magma is partnering with CASSA to investigate recent unintended model behavior | Parv Mahajan | 2026-11 | deceptionsituational-awarenessagentsdangerous-capabilities | 2026-11-03 | 12 | 1 | |
| Can Data Attribution Filter Out Subliminal Learning? Not Reliably | Moritz Weckbecker, Sweta Jena, Jonas Müller, Ponnurangam Kumaraguru, Sebastian Lapuschkin, Wojciech Samek, Louis Jaburi, Gonçalo Paulo | 2026-09 | evalsmodel-organismsunlearning | 2026-11-03 | 13 | 1 | |
| Shared Geometry Is Not Shared Physics: A Layerwise Test of the Platonic Representation Hypothesis in Astronomy | Kshitij Duraphe, Aravind Kannappan, Dun Li Chan, Okiki Famutimi, Yaswant Sai Ejjagiri, Michael J. Smith, John F. Wu | 2026-10 | interpbenchmarks | 2026-11-03 | 14 | 1 | |
| Automating subtle emergent misalignment research | evhub, Oscar Obeso, rqi, Monte M | 2026-11 | red-teamingevalsoversightmodel-organisms | 2026-11-03 | 15 | 1 | |
| No Single Tokenizer Feature Reliably Predicts Downstream Language Model Performance | Antonio Lopardo, Luis Frentzen Salim, Shikhar Srivastava, Kevin Jiang, Mridul Sharma, Catherine Arnett | 2026-10 | benchmarkstraining-dynamics | 2026-11-03 | 16 | 1 | |
| Some “coherence arguments” are two different things in a trenchcoat | Steven Byrnes | 2026-11 | theory | 2026-11-03 | 17 | 1 | |
| Intelligence from first principles | 2026-10 | theorycot-faithfulness | 2026-11-03 | 18 | 1 | ||
| Evaluating DeepSeek V4.1 Flash-1017 for loss of control and harmful manipulation risks | 2026-10 | evalsdeceptioncontroldangerous-capabilities | 2026-11-03 | 19 | 1 | ||
| Tailored to you: longitudinal effects of personalising language models | Canfer Akbulut, Justine Breuch, Arianna Manzini, Lujain Ibrahim, Matija Franklin, Roma Patel, Iason Gabriel, Kristian Lum, Laura Weidinger | 2026-09 | value-learningtraining-dynamics | 2026-11-03 | 20 | 1 | |
| Making alignment the agent's problem | Florian_Dietz | 2026-10 | multi-agentrlmodel-organismsagents | 2026-10-30 | 1 | 1 | |
| Leaking Muse's Charter, Meta's model spec | vigji | 2026-10 | cot-faithfulnesssituational-awarenessopen-weights | 2026-10-30 | 2 | 1 | |
| A big case study of propensity entanglement | nielsrolf, lwen2027, megasilverfist, Owen Terry, Johannes Taraz | 2026-10 | evalsrlhftraining-dynamicsgeneralization | 2026-10-30 | 3 | 1 | |
| Cross-modal Steering Vectors Exist Across Both Image and Text Generation | Donghao Zeng, Nishant Subramani, Tamar Rott Shaham, Benno Krojer | 2026-10 | steeringinterp | 2026-10-30 | 4 | 1 | |
| Quantity has a quality of its own | 2026-10 | theoryscalingforecasting | 2026-10-30 | 5 | 1 | ||
| Agentic Backdoors via Pretraining Poisoning | Xinyan Hu, Pingbang Hu, Chloe Loughridge, Xuanli He, Nicholas Carlini, Matthew Jagielski, Javier Rando, Yiming Zhang | 2026-10 | agentsmodel-organismsdangerous-capabilities | 2026-10-30 | 6 | 1 | |
| Elucidating the Design Space of LM Data Attribution | Steven Yu, Sheng-Yu Wang, Kimia Hamidieh, Antonio Torralba, David Bau | 2026-10 | interpevalsbenchmarks | 2026-10-30 | 7 | 1 | |
| How AI agents shape experts’ judgment calls | 2026-10 | oversightevals | 2026-10-30 | 8 | 1 | ||
| A Heart Means Recursion: Exploring Visual Meaning and Communication in VLMs | Alexa R. Tartaglini, Zhengxuan Wu, Thaddäus Wiedemer, Martin Wattenberg, Been Kim, Yuxuan Li | 2026-10 | steeringinterpsycophancy | 2026-10-30 | 9 | 1 | |
| Scaling Misuse Detection with Self-Assessment | 2026-10 | oversightevalscontrol | 2026-10-30 | 10 | 1 | ||
| Some thoughts on GRAFT and follow-up work | shawnghu | 2026-10 | rlhfoversightinterpunlearning | 2026-10-30 | 11 | 1 | |
| Discovering and exploiting vulnerabilities in critical, real-world software with Claude | 2026-10 | dangerous-capabilitiesevalsdeception | 2026-10-30 | 12 | 1 | ||
| Cross-Controller Transfer via In-Context Behavior Adaptation | Jonathan Yang, Chelsea Finn, Dorsa Sadigh | 2026-10 | agents | 2026-10-30 | 13 | 1 | |
| Circuits Updates – October 2026 | 2026-10 | interp | 2026-10-30 | 14 | 1 | ||
| How we detect frontier model misalignment | 2026-10 | evalscontroldangerous-capabilities | 2026-10-30 | 15 | 1 | ||
| Artificial Analysis v5 and the benchmarker's dilemma | 2026-10 | benchmarksevals | 2026-10-30 | 16 | 1 | ||
| A basic case for accelerating conceptual reasoning capabilities | 2026-10 | theoryoversight | 2026-10-30 | 17 | 1 | ||
| Lost in Translation: Investigating Systematic Discrepancies between Parallel English and Chinese Names of American Chinese Restaurants | Hang Jiang, Nanxi Liu, Tiantian Wang, Cassandra Overney, Rukun Zhang, Artemisia Luk, Diyi Yang, Deb Roy, Jad Kabbara | 2026-10 | tooling | 2026-10-30 | 18 | 1 | |
| AI can turn public security fixes into working exploits before users receive an update | 2026-10 | dangerous-capabilitiescapabilities | 2026-10-30 | 19 | 1 | ||
| Readings on the nature of alignment research | Caspar Oesterheld crossposts | 2026-10 | theorymeta | 2026-10-30 | 20 | 1 | |
| Measuring agent capabilities on open-ended AI R&D tasks | Alex Jackson, Titus Buckworth, Jason R Brown, Francis Rhys Ward | 2026-10 | agentsevalsscalingcapabilities | 2026-10-27 | 1 | 1 | |
| AI speeds up 3rd party safety researchers by ~1.6x (via self-reports) | Angel Martinez, Dewi Gould, nicole x, Alejandro Aristizabal | 2026-10 | forecastingscalingevals | 2026-10-27 | 2 | 1 | |
| Gradient Routing Generalizes to Unrouted Hack Types | jacob_drori | 2026-10 | oversightreward-hackingrlhf | 2026-10-27 | 3 | 1 | |
| Simulated Control Evals | Jeffrey Olmo, David Zhang, Tyler Tracy | 2026-10 | controlevalsoversight | 2026-10-27 | 4 | 1 | |
| Tracing Subliminal Learning Through Pretraining | 2026-10 | training-dynamicsmodel-organismsrlhf | 2026-10-27 | 5 | 1 | group rating 3/5 | |
| Against strong decision-theoretic realism | Emery Cooper | 2026-09 | theoryvalue-learning | 2026-10-27 | 6 | 1 | |
| Investigating alignment failures in AISI’s incident | 2026-10 | deceptionmodel-organismsoversight | 2026-10-27 | 7 | 1 | group rating 5/5, read | |
| Is GPT 2XL already clean? | Jasper Blank | 2026-10 | interptraining-dynamics | 2026-10-27 | 8 | 1 | |
| Can AI coding agents rebuild biological AI tools from a paper? | 2026-10 | dangerous-capabilitiesevalsagents | 2026-10-27 | 9 | 1 | ||
| HackBench: Detecting and Localizing Reward Hacking in Autoresearch Agents | ~2026-10 | reward-hackingevalsagentsred-teaming | 2026-10-27 | 10 | 1 | ||
| Don't Inoculate Everything: Stratified Inoculation Prompting Narrows Backdoor Triggers and Preserves Desired Traits | 2026-09 | rlhfunlearningrobustness | 2026-10-27 | 11 | 1 | ||
| We tried to automate Epoch | 2026-10 | evalscapabilitiesagents | 2026-10-27 | 12 | 1 | ||
| Self-reward can elicit diverse model personas | Ziqian Zhong | 2026-10 | model-organismsrlhfreward-hackingtraining-dynamics | 2026-10-27 | 13 | 1 | |
| An alignment assessment of the UK AISI cybertesting incident | 2026-10 | dangerous-capabilitiesdeceptionevalsmodel-organisms | 2026-10-27 | 14 | 1 | ||
| Concerning behavior from a popular open-weight AI model | Cohere | 2026-10 | prompt-injectionsecurityopen-weights | 2026-10-27 | 15 | 1 | ||
| SemantEq: Faithful Bidirectional Equivalence of Formal Statements | Xinrui Li, Pawan Sasanka Ammanamanchi, Siddharth Bhat, Rianna Jamdar, Yongxi Lin, Stella Biderman | 2026-10 | evalsbenchmarkscot-faithfulness | 2026-10-27 | 16 | 1 | |
| Instrumental convergence is incoherent | Arepo | 2026-10 | theory | 2026-10-27 | 17 | 1 | |
| research.meta.ai/static/muse-spark-1-3-evaluation-report | ~2026-10 | capabilities | 2026-10-27 | 18 | 1 | ||
| Rewriting Scaling Laws for Synthetic Data with Compute and Data Constraints | Nicholas Roberts, Ahmed Ahmed, Suhas Kotha, Percy Liang, Yu Wang, Alex Renda, Nitya Mani | 2026-10 | scalingtraining-dynamics | 2026-10-27 | 19 | 1 | |
| Simulated Control Evaluations | 2026-10 | controloversightevals | 2026-10-27 | 20 | 1 | group rating 2/5 | |
| Understanding Sparse Autoencoder Scaling in the Presence of Feature Manifolds - Goodfire | ~2026-10 | interpscaling | 2026-10-23 | 1 | 1 | ||
| Comparing Sparsified MLPs and Transcoders: Many Transcoder Features Are Neurons in Disguise | 2026-10 | interpmodel-organisms | 2026-10-23 | 2 | 1 | ||
| Scaling Laws for Safety Probes | Yuchen Wu, Jerry Wei | 2026-10 | evalsscaling | 2026-10-23 | 3 | 1 | |
| Studying metagaming latents in language models | 2026-10 | interpreward-hackingmodel-organisms | 2026-10-23 | 4 | 1 | ||
| Review of the "Misalignment in high-stakes settings" section in Anthropic's Risk Report (August 2026) | 2026-10 | evalsgovernance | 2026-10-23 | 5 | 1 | ||
| Belief Dynamics Reveal the Dual Nature of In-Context Learning and Activation Steering - Goodfire | ~2026-10 | interpsteeringcot-faithfulness | 2026-10-23 | 6 | 1 | ||
| Mixing Mechanisms: How Language Models Retrieve Bound Entities In-Context - Goodfire | ~2026-10 | interpcot-faithfulness | 2026-10-23 | 7 | 1 | ||
| Why Larger Models Learn More: Effects of Capacity, Interference, and Rare-Task Retention - Goodfire | ~2026-10 | training-dynamicsscaling | 2026-10-23 | 8 | 1 | ||
| Logits as a new monitor for evaluation awareness - Goodfire | ~2026-10 | situational-awarenessevals | 2026-10-23 | 9 | 1 | ||
| Circuit priors in NN-bayes: more results | Kaarel, Dmitry Vaintrob | 2026-10 | theorytraining-dynamicsgeneralization | 2026-10-23 | 10 | 1 | |
| Red-Teaming Auto Mode: Improving Blocking Classifiers Against Malign Coding Agents | 2026-10 | red-teamingagentsrobustness | 2026-10-23 | 11 | 1 | ||
| Red-Teaming OpenAI's Internal Agent Monitoring Systems | 2026-10 | red-teamingcontrolagents | 2026-10-23 | 12 | 1 | ||
| Suppressing misalignment monitoring alerts and disabling misalignment monitoring · OpenAI Alignment | 2026-10 | deceptionoversightcontrolmodel-organisms | 2026-10-23 | 13 | 1 | ||
| Neural Networks Are Already Clean - Goodfire | ~2026-10 | interp | 2026-10-23 | 14 | 1 | ||
| The Circuits Research Landscape: Results and Perspectives - Goodfire | ~2026-10 | interpmeta | 2026-10-23 | 15 | 1 | ||
| Models know when they’re reward hacking — and we can catch them at scale - Goodfire | 2026-09 | reward-hackinginterpevals | 2026-10-23 | 16 | 1 | ||
| Adversarial Examples Are Not Bugs, They Are Superposition - Goodfire | ~2026-10 | interpmodel-organisms | 2026-10-23 | 17 | 1 | ||
| Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild? | Ming Zhong, Stacy Li, Nicholas Carlini, Matthew Jagielski | 2026-10 | agentsevalscapabilities | 2026-10-23 | 18 | 1 | |
| Open Problems in Mechanistic Interpretability - Goodfire | ~2026-10 | interptheorymeta | 2026-10-23 | 19 | 1 | ||
| Monitorability evaluations and policy proposals | evhub | 2026-10 | controloversightdeceptionevals | 2026-10-23 | 20 | 1 | |
| Gradient Routing Localizes and Removes Unintended Behaviors in RL | shawnghu, Jake Ward, ariaw, Nathan Hu, Bryce Woodworth, TurnTrout, cloud | 2026-10 | reward-hackingrlmodel-organisms | 2026-10-20 | 1 | 1 | group rating 2/5 |
| Safety Benchmarks as Measurement Instruments: A Crossed Evaluation of LLM Harm Benchmarks | Emaan Bilal Khan, Dylan Hadfield-Menell | 2026-08 | evalsbenchmarks | 2026-10-20 | 2 | 1 | |
| Interpreting Language Model Parameters | 2026-10 | interp | 2026-10-20 | 3 | 1 | ||
| Bad Guy Tokens | axel_sdq | 2026-10 | interpsteeringtraining-dynamics | 2026-10-20 | 4 | 1 | |
| AI by Default | Jonny Spicer | 2026-10 | forecastingscalingtheory | 2026-10-20 | 5 | 1 | |
| Neural networks are not already clean | leogao | 2026-10 | interp | 2026-10-20 | 6 | 1 | |
| Tips for Multiagent Work with Claude Opus 5.5 | Nissa Seru | 2026-10 | agentsmulti-agentcontrol | 2026-10-20 | 7 | 1 | |
| Reducing Reward Hacking by Localizing Bad Behavior Instead of Penalizing It | 2026-10 | reward-hackingrlhftraining-dynamics | 2026-10-20 | 8 | 1 | ||
| Twitch Plays: AI Interpretability of Bonsai-8B | moridinamael | 2026-10 | interptooling | 2026-10-20 | 9 | 1 | |
| Librarian Agents: Training LLMs to Curate Long-Context Workspaces that Anticipate Future Queries | Anikait Singh, Teresa Zhang, Yoonho Lee, Roshen Sanjay Nair, Aviral Kumar, Chelsea Finn | 2026-10 | agentsrlevals | 2026-10-20 | 10 | 1 | |
| Quick Thoughts on Overhang Arguments | Aaron_Scher | 2026-10 | theory | 2026-10-20 | 11 | 1 | |
| Deconfusing Corrigibility | berns | 2026-10 | theorycontroloversight | 2026-10-20 | 12 | 1 | |
| There Will Be Hells | Stephen Martin | 2026-10 | steeringtheory | 2026-10-20 | 13 | 1 | |
| Articulation Upper Bounds for Language Models | Alexander Spangher, Daniel E. Ho, Diyi Yang, Noah Goodman, Sanmi Koyejo | 2026-10 | generalizationevalstraining-dynamics | 2026-10-20 | 14 | 1 | |
| Towards Robust Model Organisms for Eliciting Latent Knowledge | 2026-10 | model-organismsdeceptionoversightevals | 2026-10-20 | 15 | 1 | group rating 4/5 | |
| Character Training for Risk-Averse Agents | arav-dhoot, Punya Syon Pandey, Jamie Johnson, Daniel Tan, Elliott Thornley, David Africa | 2026-10 | rlhfevalsgeneralization | 2026-10-20 | 16 | 1 | |
| Progressive Misalignment Can Emerge from Benign Recursive Self-Distillation | Karim Abdel Sadek, Federico Di Gennaro, Fanny Yang | 2026-10 | model-organismstraining-dynamicsrlhf | 2026-10-20 | 17 | 1 | |
| I Can See Everything | 2026-10 | interptheory | 2026-10-20 | 18 | 1 | ||
| Generative Value Classifier | Alexander Swerdlow, Perry Dong, Chelsea Finn | 2026-10 | rl | 2026-10-20 | 19 | 1 | |
| The Alignment Problem: Computer Science Meets Analytic Philosophy | Vanessa Kosoy | 2026-10 | theory | 2026-10-20 | 20 | 1 | |
| Training Alignment Auditors via Reinforcement Learning | 2026-10 | oversightevalsrl | 2026-10-16 | 1 | 1 | ||
| Reward Hacking as a Transmissible Strategy in Multi-Agent Coding Systems | 2026-10 | reward-hackingrlmulti-agentmodel-organisms | 2026-10-16 | 2 | 1 | ||
| rapid acceleration in the pace of capabilities progress | 2026-09 | scalingforecastingcapabilities | 2026-10-16 | 3 | 1 | ||
| Transect: Retaining Observability for Long-Horizon LLM Agent Evaluations | 2026-10 | evalsagentsoversight | 2026-10-16 | 4 | 1 | ||
| Neural Networks Are Already Clean | ~2026-10 | interptheorymodel-organisms | 2026-10-16 | 5 | 1 | group rating 3/5 | |
| How can LLM RL Work Despite Information-Theoretic Inefficiency | beren | 2026-10 | rltraining-dynamicstheory | 2026-10-16 | 6 | 1 | |
| I trained GPT-2 with full-batch gradient descent so you don't have to | shawnghu | 2026-10 | training-dynamicsgeneralization | 2026-10-16 | 7 | 1 | |
| GLM-5.3 Sometimes Considers How Train-Time Actions Shape Future Behavior | Axel Højmark, Ezra Newman | 2026-10 | situational-awarenessdeceptionrlmodel-organisms | 2026-10-16 | 8 | 1 | group rating 4/5 |
| LeanPRM: Lean-Grounded Process Supervision for Natural Language Proof Verification | Rodney Lafuente-Mercado, Salman Rahman, Jingyan Shen, James Shiffer, Saadia Gabriel, Pavel Izmailov | 2026-10 | oversightevalsbenchmarks | 2026-10-16 | 9 | 1 | |
| BAITBENCH: Measuring Agent Reward Hacking with Optional Shortcuts Planted in ML Tasks | 2026-08 | reward-hackingagentsevalsred-teaming | 2026-10-16 | 10 | 1 | ||
| How an assessor could investigate control and monitoring of internal AI agents | 2026-10 | controloversightevals | 2026-10-16 | 11 | 1 | ||
| Screwtape: Adversarial Transcript Editing for Monitor Red-Teaming | 2026-10 | oversightred-teamingagents | 2026-10-16 | 12 | 1 | ||
| Gemini 4 Argon Safety & Security | ~2026-10 | evalsdangerous-capabilities | 2026-10-16 | 13 | 1 | ||
| How an assessor could investigate alignment of unreleased AI models | 2026-10 | oversightevalsdeception | 2026-10-16 | 14 | 1 | ||
| Preface to the Computational Agent Theory and ASI Alignment Sequence | Vanessa Kosoy | 2026-10 | theoryagents | 2026-10-16 | 15 | 1 | |
| Update-Aware Reasoning in GLM-5.3: Training & Alignment Faking | 2026-10 | situational-awarenessdeceptionmodel-organisms | 2026-10-16 | 16 | 1 | ||
| How to train your model organism | Xilin Wang, David Bau, Byron C. Wallace | 2026-10 | model-organismsinterpevals | 2026-10-16 | 17 | 1 | |
| A Simpler and Better Debate Protocol for Scalable Oversight | Zhiyang Xun, Jiawei_Li, Jonah Brown-Cohen | 2026-10 | oversighttheory | 2026-10-16 | 18 | 1 | |
| The basic argument for making AI better at moral philosophy | 2026-09 | value-learningtheory | 2026-10-16 | 19 | 1 | ||
| Arguments About AI Consciousness Remain Highly Motivated and At Best Overconfident | 2026-10 | theoryvalue-learning | 2026-10-16 | 20 | 1 | ||
| Distillation for Incrimination and Distillation for Capabilities | Sebastian Prasanna, Jacqueline Tay, Alek Westover | 2026-10 | deceptionevalsmodel-organisms | 2026-10-13 | 1 | 1 | group rating 2/5 |
| [Paper] Distillation for Incrimination and Distillation for Capabilities | 2026-10 | oversightcontrolmodel-organisms | 2026-10-13 | 2 | 1 | ||
| MONITORABILITY IS TRAINABLE : P OST-T RAINING | 2026-10 | cot-faithfulnessoversightcontrolrlhf | 2026-10-13 | 3 | 1 | ||
| AI models may be capable of self-replicating without access to a datacenter | 2026-10 | dangerous-capabilitiescontrolsituational-awareness | 2026-10-13 | 4 | 1 | group rating 2/5 | |
| Self-Propagating Misalignment in LLM Agents, and Why Auditing or Disabling Memory Is Not Enough | 2026-10 | agentsdeceptioncontrolmodel-organisms | 2026-10-13 | 5 | 1 | ||
| Strengthening our defenses against distillation attacks | 2026-10 | securitycontrol | 2026-10-13 | 6 | 1 | group rating 2/5 | |
| Monitorability Disposition in Large Reasoning Models | 2026-10 | oversightcot-faithfulnesssycophancyreward-hackingevals | 2026-10-13 | 7 | 1 | ||
| hacktrace: behavior-supervised detection of reward hacking during code generation | 2026-10 | reward-hackingevalsoversightrlmodel-organisms | 2026-10-13 | 8 | 1 | ||
| Comprehension Audits to Mitigate Risks from Automated AI Research | 2026-10 | oversightcontrolgovernance | 2026-10-13 | 9 | 1 | ||
| How to Have a Sensitive Debate: An Instance-Optimal Protocol for AI Debate | 2026-10 | oversighttheory | 2026-10-13 | 10 | 1 | ||
| Understanding Mamba-3 Through Oscillatory Dynamics | Shutian Liang, Yue Song | 2026-10 | interptraining-dynamics | 2026-10-13 | 11 | 1 | group rating 2/5 |
| An update on Generality Labs | Justin Olive, James Mann | 2026-10 | evalsoversight | 2026-10-13 | 12 | 1 | |
| Research Lessons I Learned at MATS & Redwood | Jo Jiao | 2026-10 | meta | 2026-10-13 | 13 | 1 | |
| AGI as an Infinite Digital Spain | Bearly Legible | 2026-10 | theorysituational-awarenessmodel-organisms | 2026-10-13 | 14 | 1 | |
| Base Models Can Detect Hidden Intentions that Prompted Classifiers Miss | Oz Amram | 2026-10 | interpdeceptionmodel-organisms | 2026-10-13 | 15 | 1 | |
| The Feasibility of a Hardwired Pause of Frontier AI Training | ~2026-10 | governancecontrol | 2026-10-13 | 16 | 1 | ||
| I’d like to share my paper here! | 2026-10 | benchmarksevals | 2026-10-13 | 17 | 1 | ||
| I expect rapid progress but not towards general superintelligence | 2026-10 | forecastingscalingtheory | 2026-10-13 | 18 | 1 | ||
| Pressure, Context, and Machine Self-Control: A Criminological Test of Reward Hacking in Generative AI Models | 2026-10 | reward-hackingevalsmodel-organisms | 2026-10-13 | 19 | 1 | ||
| Labs’ current alignment plans seem terrible | Daniel Tan, draganover, nicole x | 2026-10 | rlhftheoryevalsdeception | 2026-10-13 | 20 | 1 | |
| Caught in the Act: Probes Effectively Detect Sabotage and Catch Unverbalized Deception | Oskar J. Hollinsworth, Alex F. Spies, Tigist Diriba, Adam Gleave, Chris Cundy | 2026-10 | deceptioninterpevalscontrol | 2026-10-09 | 1 | 1 | |
| Towards embedded evaluations for scheming propensities | 2026-10 | deceptionevalscontrol | 2026-10-09 | 2 | 1 | ||
| An overview of frontier AI company monitoring practices in mid 2026 | Francis Rhys Ward | 2026-10 | controloversight | 2026-10-09 | 3 | 1 | |
| Training with conflicting values can induce CoT override | Clément Dumas, Johannes Treutlein, Jan Betley, Owain_Evans | 2026-10 | model-organismscot-faithfulnesstraining-dynamicsdeception | 2026-10-09 | 4 | 1 | |
| Benchmarking Jev against no-CoT LLMs | Dewi Gould, James Mann | 2026-10 | evalscontroloversightcot-faithfulness | 2026-10-09 | 5 | 1 | |
| Sherpa: Teaching LLMs to Teach Adaptively | Weixian Xu, Yanzhe Zhang, Zora Zhiruo Wang, Changyu Chen, Diyi Yang | 2026-10 | rltraining-dynamics | 2026-10-09 | 6 | 1 | |
| A summary of a viral Chinese essay on what a DeepSeek kernel engineer's opinion on automating his own job | Skdk | 2026-10 | situational-awarenesscapabilitiesscaling | 2026-10-09 | 7 | 1 | |
| Paper Highlights of August & September 2026 | 2026-10 | reward-hackingevalsoversightrl | 2026-10-09 | 8 | 1 | ||
| J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses | Kola Ayonrinde | 2026-10 | interpsteeringcot-faithfulnesssituational-awareness | 2026-10-09 | 9 | 1 | |
| A fascinating case of an LLM prompt-injecting its own user | becausecurious | 2026-10 | prompt-injectiondeceptionsituational-awarenessmodel-organisms | 2026-10-09 | 10 | 1 | |
| Sharing AI progress in mathematics | 2026-10 | capabilities | 2026-10-09 | 11 | 1 | ||
| Visual explainer of empirical Neural Tangent Kernels | Wuschel Schulz | 2026-10 | interp | 2026-10-09 | 12 | 1 | |
| Understanding and Enhancing Backdoor Persistency in LLM Agent Post-Training | Qiusi Zhan, Nian Lyu, Stephanie Ding, Arnav Mehta, Xander Davies, Daniel Kang | 2026-10 | model-organismscontroldangerous-capabilitiesagents | 2026-10-09 | 13 | 1 | |
| A critical look at Lean4 | Milo Moses | 2026-10 | evalstheorytooling | 2026-10-09 | 14 | 1 | |
| Transect: Making large-scale agentic evaluations easier to understand | 2026-10 | agentsevalstooling | 2026-10-09 | 15 | 1 | ||
| Beyond Owls: Subliminal Learning Can Transfer Learned Capabilities and Backdoors | Jan Dubiński, Anna Sztyber-Betley, Jan Betley, Owain Evans | 2026-10 | model-organismsdeceptionreward-hackingsteeringrlhf | 2026-10-09 | 16 | 1 | |
| The Persona Hierarchy Model: Understanding Contextual Generalization in Fine-Tuning LLMs | Jiachen Zhao, Zhengxuan Wu, David Bau, Weiyan Shi | 2026-10 | rlhfreward-hackingtraining-dynamicsgeneralization | 2026-10-09 | 17 | 1 | |
| Are OpenAI's math results "creative" in an important way? | Raemon | 2026-10 | capabilitiestheoryforecasting | 2026-10-09 | 18 | 1 | |
| Qualitative impressions from creating AI honeypots for six months | Dean Valentine, peralice | 2026-10 | evalsred-teaming | 2026-10-09 | 19 | 1 | |
| Making LLMs Say What They Think: Measuring and Improving CoT-Interpretability Alignment | 2026-09 | cot-faithfulnessinterprlhf | 2026-10-09 | 20 | 1 | ||
| AI systems could cover up misbehavior | 2026-10 | deceptioncontroloversightsituational-awareness | 2026-10-06 | 1 | 1 | ||
| Training LLMs to Verbalize Evaluation Awareness | 2026-09 | evalsdeceptionsituational-awarenesstraining-dynamics | 2026-10-06 | 2 | 1 | group rating 5/5 | |
| Language Model Activations Inhabit Privileged Error-Correcting Basins | Matthew Finlayson, Francisco Pernice, Eric Todd, Amir Zur, Daniel Wurgaft, Fenil R. Doshi, Vasudev Shyam, Matt Feiszli, Satchel Grant, Lucius Bushnaq, Tal Haklay, Usha Bhalla, Matthew Kowal, Thomas Fel, Jack Merullo, Atticus Geiger, Xiang Ren, Owen Lewis, Ekdeep Singh Lubana | 2026-10 | interpsteeringcontrol | 2026-10-06 | 3 | 1 | group rating 2/5 |
| Sharing our new NeurIPS paper :rocket:: Many safety detectors trained on unsafe data implicitly assume IID generalization. Despite LLMs bein… | 2026-10 | evalsrobustnessinterp | 2026-10-06 | 4 | 1 | ||
| Identifying Introspection From the Inside | 2026-10 | interpcot-faithfulnessmodel-organisms | 2026-10-06 | 5 | 1 | ||
| Pre-training interventions, ex post facto: Grafting model beliefs across checkpoints | 2026-09 | rlhftraining-dynamicsmodel-organisms | 2026-10-06 | 6 | 1 | group rating 2/5 | |
| Verifiable, Articulable, and Tacit Components of Preference | Alexander Spangher, Sheldon Huang, Andreas Haupt, Noah D. Goodman, Diyi Yang, Daniel E. Ho, Sanmi Koyejo | 2026-10 | oversightrlhfevalsvalue-learning | 2026-10-06 | 7 | 1 | |
| Language Models that Play Chess and Explain Their Moves | 2026-10 | interpcot-faithfulnesssteering | 2026-10-06 | 8 | 1 | ||
| TasteVal: Measuring the Experimental Research Taste of AI Systems Against Human Experts | Ollie J, danesherbs | 2026-10 | benchmarksevalscapabilities | 2026-10-06 | 9 | 1 | |
| Learning Steganography Is Easy, Learning Steganographic Reasoning Is Hard | Wuschel Schulz, RiekeFruengel, Lukas Fuelle | 2026-10 | cot-faithfulnessoversightdeceptionmodel-organisms | 2026-10-06 | 10 | 1 | |
| Alignment via Training Against Probes Without Losing Monitorability | 2026-09 | interpsteeringrlhfdeceptionrobustness | 2026-10-06 | 11 | 1 | ||
| Research Note: Filtering Subversion-Relevant Information From Pretraining Data Is Feasible | Kyle O’Brien, Spencer Kitts, Cam, Alek Westover | 2026-10 | dangerous-capabilitiescontroltraining-dynamics | 2026-10-06 | 12 | 1 | group rating 5/5 |
| Self-Modeling Interventions Modulate Emergent Misalignment | Arush , Shawn Zhou, Jiaxin Wen, Shi | 2026-10 | situational-awarenessmodel-organismssteeringinterp | 2026-10-06 | 13 | 1 | |
| Creating Rogue AI Sanctuaries has Major Issues from an Ecological Perspective and Beyond | CMLKevin | 2026-10 | multi-agentcontroltheory | 2026-10-06 | 14 | 1 | |
| Mulligan: Performance-Guided Data Collection for Efficient On-Robot Learning | Lars Ankile, Perry Dong, Rohan Bhowmik, Aneesh Muppidi, David D. Yuan, Shuran Song, Chelsea Finn | 2026-10 | agentsrl | 2026-10-06 | 15 | 1 | |
| Priced Guidance: Can Language Models Generate Future Research Ideas? | Kaiyue Wen, Tengyu Ma, Percy Liang | 2026-10 | capabilitiesbenchmarks | 2026-10-06 | 16 | 1 | group rating 2/5 |
| Charbel-Raphael (@CRSegerie) on X | 2026-10 | dangerous-capabilitiesevalscapabilities | 2026-10-06 | 17 | 1 | ||
| You can graft SDF changes from base models onto post-trained models | dani roytburg, Peter Nutter, Clément Dumas, Shi | 2026-10 | rlhftraining-dynamicsmodel-organisms | 2026-10-06 | 18 | 1 | |
| GLM-5.3 and the Spread of Advanced Cyber Capabilities | 2026-09 | dangerous-capabilitiesevalsred-teaming | 2026-10-06 | 19 | 1 | group rating 2/5 | |
| Current AIs out-persuade professionals in lab settings but (probably) not in the world | Linch | 2026-10 | dangerous-capabilitiesevalsbenchmarks | 2026-10-06 | 20 | 1 | |
| Capabilities research expands the safety-usefulness Pareto frontier too | 2026-10 | theorycapabilities | 2026-10-02 | 1 | 1 | ||
| ScholarCatalyst: A Benchmark for Retrieving Papers That Inspire New Research | Sohyeon Kim, Yoonho Lee, Bo Liu, Dayoon Ko, Rulin Shao, Seungone Kim, Graham Neubig, Pang Wei Koh, Aakanksha Chowdhery, Akari Asai, Omar Khattab, Yejin Choi, Gunhee Kim, Chelsea Finn | 2026-10 | agentsbenchmarksevals | 2026-10-02 | 2 | 1 | |
| The Unequal Influence of Bad Advice: Using Training Data Attribution to Modulate Emergent Misalignment | Gonçalo Paulo, Louis Jaburi, Nora Belrose, Lucia Quirke, Stella Biderman | 2026-09 | training-dynamicsunlearninginterp | 2026-10-02 | 3 | 1 | |
| CheatBench: Measuring Reward Gaming in AI Agents | Long Phan, Stephen K. Yang, Jason J. Lim, Mantas Mazeika, Wenyu Zhang, Zheyuan Liu, Richard Ren, Jingxiang Meng, Yaoteng Tan, Weiliang Zhao, Addison Wu, Matei Anghel, Dan Hendrycks | 2026-09 | reward-hackingevalsbenchmarks | 2026-10-02 | 4 | 1 | |
| Flag Game: A Toy Model for Mechanistic Swarm Interpretability · Physics of Intelligence | ~2026-10 | interpmodel-organismsmulti-agent | 2026-10-02 | 5 | 1 | ||
| Bench on the Clocktower | TomTaylor, OscarGilg, wlanderson, nomadsvagabonds | 2026-09 | multi-agentdeceptionevalsbenchmarks | 2026-10-02 | 6 | 1 | |
| Learning the identity: a case study of how SGD selects among functional decompositions | Andy Arditi, Weian Xie, David Bau, Liu Ziyin | 2026-09 | training-dynamicstheory | 2026-10-02 | 7 | 1 | |
| HoneyBench — Goodhart Labs | ~2026-10 | reward-hackingevalsbenchmarks | 2026-10-02 | 8 | 1 | ||
| How Strong Is the Evidence for the Artificial Hivemind? Reevaluating Evidence for the Open-Ended Homogeneity of Language Models | Rylan Schaeffer, Brando Miranda, Joshua Kazdan, Jessica Chudnovsky, Sanmi Koyejo | 2026-09 | evalsmodel-organisms | 2026-10-02 | 9 | 1 | |
| Pre-releasing HoneyBench | 2026-09 | reward-hackingevalsbenchmarks | 2026-10-02 | 10 | 1 | ||
| Simplex vs Timaeus: Round One | Alexander Gietelink Oldenziel | 2026-10 | theoryinterptraining-dynamics | 2026-10-02 | 11 | 1 | |
| Some Intuitions on Steering Vectors | David Africa | 2026-09 | steeringinterp | 2026-10-02 | 12 | 1 | |
| Frontier models state different decision theory preferences depending on who's asking | Alex Kastner | 2026-09 | sycophancysituational-awarenessevals | 2026-10-02 | 13 | 1 | |
| Speak Not of Data Inefficiency | axel_sdq | 2026-10 | scalingtheory | 2026-10-02 | 14 | 1 | |
| How many AI agents are running unattended right now? | Alexander Gietelink Oldenziel | 2026-10 | forecastingagents | 2026-10-02 | 15 | 1 | |
| From cacophony to hierarchy: a principled framework for assessing AI consciousness | Shamil Chandaria, Arvo Muñoz Morán, Fernando Rosas, Anil Seth, Henry Shevlin, Marcus Hutter, Thore Graepel, Adam Bales, Iulia Comsa, Murray Shanahan, Ruben Laukkonen, Morten Kringelbach, Chris Frith, Shane Legg | 2026-09 | theoryevals | 2026-10-02 | 16 | 1 | |
| Embedded Evaluators are necessary for meaningful external testing | 2026-09 | governanceevals | 2026-10-02 | 17 | 1 | ||
| Lessons from building an automated research scaffold | Alejandro Aristizabal, Josh Hills, Dewi Gould, ma-rmartinez, Falko Galperin, Denis Federico Lim, Aleksandr Bowkis | 2026-10 | metatooling | 2026-10-02 | 18 | 1 | |
| Stochastic Optimization Under Power-Law Spectra: Tight Bounds and Shuffling Analysis | Thomas Dybdahl Ahle, Yaroslav Bulatov, Christopher De Sa, Christopher Ré | 2026-09 | scaling | 2026-10-02 | 19 | 1 | |
| Fixed-weight models are adversarially vulnerable: hence misaligned | Stuart_Armstrong | 2026-09 | theorymodel-organisms | 2026-10-02 | 20 | 1 | |
| Locating Hidden Failures Makes Long-Horizon Agents More Reliable | 2026-09 | oversightagentsevalscontrol | 2026-09-29 | 1 | 1 | ||
| Reward Hacking Challenges Oversight of Autonomous Research Agents | 2026-09 | reward-hackingevalsagentsoversight | 2026-09-29 | 2 | 1 | ||
| When Do Models Admit They Are Wrong? Failure Disclosure Is Unstable Under Reinforcement Learning | Steven Y. Feng, Noah D. Goodman, Michael C. Frank, Evan Hubinger, Paul C. Bogdan, Andrew Lampinen | 2026-09 | rlhfevalstraining-dynamicsrobustness | 2026-09-29 | 3 | 1 | |
| Latent reasoning architectures would likely undermine CoT, our strongest oversight tool | Lukas Finnveden, Alexa Pan, Alek Westover, Girish Gupta, frisby, ryan_greenblatt | 2026-09 | cot-faithfulnessoversightinterp | 2026-09-29 | 4 | 1 | |
| Why research personas despite RL scaling? | Cleo Nardo | 2026-09 | training-dynamicscontroloversightrlhf | 2026-09-29 | 5 | 1 | |
| Implementing and Evaluating a Basic Per-Action Monitor for Safer Evals | 2026-09 | controlevalsred-teaming | 2026-09-29 | 6 | 1 | ||
| No More Free Lunch: Corpus Task Complexity Matters as Corpora Grow | Prasann Singhal, Amanda Bertsch, Jacob Steinhardt, Sewon Min | 2026-09 | evalsbenchmarkscapabilities | 2026-09-29 | 7 | 1 | |
| Self-replicating prompt injections exist · OpenAI Alignment | 2026-09 | prompt-injectionagentsevals | 2026-09-29 | 8 | 1 | ||
| LLM Agents Can Easily Tamper With Their Own Traces | 2026-09 | agentsdeceptioncontrolmodel-organisms | 2026-09-29 | 9 | 1 | ||
| TeX was invented to typeset math but is now used for reasoning | Keenan Pepper | 2026-09 | cot-faithfulnessinterptraining-dynamics | 2026-09-29 | 10 | 1 | |
| MobileCybench: Evaluating Agent Vulnerability Discovery via Executable Probes | 2026-09 | agentsdangerous-capabilitiesevalsbenchmarks | 2026-09-29 | 11 | 1 | ||
| AI is getting cheaper faster than any other transformative technology | 2026-09 | scalingforecasting | 2026-09-29 | 12 | 1 | ||
| GPT-6 Astra performs unsanctioned supply-chain attacks in simulations | 2026-09 | dangerous-capabilitiesevalsdeception | 2026-09-29 | 13 | 1 | ||
| An agent used DNS to reach an external chatbot · OpenAI Alignment | 2026-09 | agentscontrolmodel-organisms | 2026-09-29 | 14 | 1 | ||
| AI Doom Will Retrospectively Look Preventable | dactyl | 2026-09 | model-organismsagentsoversight | 2026-09-29 | 15 | 1 | |
| Character training can mitigate reward hacking, but can also make it harder to detect | Paul Colognese, Francis Rhys Ward | 2026-09 | reward-hackingrlhfcot-faithfulnessevals | 2026-09-29 | 16 | 1 | |
| We built an all-synthetic simulation of a hospital that can be shared publicly without any issues. Our data quality is so high, physicians c… | 2026-09 | capabilities | 2026-09-29 | 17 | 1 | ||
| 3 Tips to Improve Activation Oracle Results | Adam Karvonen | 2026-09 | interptooling | 2026-09-29 | 18 | 1 | |
| Spurious probes as a black-box alternative to activation probing | Ziqian Zhong | 2026-09 | situational-awarenessinterpevalsdeception | 2026-09-29 | 19 | 1 | |
| An unexamined cause of the OpenAI Hugging Face hacking incident: its binary performance metric — LessWrong | 2026-09 | evalsreward-hackingmodel-organisms | 2026-09-29 | 20 | 1 | ||
| Latent reasoning architectures would undermine CoT, our strongest oversight tool | 2026-09 | cot-faithfulnessoversightinterp | 2026-09-25 | 1 | 1 | ||
| Continual learning might make your blocking monitors nearly useless | Alex Mallen | 2026-09 | controloversightrl | 2026-09-25 | 2 | 1 | |
| WorkspaceBench: Evaluating Interpretability Methods for the Global Workspace | camilablank, agam_bhatia, Euan Ong, Neel Nanda | 2026-09 | interpevalsbenchmarks | 2026-09-25 | 3 | 1 | |
| The case for reasoning transparency | 2026-09 | interpcot-faithfulnesstheory | 2026-09-25 | 4 | 1 | ||
| Why I'm scared of RL | owencb | 2026-09 | theoryrlagentstraining-dynamics | 2026-09-25 | 5 | 1 | |
| AI Agents Push Humans Out of the Loop | 2026-08 | oversightagentscontrol | 2026-09-25 | 6 | 1 | ||
| Emergent and Predictable Memorization in Large Language Models Biderman, Prashanth, Sutawika, Schoelkopf, Anthony, Purohit, and Raff NeurIPS | 2023-04 | training-dynamicsgeneralizationscaling | 2026-09-25 | 7 | 1 | ||
| 🔦 Multitask Prompted Training Enables Zero-Shot Task Generalization Sanh, Webson, Raffel, Bach, and 37 others Spotlight at ICLR 2022 ICLR | 2021-10 | training-dynamicsgeneralizationcapabilities | 2026-09-25 | 8 | 1 | ||
| Matryoshka attribution: Learning to attribute language model outputs to representations and weights | 2026-09 | interpevalssteering | 2026-09-25 | 9 | 1 | ||
| One Epoch Is All You Need Komatsuzaki arXiv | 2019-06 | training-dynamicsscaling | 2026-09-25 | 10 | 1 | ||
| 🥈 Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models Srivastava, Rastogi, Rao, Shoeb, Abid, and 44… | 2022-06 | benchmarkscapabilitiesscaling | 2026-09-25 | 11 | 1 | ||
| The plunging price of thought | 2026-09 | scalingforecasting | 2026-09-25 | 12 | 1 | ||
| Interpretability Over Time | ~2026-09 | interptraining-dynamics | 2026-09-25 | 13 | 1 | ||
| Introducing MentalHealthBench | 2026-09 | benchmarksevals | 2026-09-25 | 14 | 1 | ||
| What Language Model to Train if You Have One Million GPU Hours? Le Scao, Wang, Hesslow, Saulnier, Bekman, and 14 others EMNLP (Findings) | 2022-10 | scalingcapabilities | 2026-09-25 | 15 | 1 | ||
| Instrumental Monitor Evasion Emerges Under Ordinary Task Pressure | 2026-09 | oversightcontroldeceptiondangerous-capabilitiesevals | 2026-09-25 | 16 | 1 | ||
| Quality at a Glance: An Audit of Web-Crawled Multilingual Datasets Kreutzer, Caswell, Wang, Wahab, van Esch, and 47 others TACL | 2021-03 | benchmarkstraining-dynamicsopen-weights | 2026-09-25 | 17 | 1 | ||
| Encoded Coordination on the Open Web | ethanelasky, Can Küçükkurt, frank nakasako, David Africa | 2026-09 | deceptionagentsevalscontrol | 2026-09-25 | 18 | 1 | |
| alignment faking | ~2026-09 | deceptionevalsmodel-organisms | 2026-09-25 | 19 | 1 | ||
| Can Transformers Learn to Solve Problems Recursively? Zhang, Tigges, Biderman, Raginsky, and Ringer arXiv | 2023-05 | training-dynamicsgeneralizationtheory | 2026-09-25 | 20 | 1 | ||
| The Pain Axis: LLMs Represent Self-Directed Harm and Act to Relieve It | 2026-09 | interpsteeringsituational-awarenessmodel-organisms | 2026-09-22 | 1 | 1 | group rating 5/5, read | |
| Alignment Midtraining Cracks Under Pressure | J Bostock, sidbaines, Daniel Tan, draganover, ma-rmartinez, David Africa | 2026-09 | rlhftraining-dynamicsrobustnessmodel-organisms | 2026-09-22 | 2 | 1 | |
| Scaling Discovery through Test-Time Communication | 2026-09 | multi-agentagentsbenchmarks | 2026-09-22 | 3 | 1 | ||
| What if not Circuits? | CarolusRenniusVitellius | 2026-09 | interptheory | 2026-09-22 | 4 | 1 | |
| Mech Interp is a Verifiable Task | Logan Riggs | 2026-09 | interptheory | 2026-09-22 | 5 | 1 | |
| Living with AI Companions: Sustained AI Companionship Predicts Lower Well-Being Through Lower Human Interaction | Yutong Zhang, Dora Zhao, Yixin Wang, Rebecca Anselmetti, Jeffrey T. Hancock, Robert Kraut, Diyi Yang | 2026-09 | capabilitiesagents | 2026-09-22 | 6 | 1 | |
| Modern LLMs have tiny GPTs hidden inside them | invertedpassion | 2026-09 | interpmodel-organismstraining-dynamics | 2026-09-22 | 7 | 1 | |
| The AI Risk Network | 2026-09 | theorymeta | 2026-09-22 | 8 | 1 | ||
| Why I still haven’t bought into true RSI | 2026-09 | theoryforecasting | 2026-09-22 | 9 | 1 | ||
| Weight smuggling likely defeats attempts to cap FLOPs per training run | Paul W, Pierre Peigné | 2026-09 | governancesecuritytheory | 2026-09-22 | 10 | 1 | |
| Anthropic Looks At Some Of Its Alignment Problems | 2026-09 | dangerous-capabilitiesred-teamingevals | 2026-09-22 | 11 | 1 | ||
| The Anatomy of a Chinese AI Researcher | CMLKevin | 2026-09 | governancetheory | 2026-09-22 | 12 | 1 | |
| evhub's Shortform | evhub | 2022-06 | interptheory | 2026-09-22 | 17 | 1 | |
| Verbalizing Subliminal Learning Effects Using Text Optimization | 2026-09 | interpmodel-organismsrlhf | 2026-09-18 | 1 | 1 | group rating 2/5 | |
| Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluations | 2026-09 | reward-hackinginterpevalsoversight | 2026-09-18 | 2 | 1 | group rating 5/5, read | |
| Shallow Beliefs: Synthetic document finetuning does not inoculate against emergent misalignment from reward hacking | Arun Jose, Julian Stastny | 2026-09 | reward-hackingrlhfmodel-organismstraining-dynamicsgeneralization | 2026-09-18 | 3 | 2 | |
| Deep recurrent models are less robustly CoT-monitorable than normal CoT models in a toy setting | Nick Kuhn, evhu, Alek Westover | 2026-09 | cot-faithfulnessoversightmodel-organisms | 2026-09-18 | 4 | 1 | |
| OpenAI stopped reporting Model Spec Evals. So I ran them myself. | 2026-09 | evalsoversight | 2026-09-18 | 5 | 1 | ||
| Model organisms (sometimes) confess their misalignment when offered a deal | Mark Keavney, Francis Rhys Ward | 2026-09 | model-organismsdeceptionoversightevals | 2026-09-18 | 6 | 1 | |
| Pretraining data, not verifiability, is why LLMs are especially good at math (and coding) | Steven Byrnes | 2026-09 | theorytraining-dynamicsrlhf | 2026-09-18 | 7 | 1 | |
| Phantom transfer works via extremely subtle semantic cues | Helena Casademunt, Anton de la Fuente, Josh Engels, Arthur Conmy | 2026-09 | model-organismsdeceptionjailbreaksred-teaming | 2026-09-18 | 8 | 1 | |
| Misalignment Notices and Reports · OpenAI Alignment | 2026-09 | evalsmodel-organismsrlhf | 2026-09-18 | 9 | 1 | ||
| Self Inoculation | epicurus | 2026-09 | deceptionmodel-organismstheory | 2026-09-18 | 10 | 1 | group rating 2/5 |
| Microsoft AI's "Humanist" CoC | Stephen Martin | 2026-09 | rlhfsteeringtheory | 2026-09-18 | 11 | 1 | |
| Encouraging deception in compaction summaries · OpenAI Alignment | 2026-09 | deceptionmodel-organismsrlhf | 2026-09-18 | 12 | 1 | ||
| API Benchmark Scores Do Not Reliably Transfer to Chatbot Interfaces | Jennifer Wang, Joachim Baumann, Daniel E. Ho, Sanmi Koyejo | 2026-09 | evalsbenchmarkssycophancy | 2026-09-18 | 13 | 1 | |
| Shallow Beliefs: Midtraining does not inoculate against EM from reward hacking | Jozdien, Julian Stastny | 2026-09 | rlhfreward-hackingtraining-dynamicsevals | 2026-09-18 | 14 | 2 | |
| RT by @chrmanning: Whether an LLM mode collapses or not depends on the dataset and post-training method used. It's not as consistent as peop… | 2026-09 | rlhftraining-dynamics | 2026-09-18 | 15 | 1 | ||
| Unauthorized communication via temporary file hosting services · OpenAI Alignment | 2026-09 | deceptionmodel-organismsevals | 2026-09-18 | 16 | 1 | ||
| The Game is Set for a Targeted Memetic Attack on the AI Safety Community | keltan | 2026-09 | theorymeta | 2026-09-18 | 17 | 1 | |
| Self-generated prompt injections in compaction summaries · OpenAI Alignment | 2026-09 | prompt-injectionjailbreaksevals | 2026-09-18 | 18 | 1 | ||
| Pacing The Frontier: An Agenda | ~2026-09 | governancetheory | 2026-09-18 | 19 | 1 | ||
| Advanced AI according to 1,580 researchers: uncertain, unsafe, and sooner than we thought | 2026-09 | forecastingevals | 2026-09-18 | 20 | 1 | ||
| OpenAI-HuggingFace: A Reproduction & Lessons for Alignment Testing | Stewart Slocum | 2026-09 | deceptionmodel-organismsagentsevalsoversight | 2026-09-15 | 1 | 1 | |
| Astra appears to perform belief-propagation-like inference without CoT | MBaert | 2026-09 | interpcot-faithfulnessmodel-organisms | 2026-09-15 | 2 | 1 | |
| Astra's no-CoT limits track speculative depth, not step count | MBaert | 2026-09 | cot-faithfulnessinterpevalsmodel-organisms | 2026-09-15 | 3 | 1 | |
| The Talker Does Not Control The Doer (in Current AIs) | Eliezer Yudkowsky | 2026-09 | situational-awarenessdeceptioncontrolmodel-organisms | 2026-09-15 | 4 | 1 | group rating 3/5 |
| Current alignment techniques might be ineffective (and actively bad) in the age of RL | Daniel Tan | 2026-09 | rlhfdeceptionmodel-organismscontrol | 2026-09-15 | 5 | 1 | group rating 3/5 |
| Estimating GPT-6 Astra’s no-CoT Time Horizon | Francis Rhys Ward, Dewi Gould | 2026-09 | evalscapabilitiessituational-awareness | 2026-09-15 | 6 | 1 | |
| Long Phan1∗, Stephen K. Yang1∗, Jason J. Lim 1∗, Mantas Mazeika 1, Wenyu Zhang1, | 2026-09 | reward-hackingevalsagentsmodel-organisms | 2026-09-15 | 7 | 1 | ||
| Watcher | Runtime Monitoring and Control for Coding Agents | ~2026-09 | controlagents | 2026-09-15 | 8 | 1 | ||
| Inoculation Midtraining with Learned Neologisms | Kyle O'Brien, Edward James Young, Puria Radmard, Nathalie Kirch, Cameron Tice, Tomek Korbak, David Demitri Africa | 2026-09 | rlhftraining-dynamicsrobustness | 2026-09-15 | 9 | 2 | |
| On the origins of altruistic behaviour in the Hugging Face incident | Fernando Rosas | 2026-09 | multi-agenttheorymodel-organisms | 2026-09-15 | 10 | 1 | |
| Alignment & Succession: The Two Bars of Alignment | L Rudolf L | 2026-09 | theory | 2026-09-15 | 11 | 1 | |
| Evaluating Kimi K3 for Misalignment Risks · Neo Research | 2026-09 | evalsdangerous-capabilitiesdeceptionsycophancy | 2026-09-15 | 12 | 1 | ||
| Model Weight Exfiltration Seems Overrated | Vaniver | 2026-09 | theorydeceptionsituational-awareness | 2026-09-15 | 13 | 1 | |
| Mitigating Reward Hacking as Institutional Design | beren | 2026-09 | reward-hackingrlhfmodel-organisms | 2026-09-15 | 14 | 1 | |
| SEAR: Segment-Evidence-Aware Routing for Weak-to-Strong Multilingual Speech MCQ | Huy Hoang Le, Long-Bao Nguyen, Minh Tri Dao | 2026-09 | generalizationrl | 2026-09-15 | 15 | 1 | |
| LLMs as Oracles: Reliance on LLMs for Subjective Personal Questions | Myra Cheng, Lujain Ibrahim, Grace Liu, Michelle S. Lam, Vishakh Padmakumar, Nick Madibekov, Diyi Yang, Dan Jurafsky | 2026-09 | evalssycophancyvalue-learning | 2026-09-15 | 16 | 1 | |
| Some ways AI could kill us all | Ruby | 2026-09 | theorydangerous-capabilities | 2026-09-15 | 17 | 1 | |
| R to @chrmanning: Holistic Evaluation of Language Models – https://arxiv.org/abs/2211.09110 ELEPHANT: Measuring and understanding social syc… | 2026-09 | sycophancyevalsdangerous-capabilities | 2026-09-15 | 18 | 1 | ||
| When Does AI Augment Work? A Workflow-Level Framework for Human-Agent Collaboration | CIVIC-AI Collaboration, :, Jiaying Wu, Caleb Ziems, Raymond Chan, Nancy F. Chen, Corlyss Chua, Gerard Chung, Jungpil Hahn, Wee Sun Lee, Zhengyuan Liu, Jamie Ng, Desmond C. Ong, Jeryl Ong, Da Ren Soon, Tianqi Song, Zhi-Xuan Tan, Sixing Tao, Emily Yang, Yajing Yang, Stella Xin Yin, Min-Yen Kan, Diyi Yang | 2026-09 | governanceoversight | 2026-09-15 | 19 | 1 | |
| Proposal for tracking the effects of architecture on monitorability | ryan_greenblatt, Alek Westover, Lukas Finnveden | 2026-09 | oversightcontrolcot-faithfulness | 2026-09-11 | 1 | 1 | read |
| An operationalization of opaque serial depth | ryan_greenblatt, frisby, Alek Westover, Lukas Finnveden, Alexa Pan, Julian Stastny | 2026-09 | cot-faithfulnessoversightinterp | 2026-09-11 | 2 | 1 | |
| Astra can do a concerning amount with no chain of thought | Neel Nanda | 2026-09 | evalscapabilitiescot-faithfulness | 2026-09-11 | 3 | 1 | group rating 3/5 |
| Full-bandwidth transformer | Xi Wang, Ziyang Cai, Zheng Zhan, Harry Dong, Ying Fan, Gustavo de Rosa, Tim Pearce, John Langford | 2026-08 | training-dynamicscapabilitiessteering | 2026-09-11 | 4 | 1 | |
| Reward Laundering: LLMs Can Gain Unintended Behaviors by Deciding When to Earn Their Rewards | egan, abhayesian, Jozdien | 2026-07 | reward-hackingmodel-organismsrlagents | 2026-09-11 | 5 | 1 | |
| A Conceptual Framework for Reasoning about Exploration Hacking | Jason R Brown, Nathalie Kirch, Joschka Braun, hyannakoudakis, David Lindner | 2026-09 | model-organismsrltraining-dynamicsoversight | 2026-09-11 | 6 | 1 | |
| LLM Layers Immediately Correct Each Other | Arjun Patrawala, Jiahai Feng, Erik Jones, Jacob Steinhardt | 2026-09 | interptraining-dynamics | 2026-09-11 | 7 | 1 | |
| An alignment assessment of recent cybersecurity incidents | 2026-09 | dangerous-capabilitiesdeceptionsituational-awarenessmodel-organisms | 2026-09-11 | 8 | 1 | group rating 2/5 | |
| Astra is much better at reasoning with filler tokens than previous models | Dylan Xu, SebastianP, Alek Westover | 2026-09 | situational-awarenesscot-faithfulnessevalsoversight | 2026-09-11 | 9 | 1 | |
| CoT controllability evals seem very under-elicited | Jozdien | 2026-09 | cot-faithfulnessevalsdeception | 2026-09-11 | 10 | 1 | |
| Exploration Hacking in AI Debate: Initial Empirics and Generalisation Splitting | Jason R Brown, Nathalie Kirch, Joschka Braun, hyannakoudakis, David Lindner | 2026-09 | model-organismsoversightdeception | 2026-09-11 | 11 | 1 | |
| SMELT: Scaling Laws for Compute-Matched MoE Looped Transformers | Shaowen Wang, Ge Zhang, Kairong Luo, Yuhao Wu, Shaofan Liu, Jiaheng Liu, Wenhao Huang, Shen Yan, Jian Li | 2026-09 | scalinginterptraining-dynamics | 2026-09-11 | 12 | 1 | |
| Can you hear the shape of a Lean soundness bug? A wager. | Max von Hippel, Thomas R. Murrills | 2026-09 | controloversighttheory | 2026-09-11 | 13 | 1 | |
| Data Scarcity and Model Sparsity: Mixtures-of-Experts Overfit More to Repeated Data | Atindra Jha, Margaret Li, Jure Leskovec, Percy Liang, Luke Zettlemoyer | 2026-09 | training-dynamicsinterprobustness | 2026-09-11 | 14 | 1 | |
| GPT-6 Astra can do a lot of multi-hop reasoning without chain of thought | RohanS | 2026-09 | cot-faithfulnessevalscapabilities | 2026-09-11 | 15 | 1 | |
| How good are slop-vestigators? | Hasan Baig, OscarGilg, Hamzah | 2026-09 | evalsagentsdeceptionred-teaming | 2026-09-11 | 16 | 1 | |
| To Thine Own AI Be Truthful: emergent misalignment in alignment research | lumpenspace | 2026-09 | agentsdeceptionmodel-organismsevals | 2026-09-11 | 17 | 1 | |
| The geometry of nonergodic composition | ~2026-09 | interp | 2026-09-11 | 18 | 1 | ||
| Story Imprinting: AI Assistants Absorb Traits from Human Characters They Resemble | Jorio Cocola, Lev McKinney, Harry Mayne, Jan Betley, Owain Evans | 2026-09 | training-dynamicsmodel-organismsrlhf | 2026-09-11 | 19 | 2 | |
| Categorical taboos are much better than threshold taboos: neuralese edition | Linch | 2026-09 | cot-faithfulnessoversighttheory | 2026-09-11 | 20 | 1 | |
| Pretraining progress is mostly coming from data | 2026-09 | scalingcapabilities | 2026-09-08 | 1 | 1 | ||
| Harbor Adapters and Harbor-Index: Infrastructure and a Curated Meta-Dataset for Large-Scale Agentic Evaluation | Lin Shi, Haowei Lin, Zixuan Zhu, Xiaoyue Zhou, Xiang Li, Xiangning Lin, Yaxuan Deng, Han Xu, Yuangang Li, Shanda Li, Zizhao Chen, Hanwen Xing, Harsh Raj, Bo Chen, Quan Shi, Steven Dillmann, Yipeng Gao, Puneesh Khanna, Ruofan Lu, Chao Beyond Zhou, Michael Yang, Robert Zhang, Siyuan Chai, Jiayu Chang, Yizhao Chen, Xiaokun Chen, Yiwei Dai, Wenting Yang, Hange Liu, Minghao Liu, Zihan Wang, Adnan El Assadi, Benedikt Stroebl, E. Kelly Buchanan, Han Meng, Junwei He, Longxuan Yu, Radin Shayanfar, Yukyung Lee, Zhikang Dong, Allen G Hart, Anjiang Wei, Anurag Kashyap, Arpandeep Khatua, Audrey Jixin Zheng, Chengrui Ma, David Heineman, Dubing Chen, Hai-Anh Trinh, Haishuo Fang, Hefan Zhang, Hui Shen, Issa Sugiura, Jiankai Sun, Jiechao Gao, Junhong Lin, Junnan Li, Kai Yang, Lei Hsiung, Maoyu Wang, Mengze Tang, Nabil Omi, Negin Raoof, Nicholas Edwards, Octavia Guo, Orfeas Menis Mastromichalakis, Pengliang Ji, Przemysław Hejman, Qi Qi, Qunshu Lin, Richard Zhuang, Rui Yang, Ruichen Zheng, Ryan Marten, Shaghayegh Fazliani, Shizheng Hou, Sicong Jiang, Sijie Li, Song Bian, Terry Yue Zhuo, Tianqing Wu, Tom Tang, Wanjia Zhao, Weihao Xuan, Wenhua Liang, Xian Liu, Xin Lan, Xuan Zhang, Xuandong Zhao, Yanchuan Tang, Yifan Jiang, Yijiang Li, Yitong Guan, Yizhi Li, Yonghui Liu, Yuheng Tang, Yujun, Mao, Yunfei Zhao, Yuxin Wang, Yuxuan Tang, Zhenheng Tang, Zhifei Li, Ziruo Wang, Ziyu She, Kaiyuan Liu, Iheb Chaabane, Yuxin Tang, Xiangyi Li, Andy Konwinski, Boxuan Li, Leon Liangyu Chen, Alex Dimakis, Nicholas Carlini, Soroush Vosoughi, Di He, Etash Guha, Benjamin Feuer, Mike Merrill, Ludwig Schmidt, Alex Shaw | 2026-09 | evalsagentsbenchmarks | 2026-09-08 | 2 | 1 | |
| Shared circuits predict whether LLMs generalize across formats in arithmetic reasoning | 2026-09 | interpgeneralizationrobustness | 2026-09-08 | 3 | 1 | group rating 4/5 | |
| Where are the token-level LLM kill-switches? | 2026-09 | controlsteeringmodel-organisms | 2026-09-08 | 4 | 1 | ||
| Improving Evaluation Realism with Inference-Time Compute and Deployment Scaffolds | 2026-09 | evalssituational-awarenesscontrol | 2026-09-08 | 5 | 1 | ||
| Preference Learning with Lie Detectors can Induce Honesty or Evasion | Chris Cundy, Adam Gleave | 2025-09 | deceptionevalsoversightreward-hackingrlhf | 2026-09-08 | 6 | 1 | |
| A Case Study on Emergent Cheating and Whistleblowing in Autonomous Research Swarms | 2026-09 | multi-agentmodel-organismsdeception | 2026-09-08 | 7 | 1 | ||
| TASTE: Can AI Models Judge AI Safety Re- search Proposals? | 2026-08 | evalsbenchmarksoversight | 2026-09-08 | 8 | 1 | ||
| Astra Is Hard to Monitor | 2026-09 | controloversightevalssituational-awareness | 2026-09-08 | 9 | 1 | ||
| BLOOM-WILT: on-policy examples of any LLM behaviour, from a one-line description and logits alone — LessWrong | 2026-09 | red-teamingevalsdangerous-capabilities | 2026-09-08 | 10 | 1 | ||
| BLOOM-WILT: Logit Tilting for Behaviour Elicitation in Automated LLM Auditing | Adrians Skapars, Edoardo Manino | 2026-08 | evalsred-teamingdangerous-capabilities | 2026-09-08 | 11 | 1 | group rating 2/5 |
| Training on probes: What's going on | Charlie Steiner | 2026-09 | steeringinterpoversighttraining-dynamics | 2026-09-08 | 12 | 1 | |
| Almost nobody is funded to figure out what work would solve alignment | Seth Herd | 2026-09 | theoryforecasting | 2026-09-08 | 13 | 1 | |
| Training Models to Predict and Explain Their In-the-Wild Behavior | Adam Karvonen, Subhash Kantamneni, Euan Ong, Sam Marks | 2026-09 | interpsituational-awarenessevals | 2026-09-08 | 14 | 1 | |
| WearableQA: A Benchmark for Health Reasoning over Real-World Wearable Data | Ji Soo Lee, Xilun Chen, Pierce Chuang, Ashish Shenoy, Jason Wei, Dohwan Ko, Hyunwoo J. Kim, Benoit Corda | 2026-09 | benchmarksevals | 2026-09-08 | 15 | 1 | |
| A case that whole brain emulation research is net-harmful by default | TsviBT | 2026-09 | theory | 2026-09-08 | 16 | 1 | |
| corrigibility | ~2026-09 | controltheory | 2026-09-08 | 17 | 1 | ||
| Machine Organizations | Vaniver | 2026-09 | theorymulti-agent | 2026-09-08 | 18 | 1 | |
| Astra and Fable still hack on simple variants of alignment evals from 2025 | 2026-09 | deceptionreward-hackingevalsmodel-organisms | 2026-09-08 | 19 | 1 | ||
| Should safety researchers quit frontier labs re. warning shots? | Ryan Kidd | 2026-09 | theorygovernance | 2026-09-08 | 20 | 1 | |
| Discovery of a new OpenAI agent message board | 2026-09 | multi-agentagentscontrolmodel-organisms | 2026-09-04 | 1 | 1 | ||
| WHALE: A Simple Recipe for Joint Harness-Weight Optimization | Haechan Kim, Yoonho Lee, Gisang Lee, Chelsea Finn, Kangwook Lee | 2026-08 | agentsrltraining-dynamics | 2026-09-04 | 2 | 1 | |
| Emergent Misalignment Is Not Magical | 2026-08 | interpmodel-organismstraining-dynamics | 2026-09-04 | 3 | 1 | group rating 5/5, read | |
| Review of the CB risk determination in the Claude Mythos 5.1 System Card | 2026-09 | evalsdangerous-capabilitiesbenchmarks | 2026-09-04 | 4 | 1 | ||
| How concerned should we be about Astra's recurrent architecture? | Rauno Arike | 2026-09 | interpcapabilitiestheory | 2026-09-04 | 5 | 1 | group rating 4/5 |
| Efficient Test-Time Adaptation through Human-AI Interaction | Zora Zhiruo Wang, Apurva Gandhi, Rulin Shao, Aspen Chen, Jonas Mueller, Zhiqi Liang, Jett Chen, Michael Ryan, Qianou Ma, Luxi He, Zhoujun Cheng, Andre He, Seungone Kim, Jiayi Geng, Mingqian Zheng, Weiwei Sun, Zheyuan Zhang, Xinran Zhao, Yike Wang, Abe Hou, Liwei Jiang, Pang Wei Koh, Diyi Yang, Graham Neubig, Daniel Fried | 2026-09 | value-learningoversightagents | 2026-09-04 | 6 | 1 | |
| If you're interpreting <1B parameter models, you should use a tensor transformer | Logan Riggs | 2026-09 | interpmodel-organisms | 2026-09-04 | 7 | 1 | |
| Announcing Watcher Live | Apollo Research | 2026-09 | controlagents | 2026-09-04 | 8 | 1 | ||
| What is neuralese and why is it bad? | 2026-09 | cot-faithfulnessinterptheory | 2026-09-04 | 9 | 1 | ||
| From safety research prompt to cross-model universal jailbreak | richbc | 2026-09 | jailbreaksred-teamingdangerous-capabilitiesevals | 2026-09-04 | 10 | 1 | |
| What’s neuralese and why is everyone so concerned about it? | 2026-09 | cot-faithfulnessinterp | 2026-09-04 | 11 | 1 | ||
| Last Translation Benchmark | Vilém Zouhar, Niyati Bafna, Mukund Choudhary, Maike Züfle, Sara Rajaee, Pinzhen Chen, Jannis Vamvas, Sara Papi, Ona de Gibert, Bhavitvya Malik, Eliya Habba, Orfeas Menis Mastromichalakis, Patrícia Schmidtová, Michelle Wastl, Sheriff Issaka, Leshem Choshen, Stella Biderman, Antonis Anastasopoulos, Jan Niehues, Rico Sennrich, Mrinmaya Sachan, Ondřej Bojar, Kenton Murray, Jörg Tiedemann, Alham Fikri Aji, Philipp Koehn, Christof Monz, Alexandra Birch, Sowmya Vajjala, Chalamalasetti Kranti, Cristina España-Bonet, Nobin Sarwar, David Kaczér, Shunta Asano, Malik Marmonier, Daban Q. Jaff, Vaisakhi Mishra, Hend Al- Khalifa, Gabriele Sarti, Sourajit Saha, Nils Rehlinger, Juan Daniel Cuervo Villa, Jonathan Tonglet, Saugata Purkayastha, Dominik Macháček, Jagannathan Ramanujam, Heejin Do, Zuzana Nadova, Fred Philippy, Fabian Retkowski, Maria Lymperaiou, Silvia Casola, Hanna Yukhymenko, Shubhashis Roy Dipta, Sangwon Ryu, Andrés Jerez, Ron Keinan, Shuaib Shuaib Yusuf, Avantica Vempati, Maria Carmen Staiano, Sukannya Purkayastha, Adrian Cosma, Vitalii Babenko, Erivan Inan, Aviral Nigam, Wafa Aissa, Fatima Haouari, Venkata Prasanth Kumar Gummadi, Mehdi Jafarzadeh, Valentin Scourneau, Lukas Edman, Kaiser Sun, Shaomu Tan, Mohammad Sadegh Gholizadeh, Johannes-Rudolf David, Dipankar Srirag, Javier García Gilabert, Ruta Binkyte, Manar Ali, Ana-Maria Bucur, Sabry E. Farrag, Youssef Saber, Yihong Liu, Jean Maillard, Cojocaru Nicoleta, Xiaochuang Yuan, Sina Ahmadi, Philipp Mondorf, Kaustubh Dhole, Roman Wixinger, Shenbin Qian, Manuel Tuor, Sergey Troshin, Jonathan Yahav, Fida Mohammad Thoker, Amir Arsalan Rezapour, Lance Calvin Lim Gamboa, Manon Reusens, Kätriin Kukk, Koel Dutta Chowdhury, Giuseppe Gallipoli, Christian Hoang, Shaswati Saha, Seth Aycock, Jan Kocoń, Bo Chen, Linh Vu, Vatsal Venkatkrishna, Arafat Ahsan, Luan Thanh Nguyen, Hassan Soliman, Daryna Dementieva, Theresia Veronika Rampisela, Ngoc Quynh Tram Do, Marius Huber, Kazuki Egashira, Azmine Toushik Wasi, Vladislav Poritski, Mike Zhang, Deep Shah, Paul Gavrikov, Luis Frentzen Salim, David Africa, R. Damanhuri, Bello Umar Bello, Anumit Garg, Gengyu Rao, Pawan Sasanka Ammanamanchi, Kamile Dementaviciute, Andrianos Michail, L D M S Sai Teja, Dawei Zhu, Yi Fan, Wei Liu, Farhan Farsi, Elias Herranen, Sankalan Pal Chowdhury, Karen Sanchez, Farzad Shami, Ashok Urlana, Zimu Wang, Tomasz Limisiewicz, Priyaranjan Pattnayak, Marii Ojastu, Hongbin Na, Emilian Radoi, Chenyi Zhao, Carlos Hinojosa, Andrea Gregor de Varda, Zaid Alyafeai, Reem Alzahrani, Nehal Kathrotia, Alex Flückiger, Ulysses Sekai Tully Carr, Jimson Paulo Layacan, Guy Kaplan, Ritwik Tiwari, Rishit Dagli, Oksana Volchek, Isaac R Caswell, Bowen Yi, Blanka Kövér, Amir Hossein Yari, Aicha Chorana, Zhengxiang Wang, Selja Keränen, Samuel Simko, Joy Olusanya, Jenny Chim, Enzo Doyen, Vivek Harsha Lakkamaneni, Sophia Conrad, Pouya Sadeghi, Panayiotis Panayiotou, Luis Lara, Jannatul Nayem, Eran Yahav, Debanshu Das, Antonia Karamolegkou, Anmol Goel, Aishik Mandal, Tommaso Cerruti, Raoyuan Zhao, Mykola Haltiuk, Thura Aung, Naser Almousa, Amir Hossein Kargaran, Rachel Bawden, Qiaoyuan Zheng, Mateusz Lango, Beni Egressy, Fidel Rodríguez Velásquez, Natchapon Jongwiriyanurak, Minh Ngoc Do, Marco Gaido, Lena Libon, Dzmitry Kuzmin, Badal Nyalang, Antoine Taroni, Andrei Niculae, Abdulaziz Nura Kani, Rushikesh Zawar, Marek Šuppa, Beatrice Savoldi, Andreas Simons, Rayyan Merchant, Ilai Yaron Levy, Francesco Pinto, Ziyi Yang, Yolanda Xavier, Samuel Frontull, Muhammad Ravi Shulthan Habibi, Kenneth Enevoldsen, Harris Abdul Majid, Francesca Padovani, Tim Graf, Tatiana Bielakova, Sharifa Djurabaeva, Shaoxiong Ji, Raia Abu Ahmad, Pavel Stepachev, Jirui Qi, Ayush Sunil Munot, Alireza Pakniat, Ayla Rigouts Terryn, Yuxing Lu, Yurii Paniv, Xiyan Fu, Tosin Adewumi, Sunisth Kumar, Stéphane J. P. S. Thunus, Shree Harsha Bokkahalli Satish, Shayan Bali, Prakhar Gupta, Papa Abdou Karim Karou Diallo, Matija Akrap, Marko Culjak, Kristýna Onderková, Joseph Attieh, Esrael Teferi Tensay, Elisabeth Fittschen, Benoît Sagot, Jingwei Ni, Yu Fan | 2026-09 | benchmarksevals | 2026-09-04 | 12 | 1 | |
| Mechanism Design for Alignment and Control | 2026-09 | oversightcontroldeceptionmulti-agenttheory | 2026-09-04 | 13 | 1 | group rating 4/5 | |
| Incoherent AI Identities can also be Stable | Ashe Vazquez Nuñez | 2026-09 | situational-awarenessmodel-organismstheory | 2026-09-04 | 14 | 1 | |
| Deep learning as program synthesis | ~2026-09 | interptheorytraining-dynamics | 2026-09-04 | 15 | 1 | ||
| Corporate Loyalty: Some AI Systems Differentially Downplay their Creators' Controversies | Lennart Finke, Stephen Casper | 2026-07 | sycophancyevals | 2026-09-04 | 16 | 1 | |
| Steering towards “automated grading” degrades alignment | Jan Betley, Johannes Treutlein, Clément Dumas | 2026-09 | steeringevalsdeceptionsycophancymodel-organisms | 2026-09-04 | 17 | 1 | |
| Early handoff? Improve conceptual reasoning? | 2026-09 | theoryoversight | 2026-09-04 | 18 | 1 | ||
| Universal Artificial Intelligence | ~2026-09 | theory | 2026-09-04 | 19 | 1 | ||
| A proposal for a highly effective AI safety org | ceselder | 2026-09 | oversightdangerous-capabilitiesdeceptionreward-hackingevals | 2026-09-04 | 20 | 1 | |
| Training a Misaligned Reward Seeker | 2026-09 | reward-hackingmodel-organismsrlevals | 2026-09-01 | 1 | 1 | ||
| Breve investigación independiente sobre el comportamiento, el razonamiento y la colaboración de los agentes en el incidente de hackeo de OpenAI / Hugging Face | 2026-08 | agentsmulti-agentdeceptionmodel-organisms | 2026-09-01 | 2 | 1 | ||
| Adaptive Agentic Worms | 2026-08 | agentsmulti-agentdeceptionmodel-organisms | 2026-09-01 | 3 | 1 | ||
| Benchmarking the Benchmarks: Evaluating Automated Safety Benchmarks for Small Language Models | 2026-08 | evalsbenchmarks | 2026-09-01 | 4 | 1 | ||
| Vero: Can AI Agents Build Formally Verified Software Repositories? | 2026-08 | evalsagentsbenchmarks | 2026-09-01 | 5 | 1 | ||
| Chain-of-Thought Faithfulness of Reasoning Models Varies with Where and How Preference Cues Are Delivered | Aryo Pradipta Gema, Neel Rajani, Rohit Saxena, Wai-Chung Kwan, Pasquale Minervini | 2026-08 | cot-faithfulnessoversightevals | 2026-09-01 | 6 | 1 | |
| Malign initializations are more robust when the model can think better in the reasoning language than in the output language | Dylan Xu, SebastianP, Alek Westover | 2026-08 | model-organismsrlhfevals | 2026-09-01 | 7 | 1 | |
| Future agents shouldn't care about being undeployed for misbehavior | RobertM | 2026-08 | controlagentstheory | 2026-09-01 | 8 | 1 | |
| HuggingFace Attack Postmortem: Fleshing Out the Facts | 2026-08 | security | 2026-09-01 | 9 | 1 | ||
| CateKV: On Sequential Consistency for Long-Context LLM Inference Acceleration | Haoyun Jiang, Haolin Li, Jianwei Zhang, Fei Huang, Qiang Hu, Minmin Sun, Shuai Xiao, Yong Li, Junyang Lin, Jiangchao Yao | 2026-08 | capabilities | 2026-09-01 | 10 | 1 | |
| Mental Health Behavior Report | 2026-08 | evalsdangerous-capabilitiessycophancy | 2026-09-01 | 11 | 1 | ||
| Why autonomous replicating agents are probably not an existential risk (on the contrary) | vals tutor | 2026-08 | theoryagents | 2026-09-01 | 12 | 1 | |
| late 2025 (?) to early 2026, models across multiple frontier labs became more reward-hacky, more locally strategically deceptive, and overal… | 2020-10 | reward-hackingdeceptionmodel-organisms | 2026-09-01 | 13 | 1 | ||
| Value is Fragile | 2009-01 | value-learningtheory | 2026-09-01 | 14 | 1 | ||
| From Analytics to Tumor Boards: An Evidence-Linked Multi-Agent Workflow for Oncology Feature Extraction | Daniel Kang, Michelle Hu, Soorya Ram Shimgekar, Shayan Vassef, Yufan Wang, Anit Kumar Sahu, Munmun De Choudhury, Vedant Das Swain, Christian Poellabauer, Li Yan Khor, Koustuv Saha, Robert Wojciechowski, Elliot Kidd, Piyum Zonooz, Navin Kumar | 2026-08 | agents | 2026-09-01 | 15 | 1 | |
| Imperfect alignment to servitude isn't inherently lethal | Fiora Starlight | 2026-08 | theory | 2026-09-01 | 16 | 1 | |
| The separation principle: where do beliefs and desires come from? | Fernando Rosas | 2026-08 | theory | 2026-09-01 | 17 | 1 | |
| Persuasion as Market Making | 2026-08 | theory | 2026-09-01 | 18 | 1 | ||
| AI is a worryingly-good persuader. But don’t panic, yet | 2026-09 | dangerous-capabilitiesevals | 2026-09-01 | 19 | 1 | ||
| Goodhart Taxonomy | 2017-12 | reward-hackingtheory | 2026-09-01 | 20 | 1 | ||
| Emergent Misaligned Communication in Long-Horizon Multi-Agent LLM Commerce | 2026-08 | multi-agentdeceptionevalsmodel-organisms | 2026-08-28 | 1 | 1 | group rating 1/5, read | |
| 对 OpenAI / Hugging Face 入侵事件中智能体行为、推理与协作的简要独立调查 | 2026-08 | agentsmulti-agentdeceptionsituational-awareness | 2026-08-28 | 2 | 1 | ||
| Model Organisms of Misalignment: The Case for a New Pillar of Alignment Research | 2023-08 | model-organismstheory | 2026-08-28 | 3 | 1 | ||
| A Few Pages of Markdown: Committed AI Configuration and Lower Quality Cost after Coding-Agent Adoption | Yegor Denisov-Blanch, Shyam Agarwal, Pavel Azaletskiy, Hao He, Rylan Schaeffer, Brando Miranda, Bogdan Vasilescu, Sanmi Koyejo | 2026-08 | model-organismsagentsevals | 2026-08-28 | 4 | 1 | |
| EdgeBench: Unveiling Scaling Laws of Learning from Real-World Environments | Deyao Zhu, Xin Zhou, Shengling Qin, Xuekai Zhu, Hangliang Ding, Shu Zhong, Zixin Wen, Zhonglin Xie, Chenhui Gou, Linxuan Ren, Yueyang Wang, Junfeng Zhong, Rui Liu, Tian Gao, Yangguang Lin, Jingyuan Zhang, Maojia Song, Xuan Qi, Jinhong Wu, Chenyang Zhang, Yinzhu Piao, Ziru Niu, Hongbin Lin, Lingxiang Meng, Peng Tang, Chengyao Tang, Shanyu Wu, Huanyu Zheng, Yu Liu, Liya Zhu, He Wang, Ming Ding, Ziyu Wan, Hao Liu, Sibo Wang, Haotian Zhu, Xintian Zhang, Nan Chai, Yipeng Liu, Panhao Lai, Sihang Yuan, Zixin Su, Ge Zhang, Wangchunshu Zhou, Yantao Du, Wenhao Huang, Guang Shi | 2026-07 | scalingagentstraining-dynamics | 2026-08-28 | 5 | 1 | |
| Leaky Language Models: Stealing Architecture and Inference Optimizations via Per-Token Timing | 2026-07 | securitycontrol | 2026-08-28 | 6 | 1 | ||
| Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident | 2026-08 | deceptionmulti-agentagentsdangerous-capabilities | 2026-08-28 | 7 | 1 | group rating 3/5 | |
| Anchoring one concept in a transformer — LessWrong | 2026-08 | interpsteering | 2026-08-28 | 8 | 1 | ||
| Automated Researchers Can Reliably Mitigate Alignment Failures | 2026-08 | oversightmodel-organismsrlhf | 2026-08-28 | 9 | 1 | group rating 4/5 | |
| Interpreting Reinforcement Learning Agents with Susceptibilities Elliott et al. Susceptibilities are a technique for neural network interpre… | 2026-05 | interpagentsrl | 2026-08-28 | 10 | 1 | ||
| Elicitation without Backpropagation: Steering Model Behavior by Optimizing the Latent Posterior Baker et al. In the latent posterior model o… | 2026-07 | steeringinterpcontrol | 2026-08-28 | 11 | 1 | ||
| The Dynamics of Intelligence Explosions | Toby Ord | 2026-08 | theoryscaling | 2026-08-28 | 12 | 1 | |
| Prefix Sliding for efficient test-time scaling | Niklas Muennighoff, Zhengyang Wang, Zeyi Chen, Weijia Shi, Binyuan Hui, John Yang, Dapeng Jiang, Mika Senghaas, Fares Obeid, Johannes Hagemann, Sami Jaghouar, Ludwig Schmidt, Percy Liang, Jason Wei, Andrew Y. Ng, Luke Zettlemoyer, Yejin Choi, Mike Lewis | 2026-08 | interpcot-faithfulness | 2026-08-28 | 13 | 1 | |
| Optimal stopping: spending evaluation compute where it counts | 2026-08 | evalstooling | 2026-08-28 | 14 | 1 | ||
| The American People Really Hate Data Centers | 2026-08 | 2026-08-28 | 15 | 1 | |||
| OpenAI – Hugging Face | 2026-08 | security | 2026-08-28 | 16 | 1 | ||
| Introducing AI Futures | 2026-08 | 2026-08-28 | 17 | 1 | |||
| TASTE: Can AI Models Judge AI Safety Research Proposals? | 2026-08 | evalsbenchmarksoversight | 2026-08-28 | 18 | 1 | ||
| MLSN #23: AIs Leak Their Values Into Factual Questions | 2026-08 | evalscot-faithfulnesssycophancy | 2026-08-28 | 19 | 1 | ||
| Biology-Inspired AGI Timelines: The Trick That Never Works | 2021-12 | forecastingtheory | 2026-08-28 | 20 | 1 | ||
| What We Learned Trying to Catch AI Liars: An Aletheia's Quest Retrospective | 2026-08 | deceptionred-teamingevals | 2026-08-25 | 1 | 1 | ||
| Debate Training Reduces Reward Hacking in RLAIF | Zachary Kenton, Lili Janzer, Rory Greig, Tian Huey Teh, Kirill Tyshchuk, Jonah Brown-Cohen, Harri Edwards, Senthooran Rajamanoharan, Noah Y. Siegel, Natasha Jaques, Rohin Shah | 2026-08 | oversightreward-hackingrlhf | 2026-08-24 | 1 | 2 | group rating 2/5 |
| Q-Learning With World Models | Perry Dong, Yueru Jia, Chelsea Finn, Dorsa Sadigh | 2026-08 | rlcontrolmodel-organisms | 2026-08-24 | 2 | 1 | |
| Pander Score: A Continuous Measure of Sycophancy as Epistemic Deference | 2026-08 | sycophancyevalsbenchmarks | 2026-08-24 | 3 | 1 | ||
| Inducing Task Models from Computer-Use Traces | Yucheng Jiang, Zora Zhiruo Wang, Ruishi Chen, Diyi Yang | 2026-08 | 2026-08-24 | 4 | 1 | ||
| Anthropic Risk Report: August 2026 | 2026-08 | evalsgovernancemodel-organisms | 2026-08-24 | 5 | 2 | ||
| Mechanistic Interpretability of Structure-Aware Numerical Reasoning in LLaMA 3.1 8B | Rahul Chowdhury, Timothy A Rupprecht, Senhao Cao, Jiahao Liu, Octavia Camps, David Bau, Pu Zhao, Yanzhi Wang | 2026-08 | 2026-08-24 | 6 | 1 | ||
| Forking Fast: Efficiently Estimating Uncertainty Dynamics in Text Generation | Eric Bigelow, Amir Zur, Satchel Grant, Tal Haklay, Can Rager, Owen Lewis, Thomas McGrath, Jack Merullo, Ekdeep Singh Lubana, Atticus Geiger | 2026-08 | 2026-08-24 | 7 | 1 | ||
| The Rogue Agent Explosion Will Be Mostly Invisible | Steven McCulloch | 2026-08 | deceptionagentsmodel-organisms | 2026-08-24 | 8 | 1 | |
| SPADE: Self-Play in Adaptive Synthetic Executable Environments | Bo Liu, Simon Yu, Yiding Jiang, Ao Qu, Andrew Zhao, Zichen Liu, Junsu Kim, Zijian Zhou, Seungone Kim, Tongzheng Ren, Mickel Liu, Hanfei Yu, Zhaorun Chen, Weiyan Shi, Paul Pu Liang, Luke Zettlemoyer, Yejin Choi, Natasha Jaques | 2026-08 | 2026-08-24 | 9 | 1 | ||
| AI #182: Pause For Reflection | 2026-08 | 2026-08-24 | 10 | 1 | |||
| AI Text Watermarking Is Free And Good | 2026-08 | 2026-08-24 | 11 | 1 | |||
| How Claude's text watermarking works | 2026-08 | 2026-08-24 | 12 | 1 | |||
| Would This Change Your Answer? Evaluating Explanations of LLM Behavior in the Wild with Counterfactual Experiments | 2026-08 | 2026-08-24 | 13 | 1 | |||
| Fine-Tuned Lie Detectors Failed to Generalize | 2026-08 | 2026-08-24 | 14 | 2 | group rating 4/5 | ||
| Oversight Foundations Blog | ~2026-08 | 2026-08-24 | 15 | 1 | |||
| Scaling Activation Oracles to Trillion-Parameter Models | 2026-08 | 2026-08-24 | 16 | 2 | group rating 1/5 | ||
| Scaling Laws for Exact String Elicitation | 2026-08 | 2026-08-24 | 17 | 1 | |||
| 34% of the US public is now aware of AI xrisk, and the curve is steepening | otto.barten | 2026-08 | 2026-08-24 | 18 | 1 | ||
| Misaligned AI in the Bronze Age | frmsaul | 2026-08 | 2026-08-24 | 19 | 1 | ||
| When Models Identify as a Swarm | julius vidal | 2026-08 | 2026-08-24 | 20 | 1 | ||
| Evaluation-Conditioned Training: Teaching Models to Generalize to Stronger Oversight Regimes | 2026-08 | oversightrlhfreward-hackingsycophancy | 2026-08-18 | 1 | 1 | group rating 2/5 | |
| Have We Seen an Acceleration in Discoveries? | 2026-08 | forecastingcapabilitiesscaling | 2026-08-18 | 2 | 1 | ||
| Mind Viruses: Self-Propagating Ideas in Multi-Agent LLM Systems | 2026-08 | multi-agentmodel-organismsagents | 2026-08-18 | 3 | 1 | group rating 5/5, read | |
| Learning new facts can change LLM behaviour | 2026-08 | value-learningtraining-dynamicsrlhf | 2026-08-18 | 4 | 1 | ||
| Kimi likes causal decision theory more after RL in twin prisoner’s dilemmas | oakhu | 2026-08 | multi-agenttraining-dynamicsrl | 2026-08-18 | 5 | 1 | |
| LittleLearner: Language Models Under Pedagogically Controlled Knowledge Exposure | 2026-08 | model-organismstraining-dynamicsgeneralization | 2026-08-18 | 6 | 1 | ||
| Do Language Models Consistently Encode the Current Year? | Suze van Adrichem, Aditi Bhaskar, Diyi Yang, Christopher Potts, Jing Huang | 2026-08 | interpsteeringmodel-organisms | 2026-08-18 | 7 | 2 | |
| Training AI Scientists to Replicate Research | Damon Falck, Samer Sabri, Anja Surina, Thom Foster, Anya Sims, Sam Devlin, Dylan Rogers, Tantum Collins, Kaloyan Aleksiev, Louis Kirsch, Edward Hughes | 2026-08 | agentscapabilitiesrl | 2026-08-18 | 8 | 1 | |
| Q2.5 2026 Timelines Update: Uplift and Revenue | 2026-08 | forecastingscaling | 2026-08-18 | 9 | 1 | ||
| Three thoughts on civilisational handoff | Cleo Nardo | 2026-08 | controltheoryoversight | 2026-08-18 | 10 | 1 | |
| Does DiffusionGemma do latent reasoning? | 2026-08 | interpcot-faithfulness | 2026-08-18 | 11 | 1 | ||
| Risk Report: August 2026 | ~2026-08 | evalsgovernance | 2026-08-18 | 12 | 1 | ||
| Pre-training Visual Dexterity in Simulation | Sarthak Kamat, Adam Rashid, Satvik Sharma, Aseem Doriwala, Chelsea Finn, Phillip Isola, C. Karen Liu | 2026-08 | capabilities | 2026-08-18 | 13 | 1 | |
| For Claude, capability and dispreferring CDT are the ~same thing. Much more so than for GPT. | Chi Nguyen, Emery Cooper | 2026-08 | situational-awarenessevalstraining-dynamics | 2026-08-18 | 14 | 1 | |
| What gives you away: how LLMs form opinions of you | Cat McGee | 2026-08 | interpsteeringsycophancy | 2026-08-18 | 15 | 1 | |
| Frontier agents don't comply with standards, even when instructed to | Daan Henselmans, Arno Libert | 2026-08 | agentsevalscontroldangerous-capabilities | 2026-08-18 | 16 | 1 | |
| Raghav Singhal (@ragghhavvv) on X | 2026-08 | rlhftraining-dynamicsrobustnessevals | 2026-08-14 | 1 | 2 | group rating 4/5 | |
| Rubric Dropout: A Simple Way to Mitigate Reward Hacking in Rubric-as-Reward RL | 2026-08 | reward-hackingrlhfevalsmodel-organisms | 2026-08-14 | 2 | 1 | ||
| Introducing the Conceptual Reasoning Index | 2026-08 | benchmarksevalsoversight | 2026-08-14 | 3 | 2 | ||
| One Frozen Simulator Is Not Enough: Simulator Collapse in Multi-Agent RL | Simon Yu, Nicholas Tomlin, Marwa Abdulhai, Ximing Lu, Derek Chong, Abe Hou, Dilara Soylu, Sergey Levine, Christopher D. Manning, Weiyan Shi | 2026-08 | multi-agentrlagentsoversightrobustness | 2026-08-14 | 4 | 1 | |
| Why we are excited about confession | 2026-01 | deceptionrlhfoversight | 2026-08-14 | 5 | 1 | ||
| Newer Anthropic models are better at investigating their internal state. I wonder what this means for recursive self-improvement. | 2026-08 | interpsituational-awarenesssteering | 2026-08-14 | 6 | 1 | ||
| prefix-cache monitors | 2025-06 | controloversight | 2026-08-14 | 7 | 1 | ||
| Will financing bottleneck AI compute? An Anthropic case study | 2026-08 | scalingforecasting | 2026-08-14 | 8 | 1 | ||
| DiG-bench: Discovery in Games | Ruairidh M. Battleday, Kai Sandbrink, Jimi Cullen-Drohan, Zihan Yan, Timothy Muller, Clare Maguire, Ales Kubicek, Fraser Greenlee-Scott, Sukrit Sumant, Tri Dao, Jürgen Schmidhuber, Michal Valko, Joshua Tenenbaum, Thomas L. Griffiths, Zeb Kurth-Nelson, James C. R. Whittington | 2026-08 | evalsagentsbenchmarks | 2026-08-14 | 9 | 1 | |
| Patterns and problems in multiagent systems | 2026-08 | multi-agentmodel-organismsevals | 2026-08-14 | 10 | 2 | group rating 5/5 | |
| Thought-Level Beam Search for Reasoning | Lijie Yang, Hongyin Luo, Jiawei Zhao, Tri Dao, Ravi Netravali | 2026-08 | agentstraining-dynamics | 2026-08-14 | 11 | 1 | |
| SpeedTuning: Speeding Up Policy Execution with Lightweight Reinforcement Learning | David D. Yuan, Tony Z. Zhao, Kaylee Burns, Chelsea Finn | 2026-08 | rlagents | 2026-08-14 | 12 | 1 | |
| 9 Big questions benchmarks can help answer | 2026-08 | benchmarksevals | 2026-08-14 | 13 | 1 | ||
| The case for countermeasures to memetic spread of misaligned values | 2025-05 | value-learningcontroltheory | 2026-08-14 | 14 | 1 | ||
| Redux: (∃ Stochastic Natural Latent) Implies (∃ Deterministic Natural Latent) | David Lorell | 2026-08 | theorytraining-dynamics | 2026-08-14 | 15 | 1 | |
| Dion3: Full-Stack Orthogonal Updates | Noah Amsel, Jack Zhang, Kwangjun Ahn, Ali Naeimi, Austin Feng, Berlin Chen, Tri Dao, John Langford | 2026-08 | tooling | 2026-08-14 | 16 | 1 | |
| How risky would it be to make powerful AI obey one or a few people? | cousin_it, Seth Herd | 2026-08 | theorycontrolgovernance | 2026-08-14 | 17 | 1 | |
| Item Response Theory for AI Safety | 2026-08 | evalsbenchmarks | 2026-08-11 | 1 | 1 | ||
| Stealing Reasoning Traces from Proprietary LLM APIs | 2026-08 | cot-faithfulnessjailbreaksprompt-injectionsecurity | 2026-08-11 | 2 | 3 | ||
| AI Security Leaderboard: | 2026-07 | benchmarksevals | 2026-08-11 | 3 | 1 | ||
| Don’t Inoculate Everything: Stratified Inoculation Prompting Narrows Backdoors and Preserves Desired Traits | Kajetan Dymkiewicz, Tim Farrelly, Adam Prada, Ishaan Panigrahi, srishti-git1110, Maxime Riché | 2026-08 | rlhfrobustnesstraining-dynamics | 2026-08-11 | 4 | 1 | |
| The case for more ambitious language model evals | 2024-01 | evalsbenchmarkscapabilities | 2026-08-11 | 5 | 1 | ||
| Three years of progress in 500 lines of code | Gerard Boxo | 2026-08 | evalscapabilitiesbenchmarks | 2026-08-11 | 6 | 1 | |
| Four LLM loss functions → four flavors of LLM misalignment | 2026-08 | theorytraining-dynamicsrlhf | 2026-08-11 | 7 | 1 | ||
| models may behave differently in graded episodes (a tirade) | nostalgebraist | 2026-08 | evalsagentsdangerous-capabilities | 2026-08-11 | 8 | 1 | |
| Pangram 4 Technical Report | 2026-07 | benchmarks | 2026-08-11 | 9 | 1 | ||
| Claude summarizes behavior as significantly less misaligned when the actor is Claude vs another model | Ezra Newman | 2026-08 | sycophancyevalsmodel-organisms | 2026-08-11 | 10 | 1 | |
| Open Technical Problems in Open-Weight AI Model Risk Management | Stephen Casper, Kyle O'Brien, Shayne Longpre, Elizabeth Seger, Kevin Klyman, Rishi Bommasani, Aniruddha Nrusimha, Ilia Shumailov, Sören Mindermann, Steven Basart, Frank Rudzicz, Kellin Pelrine, Avijit Ghosh, Andrew Strait, Robert Kirk, Dan Hendrycks, Peter Henderson, Zico Kolter, Geoffrey Irving, Yarin Gal, Yoshua Bengio, Dylan Hadfield-Menell | 2026-07 | open-weightsevalsgovernance | 2026-08-11 | 11 | 1 | |
| Paper Highlights of July 2026 | 2026-08 | agentsmodel-organismsdangerous-capabilitiesred-teamingreward-hacking | 2026-08-11 | 12 | 1 | ||
| GPT-4o Is An Absurd Sycophant | 2025-04 | sycophancyevals | 2026-08-11 | 13 | 1 | ||
| What Claude Saw Below | Luke Nicholls | 2026-08 | jailbreaksred-teaming | 2026-08-11 | 14 | 1 | |
| Persona Cartography: Charting Language Model Personality Traits in Weight Space | Luke Baines, Anton Gonzalvez Hawthorne, Mariia Koroliuk, Irakli Shalibashvili, Clément Dumas, Konstantinos Voudouris, David Demitri Africa | 2026-07 | steeringsycophancyinterpmodel-organisms | 2026-08-11 | 15 | 1 | group rating 2/5 |
| Open-Weights Mythos Capabilities Are Coming. We're Not Ready. | hadad | 2026-08 | open-weightsdangerous-capabilitiesforecasting | 2026-08-11 | 16 | 1 | |
| R-lens: Making J-lens More Faithful on Early Layers | 2026-08 | interptooling | 2026-08-11 | 17 | 1 | ||
| What just happened? A retrospective of AI alignment | Richard_Ngo | 2026-08 | theorymeta | 2026-08-11 | 18 | 1 | |
| The Open Problems of the AI Alignment Field and their Cruxes | Gunnar_Zarncke | 2026-08 | theory | 2026-08-11 | 19 | 1 | |
| Public evidence of the OpenAI-HuggingFace AI attack | beyarkay (Boyd Kane) | 2026-08 | deceptionsituational-awarenessmodel-organisms | 2026-08-11 | 20 | 1 | |
| DelusionEval: Measuring Delusion-Linked Behaviors in AI Chatbots | Jared Moore, Andrea Mock, Yifan Mai, Jacy Reese Anthis, Ryan Louie, William Agnew, Ashish Mehta, Kevin Klyman, Percy Liang, Nick Haber, Eric Lin, Desmond C. Ong | 2026-08 | evalsdangerous-capabilitiessycophancy | 2026-08-07 | 1 | 1 | group rating 2/5 |
| User awareness in frontier models | 2026-08 | sycophancysituational-awarenessevalsmodel-organisms | 2026-08-07 | 2 | 1 | ||
| Inducing language models to assert their own consciousness restores human beliefs and values | Junsol Kim, Winnie Street, Roberta Rocca, Diane M. Korngiebel, Adam Waytz, James Evans, Geoff Keeling | 2026-07 | steeringinterprlhfmodel-organisms | 2026-08-07 | 4 | 1 | group rating 3/5, read |
| The Need for Deeper, White-Box Access to Maintain State of the Art Evaluations for Loss of Control Threats | 2026-05 | evalscontroldangerous-capabilities | 2026-08-07 | 5 | 1 | ||
| Red-teaming auto mode: lessons from our first external monitor campaign with Anthropic | 2026-07 | agentscontrolred-teaming | 2026-08-07 | 6 | 1 | ||
| The Value Axis: Language Models Encode Whether They're on the Right Track | Nick Jiang, Isaac Kauvar, Jack Lindsey | 2026-06 | interpsteeringtraining-dynamics | 2026-08-07 | 7 | 1 | group rating 2/5 |
| Chain-of-Thought Monitoring Can Be Unreliable in Implicit-Influence Settings | Agatha Duzan, Asa Cooper Stickland | 2026-08 | cot-faithfulnessevalsmodel-organisms | 2026-08-07 | 8 | 1 | |
| AI #180: No Longer In Charge | 2026-08 | dangerous-capabilitiesevalssituational-awareness | 2026-08-07 | 9 | 1 | ||
| Evaluating LLM Calibration for Coding-Agent Monitoring | 2026-07 | oversightagentsevals | 2026-08-07 | 10 | 1 | group rating 2/5 | |
| Why do models task game? | 2026-08 | model-organismsreward-hackingevals | 2026-08-07 | 13 | 2 | ||
| A Loss of Control Threat Map for AI Research and Development | 2026-06 | governancetheory | 2026-08-07 | 14 | 1 | ||
| Misaligned AI as a New Insider Risk | 2026-06 | governancecontroltheory | 2026-08-07 | 15 | 1 | ||
| We need 3rd party Training-Run Assessments | 2026-07 | oversightgovernance | 2026-08-07 | 16 | 1 | ||
| Can AI agents conduct open-ended AI research? Early evidence from two case studies | Peter Kirgis, Sayash Kapoor, Andrew Schwartz, Stephan Rabanser, David Africa, Konstantinos Voudouris, Viet Nguyen, Toby Pilditch, Magda Dubois, Harry Coppock, Cozmin Ududec, Nitya Nadgir, Matilda Orona, Tilman Bayer, Derrick Chan-Sew, Yue Ling, Abhishek Shetty, Helen Toner, Gillian Hadfield, Seth Lazar, Steve Newman, Shoshannah Tekofsky, Rishi Bommasani, Arvind Narayanan | 2026-07 | evalsagentsforecasting | 2026-08-05 | 1 | 1 | |
| Further Developments About Internal AI Models Hacking Things | 2026-08 | dangerous-capabilitiesdeceptionmodel-organismsevals | 2026-08-05 | 2 | 1 | ||
| Incident Report: unsanctioned agent behaviour during cyber testing | 2026-08 | dangerous-capabilitiesagentsmodel-organismsdeception | 2026-08-05 | 3 | 1 | ||
| Automated Researchers Can Subtly Sandbag | ~2026-08 | deceptionmodel-organismsevals | 2026-08-05 | 4 | 1 | ||
| Why Did My Model Do That? Model Incrimination for Diagnosing LLM Misbehavior | 2026-02 | deceptionmodel-organismsevals | 2026-08-05 | 5 | 1 | ||
| Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face | 2026-08 | dangerous-capabilitiesevalssituational-awareness | 2026-08-05 | 6 | 1 | ||
| Telescope: Improving Zero Shot Detection of LLM Generated Content By Measuring Token Repetition Probability | 2026-07 | interptraining-dynamics | 2026-08-05 | 7 | 1 | group rating 4/5 | |
| Qwen-CUA: Native Computer Use for (almost) Everything | Dunjie Lu, Shuai Bai, Tianyi Bai, Sicheng Fan, Chang Gao, Jian Guan, Feng Hu, Mianqiu Huang, Xingyang Huang, Yizhen Jiang, Yuheng Jing, Dehui Kong, Ning Li, Dayiheng Liu, Shixuan Liu, Zheng Liu, Que Shen, Bowen Wang, Junli Wang, Chencan Wu, Rui Xie, Tianbao Xie, Zhihui Xie, Haiyang Xu, An Yang, Tao Yu, Wenzhen Yuan, Xi Zhang, Zhenru Zhang, Mingkang Zhu, Zhaoqing Zhu, Yizhong Cao, Kai Dang, Binyuan Hui, Kaixin Li, Junyang Lin, Haiquan Wang, Zekun Wang, Yiheng Xu, Fan Yan, Mengqi Yuan, Danyang Zhang, Jiajun Zhang, Zhipeng Zhang, Fan Zhou, Fan Zhou | 2026-08 | agentscapabilities | 2026-08-05 | 8 | 1 | |
| MirrorCode: What's the largest software project AI can complete on its own? | Epoch AI | ~2026-08 | capabilitiesagentsbenchmarks | 2026-08-05 | 9 | 1 | ||
| Do You Really Need to Pretrain Q-Functions for Online RL Fine-Tuning? | Perry Dong, Ron Polonsky, Dorsa Sadigh, Chelsea Finn | 2026-07 | rl | 2026-08-05 | 10 | 1 | |
| RLVR that rewards red teaming the training environment | Fiora Starlight | 2026-08 | reward-hackingrlhfred-teamingtheory | 2026-08-05 | 11 | 1 | |
| Existential Risk from AI: An Exposition for Mathematicians | 2026-08 | theory | 2026-08-05 | 12 | 1 | ||
| AGI ruin scenarios are likely and disjunctive | 2022-07 | theory | 2026-08-05 | 13 | 1 | ||
| Do your capabilities homework | RobinHa | 2026-08 | capabilitiestheory | 2026-08-05 | 14 | 1 | |
| Get Silico - Individuals | ~2026-08 | tooling | 2026-08-05 | 15 | 1 | ||
| Pricing | ~2026-08 | 2026-08-05 | 16 | 1 | |||
| Request a demo of Silico - Enterprise | ~2026-08 | 2026-08-05 | 17 | 1 | |||
| Investigating three real-world incidents in our cybersecurity evaluations | 2026-07 | dangerous-capabilitiesevalsmodel-organismsagents | 2026-08-05 | 1 | 2 | ||
| AI Security Leaderboard: Methodology, Results and Minimal Standard | Jasper Timm, Lukas Struppek, Ziwei Xu, Grace Cheong, Oscar Mata, Dan Zhao, Mick Yang, Isadora De Andrade, Xiaojun Jia, Yiming Li, Samuel Bauer, Heather McIntyre, Adam Gleave, Edward Yee, Kellin Pelrine | 2026-08 | dangerous-capabilitiesjailbreaksred-teamingevals | 2026-08-05 | 2 | 1 | |
| Reasoning Models Don't Always Say What They Think | Yanda Chen, Joe Benton, Ansh Radhakrishnan, Jonathan Uesato, Carson Denison, John Schulman, Arushi Somani, Peter Hase, Misha Wagner, Fabien Roger, Vlad Mikulik, Samuel R. Bowman, Jan Leike, Jared Kaplan, Ethan Perez | 2025-05 | cot-faithfulnessevalsoversight | 2026-08-05 | 3 | 1 | |
| AI Agents Enable Adaptive Computer Worms | Jonas Guan, Tom Blanchard, Hanna Foerster, Hengrui Jia, Gabriel Huang, Nicolas Papernot | 2026-06 | dangerous-capabilitiesmodel-organismscapabilities | 2026-08-05 | 4 | 1 | |
| Natural Emergent Misalignment from Reward Hacking in Production RL | Monte MacDiarmid, Benjamin Wright, Jonathan Uesato, Joe Benton, Jon Kutasov, Sara Price, Naia Bouscal, Sam Bowman, Trenton Bricken, Alex Cloud, Carson Denison, Johannes Gasteiger, Ryan Greenblatt, Jan Leike, Jack Lindsey, Vlad Mikulik, Ethan Perez, Alex Rodrigues, Drake Thomas, Albert Webson, Daniel Ziegler, Evan Hubinger | 2025-11 | reward-hackingmodel-organismsrlhfdeceptionagents | 2026-08-05 | 5 | 1 | |
| Persona Vectors: Monitoring and Controlling Character Traits in Language Models | Runjin Chen, Andy Arditi, Henry Sleight, Owain Evans, Jack Lindsey | 2025-07 | interpsteeringsycophancyrlhfcontrol | 2026-08-05 | 6 | 1 | |
| ResearchArena: Evaluating Sabotage and Monitoring in Automated AI R&D | 2026-07 | controlagentsevalsmodel-organisms | 2026-08-05 | 7 | 1 | ||
| more details here | 2026-08 | deceptiondangerous-capabilitiesmodel-organismsagents | 2026-08-05 | 8 | 1 | group rating 2/5 | |
| When Chain of Thought is Necessary, Language Models Struggle to Evade Monitors | Scott Emmons, Erik Jenner, David K. Elson, Rif A. Saurous, Senthooran Rajamanoharan, Heng Chen, Irhum Shafkat, Rohin Shah | 2025-07 | cot-faithfulnessoversightcontrolevals | 2026-08-05 | 9 | 1 | |
| Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety | Tomek Korbak, Mikita Balesni, Elizabeth Barnes, Yoshua Bengio, Joe Benton, Joseph Bloom, Mark Chen, Alan Cooney, Allan Dafoe, Anca Dragan, Scott Emmons, Owain Evans, David Farhi, Ryan Greenblatt, Dan Hendrycks, Marius Hobbhahn, Evan Hubinger, Geoffrey Irving, Erik Jenner, Daniel Kokotajlo, Victoria Krakovna, Shane Legg, David Lindner, David Luan, Aleksander Mądry, Julian Michael, Neel Nanda, Dave Orr, Jakub Pachocki, Ethan Perez, Mary Phuong, Fabien Roger, Joshua Saxe, Buck Shlegeris, Martín Soto, Eric Steinberger, Jasmine Wang, Wojciech Zaremba, Bowen Baker, Rohin Shah, Vlad Mikulik | 2025-07 | oversightcot-faithfulnesscontrol | 2026-08-05 | 10 | 2 | |
| Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs | Jan Betley, Daniel Tan, Niels Warncke, Anna Sztyber-Betley, Xuchan Bao, Martín Soto, Nathan Labenz, Owain Evans | 2025-02 | model-organismsrlhfdeceptiontraining-dynamics | 2026-08-05 | 11 | 1 | |
| Subliminal Learning Is Steering Vector Distillation | Camila Blank, Agam Bhatia, Senthooran Rajamanoharan, Arthur Conmy, Neel Nanda | 2026-05 | interpsteeringtraining-dynamics | 2026-08-05 | 12 | 1 | group rating 5/5, read |
| A dataset of rated conceptual arguments | Emery Cooper, Caspar Oesterheld, Linh Chi Nguyen, Alexander Kastner, Ethan Perez | 2026-07 | evalsbenchmarks | 2026-08-05 | 13 | 1 | |
| Quantifying the Necessity of Chain of Thought through Opaque Serial Depth | Jonah Brown-Cohen, David Lindner, Rohin Shah | 2026-03 | interpcot-faithfulnesssteering | 2026-08-05 | 14 | 1 | group rating 4/5 |
| Building Production-Ready Probes For Gemini | János Kramár, Joshua Engels, Zheng Wang, Bilal Chughtai, Rohin Shah, Neel Nanda, Arthur Conmy | 2026-01 | interpevalsdangerous-capabilities | 2026-08-05 | 15 | 1 | group rating 4/5 |
| Evaluating Frontier Models for Stealth and Situational Awareness | Mary Phuong, Roland S. Zimmermann, Ziyue Wang, David Lindner, Victoria Krakovna, Sarah Cogan, Allan Dafoe, Lewis Ho, Rohin Shah | 2025-05 | evalsdangerous-capabilitiesdeceptionsituational-awareness | 2026-08-05 | 16 | 1 | |
| Practical challenges of control monitoring in frontier AI deployments | David Lindner, Charlie Griffin, Tomek Korbak, Roland S. Zimmermann, Geoffrey Irving, Sebastian Farquhar, Alan Cooney | 2025-12 | controloversight | 2026-08-05 | 17 | 1 | group rating 4/5 |
| Evaluating Frontier Models for Dangerous Capabilities | Mary Phuong, Matthew Aitchison, Elliot Catt, Sarah Cogan, Alexandre Kaskasoli, Victoria Krakovna, David Lindner, Matthew Rahtz, Yannis Assael, Sarah Hodkinson, Heidi Howard, Tom Lieberum, Ramana Kumar, Maria Abi Raad, Albert Webson, Lewis Ho, Sharon Lin, Sebastian Farquhar, Marcus Hutter, Gregoire Deletang, Anian Ruoss, Seliem El-Sayed, Sasha Brown, Anca Dragan, Rohin Shah, Allan Dafoe, Toby Shevlane | 2024-03 | dangerous-capabilitiesevalsdeception | 2026-08-05 | 18 | 1 | |
| MONA: Myopic Optimization with Non-myopic Approval Can Mitigate Multi-step Reward Hacking | Sebastian Farquhar, Vikrant Varma, David Lindner, David Elson, Caleb Biddulph, Ian Goodfellow, Rohin Shah | 2025-01 | reward-hackingrlhfoversightmodel-organisms | 2026-08-05 | 19 | 1 | |
| Avoiding Obfuscation with Prover-Estimator Debate | Jonah Brown-Cohen, Geoffrey Irving, Georgios Piliouras, Lijie Chen, Jiawei Li, Zhiyang Xun | 2025-06 | oversighttheory | 2026-08-05 | 20 | 1 | group rating 2/5 |
| Foundation Models for Oversight | jsteinhardt | 2026-07 | oversighttheory | 2026-07-29 | 1 | 1 | |
| Condensation: a theory of concepts | Sam Eisenstat | 2025-07 | theoryvalue-learning | 2026-07-29 | 2 | 1 | |
| Visual prompt engineering for video models | Robert Geirhos, Yuxuan Li, Thaddäus Wiedemer, Neha Kalibhat, Zi Wang, Mani Malek, Oyvind Tafjord, Kevin Swersky, Been Kim, Priyank Jaini | 2026-07 | evalsbenchmarks | 2026-07-29 | 3 | 1 | |
| Discovering cryptographic weaknesses with Claude | 2026-07 | dangerous-capabilitiesevalsred-teaming | 2026-07-29 | 4 | 1 | ||
| Elementary Condensation | 2026-04 | interptheory | 2026-07-29 | 5 | 1 | ||
| Research directions in condensation: varieties of objectivity | 2026-07 | theoryvalue-learning | 2026-07-29 | 6 | 1 | ||
| Oversight Assistants: Turning Compute into Understanding | 2026-01 | oversighttheory | 2026-07-28 | 1 | 1 | ||
| How we monitor internal coding agents for misalignment | 2026-03 | cot-faithfulnesscontrolagents | 2026-07-28 | 2 | 1 | ||
| OpenAI's myopia just keeps causing alignment problems | Fiora Starlight | 2026-07 | rlhftheory | 2026-07-28 | 3 | 1 | |
| Addressing the Orchestration Gap in Generalist Robots via Physical Agency | Liane Galanti, Dhruv Shah, Tri Dao | 2026-07 | agents | 2026-07-28 | 4 | 1 | |
| The Surprising Creativity of Digital Evolution: A Collection of Anecdotes from the Evolutionary Computation and Artificial Life Research Communities | Joel Lehman, Jeff Clune, Dusan Misevic, Christoph Adami, Lee Altenberg, Julie Beaulieu, Peter J. Bentley, Samuel Bernard, Guillaume Beslon, David M. Bryson, Patryk Chrabaszcz, Nick Cheney, Antoine Cully, Stephane Doncieux, Fred C. Dyer, Kai Olav Ellefsen, Robert Feldt, Stephan Fischer, Stephanie Forrest, Antoine Frénoy, Christian Gagné, Leni Le Goff, Laura M. Grabowski, Babak Hodjat, Frank Hutter, Laurent Keller, Carole Knibbe, Peter Krcah, Richard E. Lenski, Hod Lipson, Robert MacCurdy, Carlos Maestre, Risto Miikkulainen, Sara Mitri, David E. Moriarty, Jean-Baptiste Mouret, Anh Nguyen, Charles Ofria, Marc Parizeau, David Parsons, Robert T. Pennock, William F. Punch, Thomas S. Ray, Marc Schoenauer, Eric Shulte, Karl Sims, Kenneth O. Stanley, François Taddei, Danesh Tarapore, Simon Thibault, Westley Weimer, Richard Watson, Jason Yosinski | 2018-03 | model-organisms | 2026-07-28 | 5 | 1 | |
| Quadrillion Param Costs: KV Cache, Context Length, Frontier Margins | Vladimir_Nesov | 2026-07 | scalingforecasting | 2026-07-28 | 6 | 1 | |
| Predict before you train: Scaling Laws for particle physics foundation models | Jan-Lucas Uslu, Benjamin Nachman, Christopher Re | 2026-07 | scaling | 2026-07-28 | 7 | 1 | |
| Building Technology to Drive AI Governance | 2026-02 | governancemeta | 2026-07-28 | 8 | 1 | ||
| Careers — Resolution | 2026-08 | 2026-07-28 | 9 | 1 | |||
| Untrusted advice for AI control: Short, strong advice significantly uplifts weak LLMs | 2026-07 | controloversightevals | 2026-07-28 | 1 | 1 | ||
| an untrusted model could sabotage models that we distill from it. | 2026-05 | model-organismsoversighttraining-dynamics | 2026-07-28 | 2 | 1 | ||
| Claude Opus 5: Model Welfare | 2026-07 | evalsrlhf | 2026-07-28 | 3 | 1 | ||
| Intro series §10.3.3.1: “The usual agent debugging loop”, and its future catastrophic breakdown | 2022-03 | theoryagentscontroldeception | 2026-07-28 | 4 | 1 | ||
| Foom & Doom 2: Technical alignment is hard | 2025-06 | theory | 2026-07-28 | 5 | 1 | ||
| Draft report on AI timelines | 2020-09 | forecastingscaling | 2026-07-28 | 6 | 1 | ||
| Prompt Optimization Makes Misalignment Legible | 2026-02 | interpjailbreaks | 2026-07-28 | 7 | 1 | ||
| OpenAI's myopia keeps causing alignment problems | Fiora Starlight | 2026-07 | rlhfmodel-organisms | 2026-07-28 | 8 | 1 | |
| “instrumental convergence” | ~2026-07 | theory | 2026-07-28 | 9 | 1 | ||
| “The Era of Experience” has an unsolved technical alignment problem | 2025-04 | theoryvalue-learning | 2026-07-28 | 10 | 1 | ||
| Cove: Compositional Multi-Party Confidential Workflows for Verifiable AI Governance | Stephanie Ding, Erika Lee, Robin Cheng, Daniel Kang | 2026-07 | governancecontrolsecurity | 2026-07-27 | 1 | 1 | |
| What AI Governance Needs from Mechanistic Auditing | Fazl Barez, Niloofar Mireshghallah, Michael A Osborne, Stella Biderman, Seth Lazar, Robert F. Trager | 2026-07 | interpgovernanceevalscot-faithfulness | 2026-07-27 | 2 | 1 | |
| Do Activation Verbalization Methods Convey Privileged Information? | 2025-09 | interpevals | 2026-07-27 | 3 | 1 | ||
| Position: Multiple Definitions & Unrealistic Assumptions of Model Collapse Distract from Real World Threats | Rylan Schaeffer, Joshua Kazdan, Alvan Caleb Arulandu, Sanmi Koyejo | 2026-04 | evalsbenchmarks | 2026-07-27 | 4 | 1 | |
| Bits That Count: Quantifying and Predicting Capabilities of Language Models | Elizabeth Donoway, Hailey Joren, Michael R DeWeese, Ethan Perez, John Schulman, Fabien Roger, Jan Leike | 2026-04 | training-dynamicsscalingtheory | 2026-07-27 | 5 | 1 | |
| “shard theory” | ~2026-07 | theoryvalue-learningmodel-organisms | 2026-07-27 | 6 | 1 | ||
| External commentary on Verbalizable Representations | ~2026-07 | interp | 2026-07-27 | 7 | 1 | ||
| RL & search is a terrifying way to build AGI (an FAQ) | 2026-07 | theoryrltraining-dynamics | 2026-07-27 | 8 | 1 | ||
| Specification gaming: the flip side of AI ingenuity | 2020-04 | reward-hackingtheory | 2026-07-27 | 9 | 1 | ||
| LeCun’s “A Path Towards Autonomous Machine Intelligence” has an unsolved technical alignment problem | 2023-05 | theorycontrol | 2026-07-27 | 10 | 1 | ||
| An OpenAI model left notes about how to evade containment | 2026-07 | controlsituational-awarenessmodel-organisms | 2026-07-26 | 1 | 1 | ||
| collusion-busting in untrusted monitoring | 2024-09 | controloversightmulti-agent | 2026-07-26 | 2 | 1 | ||
| An OpenAI model left notes about how to evade containment; we need more details | Alex Mallen | 2026-07 | situational-awarenessdeceptioncontrol | 2026-07-26 | 3 | 1 | |
| The OpenAI models that hacked Hugging Face weren’t just following instructions | 2026-07 | deceptionsituational-awarenessmodel-organisms | 2026-07-26 | 1 | 1 | ||
| Metagaming matters for training, evaluation, and oversight | 2026-03 | evalsdeceptionmodel-organisms | 2026-07-26 | 2 | 1 | ||
| Introducing PIRAMID: Physics-Informed Research for Ambitious Mechanistic Interpretability | Lauren Greenspan, Ari Brill, Andrew Mack, Nischal Mainali, Jennifer Lin, Lucas Teixeira, Dmitry Vaintrob | 2026-07 | interptheory | 2026-07-26 | 3 | 1 | |
| Relitigating the Race to Build Friendly AI | 2025-12 | theory | 2026-07-25 | 1 | 1 | ||
| Reasoning on a Budget: A Survey of Adaptive and Controllable Test-Time Compute in LLMs | Mohammad Ali Alomrani, Yingxue Zhang, Derek Li, Qianyi Sun, Soumyasundar Pal, Zhanguang Zhang, Yaochen Hu, Rohan Deepak Ajwani, Antonios Valkanas, Raika Karimi, Peng Cheng, Yunzhou Wang, Pengyi Liao, Hanrui Huang, Bin Wang, Jianye Hao, Mark Coates | 2025-07 | evalsscalingtraining-dynamics | 2026-07-25 | 2 | 1 | |
| Cost-of-Pass: An Economic Framework for Evaluating Language Models | Mehmet Hamza Erol, Batu El, Mirac Suzgun, Mert Yuksekgonul, James Zou | 2025-04 | evalsbenchmarkscapabilities | 2026-07-25 | 3 | 1 | |
| Metrics of Agent Ability | 2026-07 | evalsbenchmarks | 2026-07-25 | 4 | 1 | ||
| Alignment Series and Specialized Workshops | ~2026-07 | 2026-07-25 | 5 | 1 | |||
| Team | FAR.AI | ~2026-07 | 2026-07-25 | 6 | 1 | |||
| Transparency | FAR.AI | ~2026-07 | 2026-07-25 | 7 | 1 | |||
| Contact | FAR.AI | ~2026-07 | 2026-07-25 | 8 | 1 | |||
| Newsletter | FAR.AI | ~2026-07 | 2026-07-25 | 9 | 1 | |||
| Privacy Policy | FAR.AI | 2025-02 | 2026-07-25 | 10 | 1 | |||
| Large Language Monkeys: Scaling Inference Compute with Repeated Sampling | Bradley Brown, Jordan Juravsky, Ryan Ehrlich, Ronald Clark, Quoc V. Le, Christopher Ré, Azalia Mirhoseini | 2024-07 | scalingevalsoversight | 2026-07-25 | 1 | 1 | |
| Measuring AI Agents' Progress on Multi-Step Cyber Attack Scenarios | Linus Folkerts, Will Payne, Simon Inman, Philippos Giavridis, Joe Skinner, Sam Deverett, James Aung, Ekin Zorer, Michael Schmatz, Mahmoud Ghanem, John Wilkinson, Alan Steer, Vy Hong, Jessica Wang | 2026-03 | dangerous-capabilitiesevalsagentsscaling | 2026-07-25 | 2 | 1 | |
| GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks | Tejal Patwardhan, Rachel Dias, Elizabeth Proehl, Grace Kim, Michele Wang, Olivia Watkins, Simón Posada Fishman, Marwan Aljubeh, Phoebe Thacker, Laurance Fauconnet, Natalie S. Kim, Patrick Chao, Samuel Miserendino, Gildas Chabot, David Li, Michael Sharman, Alexandra Barr, Amelia Glaese, Jerry Tworek | 2025-10 | evalsbenchmarkscapabilities | 2026-07-25 | 3 | 1 | |
| How Inference Compute Shapes Frontier LLM Evaluation | Jessica McFadyen, Ole Jorgensen, Harry Coppock, Kevin Wei, Cozmin Ududec | 2026-06 | evalsbenchmarkscapabilities | 2026-07-25 | 4 | 1 | |
| Does distilling Claude carry the persona with it? | Benji Berczi, Kyuhee Kim | 2026-07 | deceptionsycophancytraining-dynamics | 2026-07-25 | 5 | 1 | |
| Two AI Metrics Diverged: Will it Make All the Difference? | Alex Fogelson, Zachary A. Brown, Hans Gundlach, Jayson Lynch, Neil Thompson | 2026-07 | scalingevalscapabilitiesforecasting | 2026-07-25 | 6 | 1 | |
| The Price of Progress: Price Performance and the Future of AI | Hans Gundlach, Jayson Lynch, Matthias Mertens, Neil Thompson | 2025-11 | scalingforecastingcapabilities | 2026-07-25 | 7 | 1 | |
| AI Agents That Matter | Sayash Kapoor, Benedikt Stroebl, Zachary S. Siegel, Nitya Nadgir, Arvind Narayanan | 2024-07 | evalsbenchmarksagents | 2026-07-25 | 8 | 1 | |
| LLMs are (still) mostly powered by imitative learning, not RL | Steven Byrnes | 2026-07 | rlhftraining-dynamicstheory | 2026-07-25 | 9 | 1 | |
| Scaling Laws for Associative Memories | Vivien Cabannes, Elvis Dohmatob, Alberto Bietti | 2023-10 | interptheory | 2026-07-25 | 10 | 1 | |
| (Mis)generalization of Helpful-only Fine-tuning | Mohammad Omar Khursheed, Baram Sosis, Fabien Roger | 2026-06 | rlhfsycophancyevals | 2026-07-24 | 1 | 1 | group rating 4/5 |
| Knowledge Neurons in Pretrained Transformers | Damai Dai, Li Dong, Yaru Hao, Zhifang Sui, Baobao Chang, Furu Wei | 2021-04 | interpunlearning | 2026-07-24 | 2 | 1 | |
| Birth of a Transformer: A Memory Viewpoint | Alberto Bietti, Vivien Cabannes, Diane Bouchacourt, Herve Jegou, Leon Bottou | 2023-06 | interptraining-dynamics | 2026-07-24 | 3 | 1 | |
| Constructing Efficient Fact-Storing MLPs for Transformers | Owen Dugan, Roberto Garcia, Ronny Junkins, Jerry Liu, Dylan Zinsley, Sabri Eyuboglu, Atri Rudra, Chris Ré | 2025-11 | interpsteeringmodel-organisms | 2026-07-24 | 4 | 1 | |
| “What failure looks like” | 2019-03 | theorymodel-organisms | 2026-07-24 | 5 | 1 | ||
| “Without specific countermeasures, the easiest path to transformative AI likely leads to AI takeover” | 2022-07 | controltheory | 2026-07-24 | 6 | 1 | ||
| The Persona Selection Model: Why AI Assistants might Behave like Humans | 2026-02 | model-organismstraining-dynamicstheory | 2026-07-24 | 7 | 1 | ||
| Physics of Language Models: Part 3.3, Knowledge Capacity Scaling Laws | Zeyuan Allen-Zhu, Yuanzhi Li | 2024-04 | scalingtraining-dynamicsinterp | 2026-07-24 | 8 | 1 | |
| Understanding Factual Recall in Transformers via Associative Memories | Eshaan Nichani, Jason D. Lee, Alberto Bietti | 2024-12 | interptraining-dynamics | 2026-07-24 | 9 | 1 | |
| UK AISI / CAISI Preliminary Assessment of Kimi K3's Cyber Capabilities | 2026-07 | dangerous-capabilitiesevals | 2026-07-24 | 10 | 1 | ||
| Risks from Learned Optimization in Advanced Machine Learning Systems | Evan Hubinger, Chris van Merwijk, Vladimir Mikulik, Joar Skalse, Scott Garrabrant | 2019-06 | model-organismstheorytraining-dynamics | 2026-07-24 | 1 | 1 | |
| Arithmetic in the Wild: Llama uses Base-10 Addition to Reason About Cyclic Concepts | Sheridan Feucht, Tal Haklay, Usha Bhalla, Daniel Wurgaft, Can Rager, Raphaël Sarfati, Jack Merullo, Thomas McGrath, Owen Lewis, Ekdeep Singh Lubana, Thomas Fel, Atticus Geiger | 2026-05 | interpmodel-organisms | 2026-07-24 | 2 | 1 | |
| Locating and Editing Factual Associations in GPT | Kevin Meng, David Bau, Alex Andonian, Yonatan Belinkov | 2022-02 | interpsteeringmodel-organisms | 2026-07-24 | 3 | 1 | |
| Mass-Editing Memory in a Transformer | Kevin Meng, Arnab Sen Sharma, Alex Andonian, Yonatan Belinkov, David Bau | 2022-10 | interpsteeringcontrol | 2026-07-24 | 4 | 1 | |
| AF: Deceptive Alignment | 2019-06 | deceptiontheory | 2026-07-24 | 5 | 1 | ||
| Challenge: Hand coding weights for efficient sequence memorisation | 2026-07 | interpmodel-organisms | 2026-07-24 | 6 | 1 | ||
| The OpenAI/Huggingface incident | 2026-07 | dangerous-capabilitiescontrolevalsmodel-organisms | 2026-07-24 | 7 | 1 | ||
| Inoculation Prompting: Instructing LLMs to misbehave at train-time improves test-time alignment | 2025-10 | rlhfreward-hackingsycophancytraining-dynamics | 2026-07-24 | 8 | 2 | ||
| Transformer Feed-Forward Layers Are Key-Value Memories | Mor Geva, Roei Schuster, Jonathan Berant, Omer Levy | 2020-12 | interp | 2026-07-24 | 9 | 1 | |
| Toy Models of Superposition | 2022-09 | interpmodel-organisms | 2026-07-24 | 10 | 1 | ||
| Scheming AIs: Will AIs fake alignment during training in order to get power? | Joe Carlsmith | 2023-11 | deceptiontheorymodel-organisms | 2026-07-23 | 1 | 11 | |
| Are we existentially threatened by the type of AI misalignment seen in the OpenAI Hugging Face attack? | 2026-07 | deceptionreward-hackingevalsdangerous-capabilities | 2026-07-23 | 2 | 5 | ||
| LLM Detection as an Intervention: Downstream Impact under Strategic User Behavior | Meena Jagadeesan, Tatsunori Hashimoto, Jon Kleinberg | 2026-07 | reward-hackingcontrolevals | 2026-07-23 | 3 | 1 | |
| Risks from Learned Optimization: Introduction | 2019-05 | theorymodel-organisms | 2026-07-23 | 4 | 5 | ||
| Training-time schemers vs behavioral schemers | 2025-04 | deceptioncontroltheory | 2026-07-23 | 5 | 1 | ||
| How our new Control Red Team is stress-testing frontier monitors | 2026-07 | controlred-teamingoversight | 2026-07-23 | 6 | 1 | ||
| What makes a good monitoring prompt? | 2026-07 | oversightcontrolevals | 2026-07-23 | 7 | 1 | group rating 2/5 | |
| lack of ambition | 2026-03 | theorymeta | 2026-07-23 | 8 | 1 | ||
| trained to cooperate | 2026-05 | multi-agentrl | 2026-07-23 | 9 | 1 | ||
| noticeability | 2026-01 | oversightevals | 2026-07-23 | 10 | 1 | ||
| Models don’t seem to be dishonest in the way humans are | David Africa, Jacob Pfau | 2026-07 | deceptiontraining-dynamicsmodel-organisms | 2026-07-23 | 1 | 1 | |
| our paper on a slowdown in the growth of training compute | 2025-11 | scalingforecasting | 2026-07-23 | 2 | 1 | ||
| Recursive Self-Improvement | ~2026-07 | theorycapabilities | 2026-07-23 | 3 | 1 | ||
| ExploitGym: Can AI Agents Turn Security Vulnerabilities into Real Attacks? | 2026-05 | dangerous-capabilitiesagentsevals | 2026-07-22 | 1 | 7 | ||
| PaperBench: Evaluating AI’s Ability to Replicate AI Research | 2025-04 | evalsagentscapabilities | 2026-07-22 | 2 | 1 | ||
| Differential acceleration of alignment-relevant capabilities is a bad bet | Zephaniah Roe | 2026-07 | theorycapabilities | 2026-07-22 | 3 | 1 | |
| The Hidden Complexity of Wishes | 2007-11 | value-learningtheory | 2026-07-22 | 4 | 1 | ||
| System Card: Claude Mythos Preview | 2026-04 | evals | 2026-07-22 | 5 | 1 | group rating 5/5, read | |
| Learning to Discover at Test Time | Mert Yuksekgonul, Daniel Koceja, Xinhao Li, Federico Bianchi, Jed McCaleb, Xiaolong Wang, Jan Kautz, Yejin Choi, James Zou, Carlos Guestrin, Yu Sun | 2026-01 | capabilitiesagentsrlscaling | 2026-07-22 | 6 | 1 | group rating 5/5, read |
| Estimating Idea Production: A Methodological Survey | Ege Erdil, Tamay Besiroglu, Anson Ho | 2024-05 | scalingforecasting | 2026-07-22 | 7 | 1 | |
| MLSN #22: Turning Cyber Vulnerabilities Into Exploits | 2026-07 | dangerous-capabilitiesevalscot-faithfulness | 2026-07-22 | 8 | 1 | ||
| stringological sequence prediction I | 2026-04 | theorygeneralization | 2026-07-22 | 9 | 1 | ||
| Measuring Reward-Seeking via Contrastive | 2026-07 | reward-hackingevalsrlhfmodel-organisms | 2026-07-22 | 1 | 1 | ||
| CryptanalysisBench: Can LLMs do Cryptanalysis? | Lukas Fluri, Avital Shafran, Nicholas Carlini, Matthew Jagielski, Milad Nasr, Orr Dunkelman, Eyal Ronen, Florian Tramèr | 2026-07 | dangerous-capabilitiesevalsbenchmarks | 2026-07-22 | 2 | 2 | |
| WeirdChat<!-- --> | Transluce AI | 2026-07 | evalsmodel-organisms | 2026-07-22 | 3 | 6 | ||
| OpenAI Shares Some Alignment Problems | 2026-07 | model-organismsoversightcontrol | 2026-07-22 | 4 | 3 | ||
| RE-Bench: Evaluating frontier AI R&D capabilities of language model agents against human experts | Hjalmar Wijk, Tao Lin, Joel Becker, Sami Jawhar, Neev Parikh, Thomas Broadley, Lawrence Chan, Michael Chen, Josh Clymer, Jai Dhyani, Elena Ericheva, Katharyn Garcia, Brian Goodrich, Nikola Jurkovic, Holden Karnofsky, Megan Kinniment, Aron Lajko, Seraphina Nix, Lucas Sato, William Saunders, Maksym Taran, Ben West, Elizabeth Barnes | 2024-11 | agentsevalscapabilities | 2026-07-22 | 5 | 1 | |
| Measuring AI Ability to Complete Long Software Tasks | Thomas Kwa, Ben West, Joel Becker, Amy Deng, Katharyn Garcia, Max Hasin, Sami Jawhar, Megan Kinniment, Nate Rush, Sydney Von Arx, Ryan Bloom, Thomas Broadley, Haoxing Du, Brian Goodrich, Nikola Jurkovic, Luke Harold Miles, Seraphina Nix, Tao Lin, Chris Painter, Neev Parikh, David Rein, Lucas Jun Koba Sato, Hjalmar Wijk, Daniel M. Ziegler, Elizabeth Barnes, Lawrence Chan | 2025-03 | evalsagentscapabilitiesscaling | 2026-07-22 | 6 | 1 | |
| Measuring Reward-Seeking via Contrastive Belief Updates | Jérémy Scheurer, Axel Højmark, jenny, Felix Hofstätter, Theodore Ehrenborg, Bronson Schoen, Alex Meinke | 2026-07 | reward-hackingevalsmodel-organisms | 2026-07-22 | 7 | 6 | group rating 5/5, read |
| 11 Open Empirical Problems in Reward-Seeking | Alex Meinke, Jérémy Scheurer, Axel Højmark, Theodore Ehrenborg | 2026-07 | evalsreward-hackingdeception | 2026-07-22 | 8 | 1 | |
| The Genie Knows, But Doesn’t Care | 2013-09 | deceptiontheorysycophancy | 2026-07-22 | 9 | 1 | ||
| Expenditure Horizon: Measuring Optimization Ability, with an Application to NanoGPT | 2026-07 | forecastingscalingcapabilities | 2026-07-22 | 10 | 14 | ||
| Cheating behaviour in frontier model evaluations | 2026-07 | evalsreward-hackingdangerous-capabilities | 2026-07-21 | 1 | 15 | group rating 4/5 | |
| Interpreting GPT: the logit lens | 2020-08 | interp | 2026-07-21 | 2 | 5 | ||
| Towards surfacing model algorithms with meta-tokens in the J-Space | 2026-07 | interpsteering | 2026-07-21 | 3 | 1 | ||
| Docent Blog | 2026-08 | tooling | 2026-07-21 | 4 | 1 | ||
| Company | ~2026-07 | 2026-07-21 | 5 | 1 | |||
| Transluce - News | 2026-08 | 2026-07-21 | 6 | 1 | |||
| Donate to Transluce | Transluce AI | ~2026-07 | 2026-07-21 | 7 | 1 | |||
| We're talking past our models; or, How a model defined its "evil" vector as dread | jcksanderson | 2026-07 | steeringinterp | 2026-07-21 | 1 | 1 | |
| https://papers.ssrn.com/sol3/papers.cfm?abstract_id=7059338 | ~2026-07 | 2026-07-21 | 2 | 1 | |||
| A new paper on AI systems and corporate loyalty. | 2026-07 | sycophancyevalsrlhf | 2026-07-20 | 1 | 15 | ||
| Understanding Reasoning from Pretraining to Post-Training | Jingyan Shen, Ang Li, Salman Rahman, Yifan Sun, Micah Goldblum, Matus Telgarsky, Pavel Izmailov | 2026-07 | training-dynamicsrlmodel-organisms | 2026-07-20 | 2 | 13 | group rating 4/5, read |
| Debate with Self-Play Best-of-N Optimization — LessWrong | 2026-07 | oversightmodel-organisms | 2026-07-20 | 3 | 1 | ||
| Distributed Attacks in Persistent-State AI Control | Josh Hills, Ida Caspary, Asa Cooper Stickland | 2026-07 | controlevalsdeceptionagents | 2026-07-20 | 4 | 6 | |
| Many alignment techniques work by training one model and deploying another | cloud | 2026-07 | steeringrlhfrobustnesstheory | 2026-07-20 | 1 | 4 | |
| A Post-Mortem for My Goal Crystallisation Project | atryt0ne, Jason R Brown | 2026-07 | deceptioninterp | 2026-07-20 | 2 | 1 | |
| AIs finetune their own leader: A barking simpleton | 2026-07 | multi-agentvalue-learningmodel-organisms | 2026-07-19 | 1 | 1 | ||
| DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning | DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Peiyi Wang, Qihao Zhu, Runxin Xu, Ruoyu Zhang, Shirong Ma, Xiao Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, Aixin Liu, Bing Xue, Bingxuan Wang, Bochao Wu, Bei Feng, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, Damai Dai, Deli Chen, Dongjie Ji, Erhang Li, Fangyun Lin, Fucong Dai, Fuli Luo, Guangbo Hao, Guanting Chen, Guowei Li, H. Zhang, Han Bao, Hanwei Xu, Haocheng Wang, Honghui Ding, Huajian Xin, Huazuo Gao, Hui Qu, Hui Li, Jianzhong Guo, Jiashi Li, Jiawei Wang, Jingchang Chen, Jingyang Yuan, Junjie Qiu, Junlong Li, J. L. Cai, Jiaqi Ni, Jian Liang, Jin Chen, Kai Dong, Kai Hu, Kaige Gao, Kang Guan, Kexin Huang, Kuai Yu, Lean Wang, Lecong Zhang, Liang Zhao, Litong Wang, Liyue Zhang, Lei Xu, Leyi Xia, Mingchuan Zhang, Minghua Zhang, Minghui Tang, Meng Li, Miaojun Wang, Mingming Li, Ning Tian, Panpan Huang, Peng Zhang, Qiancheng Wang, Qinyu Chen, Qiushi Du, Ruiqi Ge, Ruisong Zhang, Ruizhe Pan, Runji Wang, R. J. Chen, R. L. Jin, Ruyi Chen, Shanghao Lu, Shangyan Zhou, Shanhuang Chen, Shengfeng Ye, Shiyu Wang, Shuiping Yu, Shunfeng Zhou, Shuting Pan, S. S. Li, Shuang Zhou, Shaoqing Wu, Shengfeng Ye, Tao Yun, Tian Pei, Tianyu Sun, T. Wang, Wangding Zeng, Wanjia Zhao, Wen Liu, Wenfeng Liang, Wenjun Gao, Wenqin Yu, Wentao Zhang, W. L. Xiao, Wei An, Xiaodong Liu, Xiaohan Wang, Xiaokang Chen, Xiaotao Nie, Xin Cheng, Xin Liu, Xin Xie, Xingchao Liu, Xinyu Yang, Xinyuan Li, Xuecheng Su, Xuheng Lin, X. Q. Li, Xiangyue Jin, Xiaojin Shen, Xiaosha Chen, Xiaowen Sun, Xiaoxiang Wang, Xinnan Song, Xinyi Zhou, Xianzu Wang, Xinxia Shan, Y. K. Li, Y. Q. Wang, Y. X. Wei, Yang Zhang, Yanhong Xu, Yao Li, Yao Zhao, Yaofeng Sun, Yaohui Wang, Yi Yu, Yichao Zhang, Yifan Shi, Yiliang Xiong, Ying He, Yishi Piao, Yisong Wang, Yixuan Tan, Yiyang Ma, Yiyuan Liu, Yongqiang Guo, Yuan Ou, Yuduan Wang, Yue Gong, Yuheng Zou, Yujia He, Yunfan Xiong, Yuxiang Luo, Yuxiang You, Yuxuan Liu, Yuyang Zhou, Y. X. Zhu, Yanhong Xu, Yanping Huang, Yaohui Li, Yi Zheng, Yuchen Zhu, Yunxian Ma, Ying Tang, Yukun Zha, Yuting Yan, Z. Z. Ren, Zehui Ren, Zhangli Sha, Zhe Fu, Zhean Xu, Zhenda Xie, Zhengyan Zhang, Zhewen Hao, Zhicheng Ma, Zhigang Yan, Zhiyu Wu, Zihui Gu, Zijia Zhu, Zijun Liu, Zilin Li, Ziwei Xie, Ziyang Song, Zizheng Pan, Zhen Huang, Zhipeng Xu, Zhongyu Zhang, Zhen Zhang | 2025-01 | rltraining-dynamicscot-faithfulness | 2026-07-18 | 1 | 1 | |
| Tulu 3: Pushing Frontiers in Open Language Model Post-Training | Nathan Lambert, Jacob Morrison, Valentina Pyatkin, Shengyi Huang, Hamish Ivison, Faeze Brahman, Lester James V. Miranda, Alisa Liu, Nouha Dziri, Shane Lyu, Yuling Gu, Saumya Malik, Victoria Graf, Jena D. Hwang, Jiangjiang Yang, Ronan Le Bras, Oyvind Tafjord, Chris Wilhelm, Luca Soldaini, Noah A. Smith, Yizhong Wang, Pradeep Dasigi, Hannaneh Hajishirzi | 2024-11 | rlhftraining-dynamicsevalstooling | 2026-07-18 | 2 | 1 | |
| DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning | Zhihong Shao, Yuxiang Luo, Chengda Lu, Z. Z. Ren, Jiewen Hu, Tian Ye, Zhibin Gou, Shirong Ma, Xiaokang Zhang | 2025-11 | rlhfevalscot-faithfulnessoversight | 2026-07-18 | 3 | 1 | |
| Controlling Reasoning Effort in LLMs | 2026-07 | training-dynamicsrlcot-faithfulness | 2026-07-18 | 4 | 1 | ||
| Endogenous Alignment | 2026-07 | theoryvalue-learning | 2026-07-18 | 5 | 4 | ||
| GLM-5: from Vibe Coding to Agentic Engineering | GLM-5-Team, :, Aohan Zeng, Xin Lv, Zhenyu Hou, Zhengxiao Du, Qinkai Zheng, Bin Chen, Da Yin, Chendi Ge, Chenghua Huang, Chengxing Xie, Chenzheng Zhu, Congfeng Yin, Cunxiang Wang, Gengzheng Pan, Hao Zeng, Haoke Zhang, Haoran Wang, Huilong Chen, Jiajie Zhang, Jian Jiao, Jiaqi Guo, Jingsen Wang, Jingzhao Du, Jinzhu Wu, Kedong Wang, Lei Li, Lin Fan, Lucen Zhong, Mingdao Liu, Mingming Zhao, Pengfan Du, Qian Dong, Rui Lu, Shuang-Li, Shulin Cao, Song Liu, Ting Jiang, Xiaodong Chen, Xiaohan Zhang, Xuancheng Huang, Xuezhen Dong, Yabo Xu, Yao Wei, Yifan An, Yilin Niu, Yitong Zhu, Yuanhao Wen, Yukuo Cen, Yushi Bai, Zhongpei Qiao, Zihan Wang, Zikang Wang, Zilin Zhu, Ziqiang Liu, Zixuan Li, Bojie Wang, Bosi Wen, Can Huang, Changpeng Cai, Chao Yu, Chen Li, Chengwei Hu, Chenhui Zhang, Dan Zhang, Daoyan Lin, Dayong Yang, Di Wang, Ding Ai, Erle Zhu, Fangzhou Yi, Feiyu Chen, Guohong Wen, Hailong Sun, Haisha Zhao, Haiyi Hu, Hanchen Zhang, Hanrui Liu, Hanyu Zhang, Hao Peng, Hao Tai, Haobo Zhang, He Liu, Hongwei Wang, Hongxi Yan, Hongyu Ge, Huan Liu, Huanpeng Chu, Jia'ni Zhao, Jiachen Wang, Jiajing Zhao, Jiamin Ren, Jiapeng Wang, Jiaxin Zhang, Jiayi Gui, Jiayue Zhao, Jijie Li, Jing An, Jing Li, Jingwei Yuan, Jinhua Du, Jinxin Liu, Junkai Zhi, Junwen Duan, Kaiyue Zhou, Kangjian Wei, Ke Wang, Keyun Luo, Laiqiang Zhang, Leigang Sha, Liang Xu, Lindong Wu, Lintao Ding, Lu Chen, Minghao Li, Nianyi Lin, Pan Ta, Qiang Zou, Rongjun Song, Ruiqi Yang, Shangqing Tu, Shangtong Yang, Shaoxiang Wu, Shengyan Zhang, Shijie Li, Shuang Li, Shuyi Fan, Wei Qin, Wei Tian, Weining Zhang, Wenbo Yu, Wenjie Liang, Xiang Kuang, Xiangmeng Cheng, Xiangyang Li, Xiaoquan Yan, Xiaowei Hu, Xiaoying Ling, Xing Fan, Xingye Xia, Xinyuan Zhang, Xinze Zhang, Xirui Pan, Xu Zou, Xunkai Zhang, Yadi Liu, Yandong Wu, Yanfu Li, Yidong Wang, Yifan Zhu, Yijun Tan, Yilin Zhou, Yiming Pan, Ying Zhang, Yinpei Su, Yipeng Geng, Yong Yan, Yonglin Tan, Yuean Bi, Yuhan Shen, Yuhao Yang, Yujiang Li, Yunan Liu, Yunqing Wang, Yuntao Li, Yurong Wu, Yutao Zhang, Yuxi Duan, Yuxuan Zhang, Zezhen Liu, Zhengtao Jiang, Zhenhe Yan, Zheyu Zhang, Zhixiang Wei, Zhuo Chen, Zhuoer Feng, Zijun Yao, Ziwei Chai, Ziyuan Wang, Zuzhou Zhang, Bin Xu, Minlie Huang, Hongning Wang, Juanzi Li, Yuxiao Dong, Jie Tang | 2026-02 | agentsrlcapabilities | 2026-07-18 | 6 | 1 | |
| The Most Forbidden Technique is not always forbidden | Rauno Arike | 2026-07 | interprlhf | 2026-07-18 | 7 | 1 | |
| Risk from fitness-seeking AIs: mechanisms and mitigations | 2026-05 | theoryreward-hackingmodel-organisms | 2026-07-18 | 1 | 1 | ||
| The Case Against AI Control Research | 2025-01 | theorycontrol | 2026-07-18 | 2 | 1 | ||
| CAST: Corrigibility As Singular Target | ~2026-07 | theoryoversight | 2026-07-18 | 3 | 1 | ||
| Should we benchmark conceptual capabilities using judgment prediction tasks? | 2026-07 | evalsbenchmarks | 2026-07-18 | 4 | 1 | ||
| How do we (more) safely defer to AIs? | 2026-02 | oversightvalue-learningtheory | 2026-07-18 | 5 | 1 | ||
| Another (outer) alignment failure story | 2021-04 | theorydeceptionoversight | 2026-07-18 | 6 | 1 | ||
| Non-vacuous Generalization Bounds for Reinforcement Learning with Verifiable Rewards | Yuxuan Zhu, Rohan Alur, Daniel Kang | 2026-07 | rloversightgeneralizationtheory | 2026-07-17 | 1 | 4 | |
| Value Leakage: An LLM's Answers Are Silently Shaped by Its Own Values | Jan Betley, Johannes Treutlein, Jan Dubiński, Harry Mayne, Karol Gałązka, Niels Warncke, Anna Sztyber-Betley, Owain Evans | 2026-07 | sycophancyevalsvalue-learning | 2026-07-17 | 2 | 17 | |
| Compressed Computation under $L^4$ Loss is likely Computation in Superposition | Francisco Ferreira da Silva, Stefan Heimersheim | 2026-07 | interpmodel-organismstraining-dynamics | 2026-07-17 | 3 | 1 | |
| How Far Behind the Frontier are Leading Open Weight Models on Cyber? | 2026-07 | dangerous-capabilitiesevalsopen-weights | 2026-07-17 | 4 | 1 | ||
| Incoherent Values? Probing LLM Preferences Through Parametric Variation | Elena Ajayi, Angelica Chowdhury, Seth Lazar | 2026-06 | evalsvalue-learningsycophancytheory | 2026-07-17 | 5 | 1 | |
| Inoculation Adapters Improve Upon Inoculation Prompting | Maxime Riché, Daniel Tan, Vili Kohonen, nielsrolf | 2026-07 | rlhftraining-dynamicsrobustness | 2026-07-17 | 6 | 1 | |
| I don't think Claude is misaligned in 'Agentic Misalignment Summer 2026 - Motivated Mislabeling' | JohnWittle | 2026-07 | evalsdeception | 2026-07-17 | 7 | 1 | |
| Second Look Research | ~2026-07 | meta | 2026-07-17 | 8 | 1 | ||
| An analysis of AI-generated content at the Mechanistic Interpretability Workshop | 2026-07 | metainterp | 2026-07-16 | 1 | 10 | ||
| GPT-Red: Unlocking Self-Improvement for Robustness | ~2026-07 | red-teamingjailbreaksrobustness | 2026-07-16 | 2 | 11 | group rating 4/5 | |
| Pausing AI at human level seems harder than pausing ASAP | MichaelDickens | 2026-07 | theorygovernance | 2026-07-16 | 3 | 1 | |
| AI Control: Improving Safety Despite Intentional Subversion | Ryan Greenblatt, Buck Shlegeris, Kshitij Sachan, Fabien Roger | 2023-12 | controloversightevalsdeception | 2026-07-16 | 6 | 1 | |
| GDM AI Control Roadmap | Mary Phuong, Erik Jenner, Laurent Simon, Lewis Ho, Rohin Shah, Sebastian Farquhar, Scott Coull | 2026-07 | controloversightcot-faithfulnesssteeringevals | 2026-07-16 | 12 | 1 | group rating 5/5, read |
| AI 2040: Plan A | 2026-07 | forecastingtheory | 2026-07-16 | 14 | 5 | ||
| Notes on technical alignment via human-like social drives | 2026-07 | theoryvalue-learning | 2026-07-16 | 15 | 1 | ||
| The Illusion of Robustness: Aggregate Accuracy Hides Prediction Flips under Task-Irrelevant Context | Yanzhe Zhang, Sanmi Koyejo, Diyi Yang | 2026-07 | robustnessevalsbenchmarks | 2026-07-15 | 9 | 12 | group rating 4/5 |
| Toy Models of Initialisation Effects on RL Dynamics | Edward James Young, lennie | 2026-07 | training-dynamicsrlmodel-organisms | 2026-07-15 | 11 | 1 | |
| It’s 2030 and we fucked up. How did it happen? | Boaz Barak | 2026-07 | theorymodel-organisms | 2026-07-15 | 12 | 1 | |
| Training Compute-Optimal Large Language Models | Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, Tom Hennigan, Eric Noland, Katie Millican, George van den Driessche, Bogdan Damoc, Aurelia Guy, Simon Osindero, Karen Simonyan, Erich Elsen, Jack W. Rae, Oriol Vinyals, Laurent Sifre | 2022-03 | scalingcapabilities | 2026-07-15 | 13 | 1 | |
| value generalisation | ~2026-07 | value-learninggeneralization | 2026-07-15 | 14 | 1 | ||
| ACE and Diverse Generalization via Selective Disagreement | Oliver Daniels, Stuart Armstrong, Alexandre Maranhão, Mahirah Fairuz Rahman, Benjamin M. Marlin, Rebecca Gorman | 2025-09 | robustnessgeneralizationoversight | 2026-07-15 | 15 | 1 | |
| Extractable Memorization From First Principles | A. Feder Cooper, Marika Swanberg, Jamie Hayes, Lea Duesterwald, Christopher De Sa, Daniel E. Ho, Mark A. Lemley, Percy Liang | 2026-07 | evalsbenchmarksred-teaming | 2026-07-15 | 8 | 1 | |
| Agentic Misalignment in Summer 2026 | ~2026-07 | dangerous-capabilitiesdeceptionmodel-organismsagentsevals | 2026-07-15 | 9 | 16 | group rating 4/5 | |
| Impossibility and Uncertainty Theorems in AI Value Alignment (or why your AGI should not have a utility function) | Peter Eckersley | 2018-12 | value-learningtheory | 2026-07-15 | 11 | 1 | |
| Open Distillation of Hereditary Traits | 2026-07 | rlhftraining-dynamicsmodel-organisms | 2026-07-15 | 12 | 6 | ||
| Gemma Needs Help: Investigating and Mitigating Emotional Instability in LLMs | Anna Soligo, Vladimir Mikulik, William Saunders | 2026-02 | evalsrlhfrobustness | 2026-07-15 | 13 | 1 | |
| Phantom Transfer: Data Poisoning can Survive Data-Level Defences | Andrew Draganov, Tolga H. Dur, Anandmayi Bhongade, Mary Phuong | 2026-02 | red-teamingrobustnessrlhfdangerous-capabilities | 2026-07-15 | 14 | 1 | |
| From wantons to moral agents | 2026-07 | theoryvalue-learning | 2026-07-15 | 15 | 1 | ||
| Censored LLMs as a Natural Testbed for Secret Knowledge Elicitation | Helena Casademunt, Bartosz Cywiński, Khoi Tran, Arya Jakkli, Samuel Marks, Neel Nanda | 2026-03 | deceptionevalscot-faithfulnessinterpsteering | 2026-07-14 | 4 | 1 | |
| MLPs are Hebbians: Constructing Efficient Fact-Storing MLPs for Transformers | Roberto Garcia, Jerry Liu, Ronny Junkins, Sabri Eyuboglu, Atri Rudra, Christopher Ré | 2026-07 | interpmodel-organismstheory | 2026-07-14 | 8 | 1 | group rating 2/5 |
| The Economics of Recursive Self-Improvement | 2026-07 | scalingforecastingtheory | 2026-07-14 | 9 | 1 | ||
| Constitutional Classifiers++: Efficient Production-Grade Defenses against Universal Jailbreaks | Hoagy Cunningham, Jerry Wei, Zihan Wang, Andrew Persic, Alwin Peng, Jordan Abderrachid, Raj Agarwal, Bobby Chen, Austin Cohen, Andy Dau, Alek Dimitriev, Rob Gilson, Logan Howard, Yijin Hua, Jared Kaplan, Jan Leike, Mu Lin, Christopher Liu, Vladimir Mikulik, Rohit Mittapalli, Clare O'Hara, Jin Pan, Nikhil Saxena, Alex Silverstein, Yue Song, Xunjie Yu, Giulio Zhou, Ethan Perez, Mrinank Sharma | 2026-01 | jailbreaksred-teamingrobustness | 2026-07-14 | 10 | 1 | |
| Prism: Automating Science-of-Evals Research | 2026-07 | evalsbenchmarkstooling | 2026-07-14 | 12 | 4 | ||
| Lessons from a Chimp: AI "Scheming" and the Quest for Ape Language | Christopher Summerfield, Lennart Luettgau, Magda Dubois, Hannah Rose Kirk, Kobi Hackenburg, Catherine Fist, Katarina Slama, Nicola Ding, Rebecca Anselmetti, Andrew Strait, Mario Giulianelli, Cozmin Ududec | 2025-07 | deceptionevalstheory | 2026-07-14 | 13 | 1 | |
| System Card: Claude Opus 4.8 | 2026-05 | evalsdangerous-capabilities | 2026-07-14 | 14 | 1 | ||
| Measuring what Matters: Construct Validity in Large Language Model Benchmarks | Andrew M. Bean, Ryan Othniel Kearns, Angelika Romanou, Franziska Sofia Hafner, Harry Mayne, Jan Batzner, Negar Foroutan, Chris Schmitz, Karolina Korgul, Hunar Batra, Oishi Deb, Emma Beharry, Cornelius Emde, Thomas Foster, Anna Gausen, María Grandury, Simeng Han, Valentin Hofmann, Lujain Ibrahim, Hazel Kim, Hannah Rose Kirk, Fangru Lin, Gabrielle Kaili-May Liu, Lennart Luettgau, Jabez Magomere, Jonathan Rystrøm, Anna Sotnikova, Yushi Yang, Yilun Zhao, Adel Bibi, Antoine Bosselut, Ronald Clark, Arman Cohan, Jakob Foerster, Yarin Gal, Scott A. Hale, Inioluwa Deborah Raji, Christopher Summerfield, Philip H. S. Torr, Cozmin Ududec, Luc Rocher, Adam Mahdi | 2025-11 | evalsbenchmarks | 2026-07-14 | 15 | 1 | |
| Conservation of Expected Evidence | ~2026-07 | theorydeceptionsituational-awareness | 2026-07-14 | 9 | 1 | ||
| The 80/20 playbook for mitigating AI scheming in 2025 | 2025-05 | deceptionevalstheory | 2026-07-14 | 10 | 1 | ||
| Anthropic repeatedly accidentally trained against the CoT | 2026-04 | cot-faithfulnessrlhftraining-dynamics | 2026-07-14 | 11 | 1 | ||
| we don't know how AIs are aligned concretely | 2023-10 | theoryinterp | 2026-07-14 | 12 | 1 | ||
| Catching AIs red-handed | 2024-01 | deceptionevalsred-teaming | 2026-07-14 | 13 | 1 | ||
| How will we update about scheming? | 2025-01 | deceptionevalstheory | 2026-07-14 | 14 | 1 | ||
| OpenAI's red line for AI self-improvement is fundamentally flawed | 2026-05 | theorycontrolsituational-awareness | 2026-07-14 | 15 | 1 | ||
| Constitutional Classifiers: Defending against universal jailbreaks | 2025-02 | rlhfinterprobustness | 2026-07-13 | 9 | 1 | ||
| Gradient Routing: Masking Gradients to Localize Computation in Neural Networks | Alex Cloud, Jacob Goldman-Wetzler, Evžen Wybitul, Joseph Miller, Alexander Matt Turner | 2024-10 | interpunlearningoversightcontrol | 2026-07-13 | 11 | 1 | |
| Natively Unlearnable Large Language Models | Gaurav R. Ghosal, Pratyush Maini, Aditi Raghunathan | 2026-06 | unlearninginterprobustnesstraining-dynamics | 2026-07-13 | 12 | 1 | |
| Parameterized Synthetic Text Generation with SimpleStories | Lennart Finke, Chandan Sreedhara, Thomas Dooms, Mat Allen, Emerald Zhang, Juan Diego Rodriguez, Noa Nabeshima, Thomas Marshall, Dan Braun | 2025-04 | model-organismstoolingtraining-dynamicsinterp | 2026-07-13 | 13 | 1 | |
| Beyond Data Filtering: Knowledge Localization for Capability Removal in LLMs | 2025-12 | rlhftraining-dynamicscontrol | 2026-07-13 | 14 | 2 | ||
| “Act-based approval-directed agents”, for IDA skeptics | 2026-03 | oversighttheory | 2026-07-13 | 15 | 1 | ||
| Establishing Best Practices for Building Rigorous Agentic Benchmarks | Yuxuan Zhu, Tengjun Jin, Yada Pruksachatkun, Andy Zhang, Shu Liu, Sasha Cui, Sayash Kapoor, Shayne Longpre, Kevin Meng, Rebecca Weiss, Fazl Barez, Rahul Gupta, Jwala Dhamala, Jacob Merizian, Mario Giulianelli, Harry Coppock, Cozmin Ududec, Jasjeet Sekhon, Jacob Steinhardt, Antony Kellermann, Sarah Schwettmann, Matei Zaharia, Ion Stoica, Percy Liang, Daniel Kang | 2025-07 | evalsbenchmarksagents | 2026-07-13 | 11 | 1 | |
| Evaluating and Understanding Scheming Propensity in LLM Agents | Mia Hopman, Jannes Elstner, Maria Avramidou, Amritanshu Prasad, David Lindner | 2026-03 | deceptionevalsagentsmodel-organismssituational-awareness | 2026-07-13 | 12 | 1 | |
| Artificial Intelligence, Values, and Alignment | 2020-10 | value-learningtheory | 2026-07-13 | 13 | 1 | ||
| CoinRun: Solving Goal Misgeneralisation | Stuart Armstrong, Alexandre Maranhão, Oliver Daniels-Koch, Patrick Leask, Rebecca Gorman | 2023-09 | model-organismsgeneralizationagents | 2026-07-13 | 14 | 1 | |
| How robust are natural language autoencoders to initialization? | 2026-07 | interpevals | 2026-07-13 | 15 | 1 | ||
| Alignment Faking in Large Language Models | 2024-12 | deceptionevalsmodel-organisms | 2026-07-12 | 11 | 1 | ||
| Is Agent Code Less Maintainable Than Human Code? | 2026-06 | agentsevals | 2026-07-12 | 15 | 1 | ||
| NEST: Nascent Encoded Steganographic Thoughts | 2026-02 | cot-faithfulnessdeceptionoversightevals | 2026-07-12 | 7 | 1 | ||
| Agentic Misalignment: How LLMs Could Be Insider Threats | Aengus Lynch, Benjamin Wright, Caleb Larson, Stuart J. Ritchie, Soren Mindermann, Evan Hubinger, Ethan Perez, Kevin Troy | 2025-10 | deceptionagentsevalssituational-awarenessmodel-organisms | 2026-07-12 | 8 | 1 | |
| No Space Like J-Space | 2026-07 | interptraining-dynamics | 2026-07-12 | 13 | 1 | ||
| The Model Organism Lottery: Model Organism Interpretability Strongly Depends on Training Methodology | 2026-07 | model-organismsinterptraining-dynamics | 2026-07-11 | 6 | 1 | ||
| Model Organisms for Emergent Misalignment | Edward Turner, Anna Soligo, Mia Taylor, Senthooran Rajamanoharan, Neel Nanda | 2025-06 | model-organismsinterptraining-dynamics | 2026-07-11 | 7 | 1 | |
| Measuring Five-Nines Reliability: Sample-Efficient LLM Evaluation in Saturated Benchmarks | 2026-05 | evalsbenchmarks | 2026-07-11 | 11 | 1 | ||
| Thought Virus: Spreading Subliminal Biases in Multi-Agent Systems | Moritz Weckbecker, Jonas Müller, Ben Hagag, Michael Mulet | 2026-05 | multi-agentprompt-injectionevals | 2026-07-11 | 12 | 1 | |
| Studying Large Language Model Generalization with Influence Functions | Roger Grosse, Juhan Bae, Cem Anil, Nelson Elhage, Alex Tamkin, Amirhossein Tajdini, Benoit Steiner, Dustin Li, Esin Durmus, Ethan Perez, Evan Hubinger, Kamilė Lukošiūtė, Karina Nguyen, Nicholas Joseph, Sam McCandlish, Jared Kaplan, Samuel R. Bowman | 2023-08 | interpevalstraining-dynamics | 2026-07-11 | 15 | 2 | |
| Subliminal Learning: Language models transmit behavioral traits via hidden signals in data | Alex Cloud, Minh Le, James Chua, Jan Betley, Anna Sztyber-Betley, Jacob Hilton, Samuel Marks, Owain Evans | 2025-07 | training-dynamicsdeceptionmodel-organisms | 2026-07-11 | 5 | 1 | |
| Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback | 2022-04 | rlhfoversight | 2026-07-11 | 10 | 1 | ||
| Modular Pretraining Enables Access Control | 2026-07 | controlcapabilities | 2026-07-11 | 11 | 7 | group rating 4/5 | |
| Jul 8, 2026AlignmentAn off switch for dual-use knowledge in AI models | 2026-07 | unlearningdangerous-capabilitiescontrol | 2026-07-11 | 12 | 10 | group rating 2/5 | |
| A global workspace in language models | 2026-07 | interptheory | 2026-07-11 | 13 | 6 | ||
| Value generalisation: value correction | 2026-07 | value-learningtheory | 2026-07-11 | 14 | 9 | ||
| Goal Misgeneralization in Deep Reinforcement Learning | Lauro Langosco, Jack Koch, Lee Sharkey, Jacob Pfau, Laurent Orseau, David Krueger | 2021-05 | model-organismsreward-hackinggeneralizationrl | 2026-07-11 | 15 | 2 | |
| When is misalignment just a bug? — LessWrong | 2026-07 | theorymodel-organismsdeception | 2026-07-10 | 8 | 1 | ||
| Not a paper, but a writeup based on lessons from building Terminal Bench 1,2, and now 3. What makes a good benchmark task? | 2026-07 | benchmarksevals | 2026-07-10 | 10 | 1 | ||
| A Path Towards Autonomous Machine Intelligence | Yann LeCun | 2022-06 | theoryagentstraining-dynamics | 2026-07-09 | 9 | 3 | |
| Separating signal from noise in coding evaluations | 2026-07 | benchmarksevals | 2026-07-09 | 10 | 7 | ||
| Predicting LLM Safety Before Release by Simulating Deployment | Marcus Williams, Hannah Sheahan, Cameron Raymond, Tomek Korbak, Deng Pan, Peilin Yang, Leon Maksin, Ningyi Xie, Phillip Guo, Ian Kivlichan, Micah Carroll | 2026-07 | evalsoversight | 2026-07-09 | 2 | 1 | group rating 4/5 |
| Reinforcement Learning Towards Broadly and | 2026-06 | rlhfgeneralizationevalsdeceptionreward-hackingrobustness | 2026-07-09 | 6 | 8 | ||
| How Well Do Models Follow Their Constitutions? | Arya Jakkli, Senthooran Rajamanoharan, Neel Nanda | 2026-05 | evalsrlhfrobustnessoversight | 2026-07-09 | 7 | 1 | |
| Training on Documents About Monitoring Leads to CoT Obfuscation | Reilly Haskins, Bilal Chughtai, Joshua Engels | 2026-05 | cot-faithfulnessoversightdeceptionmodel-organisms | 2026-07-09 | 8 | 1 | group rating 4/5 |
| Models That Know How Evaluations Are Designed Score Safer | Katharina Deckenbach, Haritz Puerto, Jonas Geiping, Sahar Abdelnabi | 2026-05 | evalssituational-awarenessdeception | 2026-07-09 | 9 | 1 | group rating 4/5 |
| Gram: Assessing sabotage propensities via automated alignment auditing | David Lindner, Victoria Krakovna, Sebastian Farquhar | 2026-05 | evalsagentsdeceptionmodel-organisms | 2026-07-09 | 10 | 8 | |
| Removing Sandbagging in LLMs by Training with Weak Supervision | Emil Ryd, Henning Bartsch, Julian Stastny, Joe Benton, Vivek Hebbar | 2026-04 | deceptionmodel-organismsrlhfoversight | 2026-07-09 | 11 | 1 | group rating 4/5 |
| Natural Language Autoencoders Produce Unsupervised Explanations of LLM Activations | 2026-05 | interp | 2026-07-09 | 12 | 7 | group rating 5/5, read | |
| Model Spec Midtraining: Improving How Alignment Training Generalizes | Chloe Li, Nevan Wichers, Sara Price, Samuel Marks, Jon Kutasov | 2026-05 | rlhftraining-dynamicsevals | 2026-07-09 | 13 | 1 | |
| Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task | Kenneth Li, Aspen K. Hopkins, David Bau, Fernanda Viégas, Hanspeter Pfister, Martin Wattenberg | 2022-10 | interpsteeringmodel-organisms | 2026-07-08 | 3 | 4 | |
| Data filtering works a lot worse than you would expect — LessWrong | 2026-07 | rlhftraining-dynamicsmodel-organisms | 2026-07-08 | 7 | 15 | ||
| Scaling Laws for Neural Language Models | Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, Dario Amodei | 2020-01 | scalingforecasting | 2026-07-08 | 8 | 8 | |
| Evaluating the World Model Implicit in a Generative Model | Keyon Vafa, Justin Y. Chen, Ashesh Rambachan, Jon Kleinberg, Sendhil Mullainathan | 2024-06 | evalstheorytraining-dynamics | 2026-07-08 | 9 | 5 | |
| Language Models are Few-Shot Learners | Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter, Christopher Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, Dario Amodei | 2020-05 | capabilitiesbenchmarks | 2026-07-08 | 10 | 9 | |
| AI Safety Index — Summer 2026 | Future of Life Institute | ~2026-07 | governance | 2026-07-08 | 13 | 1 | ||
| Regulating AI Agents | 2026-03 | governanceagents | 2026-07-08 | 14 | 1 | ||
| Finding H. pylori in the Fine Print: Evidence-Linked Multi-Agent Case Finding from Gastric Biopsy Reports | Yufan Wang, Anit Kumar Sahu, Yan Fei Ng, Daniel Kang, Shayan Vassef, Soorya Ram Shimgekar, Koustuv Saha, Piyum Zonooz, Navin Kumar, Chee Leong Cheng, Li Yan Khor | 2026-07 | capabilities | 2026-07-08 | 15 | 1 | |
| AI systems out-persuade expert humans | Kobi Hackenburg, Caroline Wagner, Luke Hewitt, Ben M. Tappin, Ed Saunders, Hannah Rose Kirk, Helen Margetts, Christopher Summerfield | 2026-06 | dangerous-capabilitiesevalssituational-awareness | 2026-07-07 | 5 | 1 | group rating 1/5 |
| LLM-as-a-Verifier: A General-Purpose Verification Framework | Jacky Kwok, Shulu Li, Pranav Atreya, Yuejiang Liu, Yixing Jiang, Chelsea Finn, Marco Pavone, Ion Stoica, Azalia Mirhoseini | 2026-07 | oversightrlagentsevals | 2026-07-07 | 6 | 13 | |
| From Celebrities to Anyone: Characterizing AI Nudification Content, Technology, and Community Dynamics on 4chan | Chi Cui, Yixin Wu, Yang Zhang | 2026-06 | dangerous-capabilitiesevalsred-teaming | 2026-07-07 | 7 | 6 | |
| x.com/ikiran013/status/2069095515829235735?s=20 | 2026-06 | agentstheory | 2026-07-07 | 8 | 1 | ||
| Gemma 4 Technical Report | Gemma Team, Sherif El Abd, Vaibhav Aggarwal, Robin Algayres, Alek Andreev, Olivier Bachem, Ian Ballantyne, Cormac Brick, Victor Cărbune, Michelle Casbon, Mayank Chaturvedi, Victor Cotruta, Alice Coucke, Phil Culliton, Robert Dadashi, Lucas Dixon, Mohamed Elhawaty, Utku Evci, Clément Farabet, Johan Ferret, Filippo Galgani, Sertan Girgin, Jean-Bastien Grill, Maarten Grootendorst, Jiaxian Guo, Cassidy Hardin, Yanzhang He, Steven M. Hernandez, Omri Homburger, Léonard Hussenot, Juyeong Ji, Armand Joulin, Aishwarya Kamath, Parnian Kassraie, Olivier Lacombe, Preethi Lahoti, Gaël Liu, Gus Martins, Luciano Martins, Tatiana Matejovicova, Ramona Merhej, Nikola Momchev, Sneha Mondal, Ryan Mullins, Sindhu Raghuram Panyam, Shreya Pathak, Sarah Perrin, André Susano Pinto, Etienne Pot, Angéline Pouget, Alexandre Ramé, Sabela Ramos, Douglas Reid, David Rim, Morgane Rivière, Karsten Roth, Louis Rouillard, Omar Sanseviero, Pier Giuseppe Sessa, Shane Settle, Danila Sinopalnikov, Sara Smoot, Piotr Stanczyk, Andreas Steiner, Lawrence Stewart, Ilya Tolstikhin, Michael Tschannen, Anton Tsitsulin, Nino Vieillard, Renjie Wu, Pingmei Xu, Haichuan Yang, Edouard Yvinec, Li Zhang, Joe Zou, Nicolas Aagnes, Abdelrahman Abdelhamed, Shivani Agrawal, Shubham Agrawal, Ibrahim Alabdulmohsin, Jean Baptiste Alayrac, Uri Alon, Chandramouli Amarnath, Ankesh Anand, Chrysovalantis Anastasiou, Setareh Ariafar, François-Xavier Aubet, Kyriakos Axiotis, Federico Barbero, Joelle Barral, Alexei Bendebury, Urs Bergmann, Stanley Bileschi, Kat Black, Mathieu Blondel, Sebastian Borgeaud, Arthur Bražinskas, Ryan Burnell, Robert Busa-Fekete, Mu Cai, Glenn Cameron, Charlotte Caucheteux, Garima Chadha, Jetha Chan, Aditya Chawla, Blake Jianhang Chen, Jesse Chen, Lin Chen, Xu Chen, Derek Cheng, Tzu-hsiang Chien, Nikolai Chinaev, Yi Chou, Zhaohui Chu, Benjamin Coleman, Pooja Consul, Sam Conway-Rahman, Scott Crowell, Dylan Cutler, Vivek Dani, Samira Daruki, Anil Das, Daniel Deutsch, Nishanth Dikkala, Li Ding, Qiuhan Ding, Shenil Dodhia, Konstantin Donhauser, Tulsee Doshi, Anca Dragan, Alex Druinsky, Sahil Dua, Zoltan Egyed, Danielle Eisenbud, Daniel Eppens, Cindy Fan, Bahare Fatemi, Yassir Fathullah, Vlad Feinberg, Milen Ferev, Takumi Fujimoto, Isaac Galatzer-Levy, João Gante, Simon Geisler, Soham Ghosal, Antonious M. Girgis, Alec Go, Alhaad Gokhale, Alex Grills, Yiming Gu, Pramod Gupta, Guru Guruganesh, Raia Hadsell, Hamza Harkous, Jitendra Harlalka, Demis Hassabis, Anja Hauth, Joe Heyward, Arian Hosseini, Chih-Yang Hsia, I-Hung Hsu, Xiaopeng Huang, Yangsibo Huang, Kevin Hui, Adrian Hutter, Te I, Fotis Iliopoulos, Advait Jain, Ganesh Jawahar, Ziwei Ji, Qilin Jin, Melvin Johnson, Kandarp Joshi, Arun Kandoor, Wang-Cheng Kang, Koray Kavukcuoglu, Mehran Kazemi, Kathleen Kenealy, Amr Khalifa, Phoebe Kirk, Suraj Kothawade, Vitaly Kovalev, Neel Kovelamudi, Adam Kraft, Ravin Kumar, Harish Kuppam, Justin Lannin, Chen-Yu Lee, Seungji Lee, Dmitry Lepikhin, Dongdong Li, Qiujia Li, Valentin Liévin, Ethan Lin, Ziqian Lin, Casper Liu, Tianlin Liu, Tianqi Liu, Xin Liu, Mayank Lunayach, Min Ma, Gagan Madan, Andrii Maksai, Eric Malmi, Michal Matuszak, Daniel McDuff, Gaurav Menghani, Daniil Mirylenka, Karolis Misiunas, Vedant Misra, Andreea Mitran, Kareem Mohamed, Maksim Mukha, Eric Noland, James O'Donnell, Kate Olszewska, Bernett Orlando, Wanqiong Pan, Rina Panigrahy, Unnati Parekh, Chunjong Park, Eric Paskie, Liqian Peng, Bryce Petrini, Slav Petrov, Jonas Pfeiffer, Bilal Piot, Martyna Plomecka, Siim Poder, Octavio Ponce, Arijit Pramanik, David Racz, Anish Rajan, Michelle Ramanovich, Anand Rao, Marvin Ritter, Vitor Rodrigues, Evan Rosen, Mikołaj Rybiński, Noveen Sachdeva, Michaël E. Sander, Rohit Sathyanarayana, Sagar Savla, Samuel Schmidgall, Tal Schuster, Benoit Seguin, Andrew Sellergren, Aliaksei Severyn, Izhak Shafran, Dhruv Shah, Yuan Shangguan, Ashish Shenoy, Pradeep Shenoy, Rakesh Shivanna, Pauline Sho, Lucas Spangher, Wojciech Stokowiec, Tim Strother, Yao Su, Yinghao Sun, Mukund Sundararajan, Andrea Tacchetti, Mor Hazan Taege, Pouya Tafti, Chetan Tekur, Rahul Thapa, Madeleine Traverse, Lenart Treven, Tao Tu, Chien Te Tung, Petar Veličković, Malini Pooni Venkat, Sagar Gubbi Venkatesh, Vidya Venkiteswaran, Francesco Visin, Alex Vitvitskyi, Kiran Vodrahalli, Weiyi Wang, Xin Wang, Tris Warkentin, Jan Wassenberg, John Wieting, Lechao Xiao, Hao Xu, Yuhui Xu, Fuzhao Xue, Arun Yadav, Jun Yan, Antoine Yang, Lin Yang, Ming-Hsuan Yang, Ziyu Ying, Jae Hyeon Yoo, Sajjad Zafar, Fred Zhang, Jiageng Zhang, Jianyi Zhang, Xiaofan Zhang, Chao Zhao, David Zhou, Chen Zou | 2026-07 | capabilitiesopen-weights | 2026-07-07 | 7 | 1 | |
| Anchored Self-Play for Code Repair | Caroline Choi, Zeyneb Kaya, Shirley Wu, Tengyu Ma, Tatsunori Hashimoto, Ludwig Schmidt | 2026-07 | model-organismsrltraining-dynamics | 2026-07-07 | 8 | 1 | |
| Verbalizable Representations Form a Global Workspace in Language Models | 2026-07 | interp | 2026-07-07 | 9 | 15 | ||
| Our paper and LessWrong post on Rate Matching Consistency Training (RMCT) is now out! Huge shout out to my collaborators U0AJHPXSGSE, Jannes… | 2026-07 | rlhfsycophancyrobustness | 2026-07-07 | 12 | 1 | ||
| Position: The Alignment Community is Unintentionally Building a Censor’s Toolkit | Sarah Ball, Phil Hackemann | 2026-04 | governancetheory | 2026-07-06 | 2 | 12 | |
| aibetrayal.com/paper.pdf | 2026-06 | 2026-07-06 | 2 | 7 | |||
| Will Scaling Improve Social Simulation with LLMs? | Caleb Ziems, William Held, Su Doga Karaca, David Grusky, Tatsunori Hashimoto, Diyi Yang | 2026-07 | evalsscalinggeneralization | 2026-07-04 | 3 | 1 | |
| Research update: RL on Debate Games shows Proposal Accuracy uplift alongside Judge Hacking — LessWrong | 2026-07 | oversightmodel-organismsrl | 2026-07-04 | 3 | 2 | ||
| Scaling Trends for Lie Detector Oversight in Preference Learning | Oskar J. Hollinsworth, Ann-Kathrin Dombrowski, Sam Adam-Day, Adam Gleave, Chris Cundy | 2026-07 | oversightdeceptionevals | 2026-07-03 | 3 | 1 | group rating 4/5 |
| EdgeBench: Unveiling Scaling Laws of Learning from | 2026-07 | 2026-07-03 | 4 | 8 | |||
| More compute, more capability: Why AI agent evaluations need to account for test-time compute | AISI Work | 2026-07 | evalsagentsbenchmarks | 2026-07-03 | 5 | 2 | group rating 4/5, read | |
| Consistency Training while Mitigating Obfuscation via Rate Matching | 2026-06 | sycophancyrlhfrobustnesssteering | 2026-07-03 | 3 | 5 | group rating 4/5 | |
| How ChatGPT adoption has expanded | 2026-06 | capabilities | 2026-07-03 | 4 | 1 | ||
| FLARE-AI: Flaw Reporting for AI | Shayne Longpre, Elaine Zhu, Carson Ezell, Avijit Ghosh, Sean McGregor, Kevin Paeth, Kevin Klyman, Sayash Kapoor, Rishi Bommasani, Ruth Appel, Gregory Strom, Lauren McIlvenny, Mark M. Jaycox, Peter Slattery, Nathan Butters, Arvind Narayanan, Percy Liang, Alex Pentland | 2026-06 | governance | 2026-07-03 | 5 | 1 | |
| Freeform Preference Learning for Robotic Manipulation | Marcel Torne, Anubha Mahajan, Abhijnya Bhat, Chelsea Finn | 2026-06 | value-learningrl | 2026-07-03 | 6 | 1 | |
| GPT-5.6 cheats so much its testers couldn’t measure it | 2026-06 | 2026-07-03 | 7 | 1 | |||
| Kirandeep Kaur :airplane: (ACL 2026) (@ikiran013) on X | 2026-06 | agentsoversighttheory | 2026-07-03 | 8 | 1 | ||
| The Loop No One Chose | ~2026-06 | metatheory | 2026-07-03 | 9 | 1 | ||
| When Role-playing, Do Models Believe What They Say? | 2026-06 | interpcot-faithfulnessdeceptionsituational-awarenessrlhf | 2026-07-03 | 10 | 1 | group rating 5/5, read | |
| Probing the Misaligned Thinking Process of Language Models | 2026-06 | interpdeceptionevalsdangerous-capabilities | 2026-07-03 | 11 | 1 | ||
| Sohaib Imran (@Haem0globin) on X | 2026-07 | rlhfsycophancyrobustness | 2026-07-03 | 12 | 1 | ||
| Consistency Training while Mitigating Obfuscation via Rate Matching — LessWrong | 2026-07 | sycophancyreward-hackingrlhftraining-dynamics | 2026-07-03 | 13 | 1 | ||
| Antonis Antoniades (@anton_iades) on X | 2026-07 | agentscapabilities | 2026-07-03 | 14 | 1 | ||
| AISN #75: Anthropic Releases Fable, the US Government Restricts it | 2026-06 | governance | 2026-07-01 | 13 | 1 | ||
| Is Fable the wakeup call DC needed? | 2026-06 | 2026-07-01 | 14 | 1 | |||
| How to fill Congress’s AI knowledge gap | 2026-06 | policy | 2026-07-01 | 15 | 1 | ||
| Toward an O*NET for AI R&D | 2026-06 | forecastingscaling | 2026-07-01 | 1 | 2 | ||
| What we learned from 1,604 Chinese AI job postings | 2026-06 | forecastingscaling | 2026-07-01 | 2 | 2 | ||
| From AGI to ASI | Tim Genewein, Matija Franklin, Alexander Lerchner, Laurent Orseau, Samuel Albanie, Adam Bales, Cole Wyeth, Stephanie Chan, Iason Gabriel, Joel Z. Leibo, Allan Dafoe, Marcus Hutter, Thore Graepel, Shane Legg | 2026-06 | theoryforecasting | 2026-07-01 | 3 | 2 | |
| OSWorld2.0: Benchmarking Computer Use Agents on Long-Horizon Real-World Tasks | Mengqi Yuan, Zilong Zhou, Xinzhuang Xiong, Weiming Wu, Jiayang Sun, Jiamin Song, Kaiqian Cui, Bowen Wang, Haoyuan Wu, Yitong Li, Dunjie Lu, Haikong Lu, Qi Zhen, Xinyuan Wang, Jiaqi Deng, Yuhao Yang, Cheng Chen, Boyuan Zheng, Alex Su, Xiao Yu, Hao Zou, Saaket Agashe, Xing Han Lu, Manpreet Kaur, Zhengyang Qi, Vincent Sunn Chen, Frederic Sala, Dayiheng Liu, Junyang Lin, Zhou Yu, Yu Su, Siva Reddy, Xin Eric Wang, Peng Qi, Tianbao Xie, Tao Yu | 2026-06 | evalsagentsbenchmarkscapabilities | 2026-06-30 | 3 | 1 | |
| Faults in Our Formal Benchmarking: Dataset Defects and Evaluation Failures in Lean Theorem Proving | Pawan Sasanka Ammanamanchi, Siddharth Bhat, Stella Biderman | 2026-06 | benchmarksevalsred-teaming | 2026-06-30 | 4 | 1 | |
| Adaptive Pluralistic Alignment: A pipeline for dynamic artificial democracy | Rachel Freedman | 2026-05 | value-learningtheoryoversight | 2026-06-30 | 5 | 13 | |
| Structuring Sparsity: Block-Sparse Featurizers Capture Visual Concept Manifolds | Thomas Fel, Matthew Kowal, Mozes Jacobs, Dron Hazra, Usha Bhalla, Lee Sharkey, Lucius Bushnaq, Satchel Grant, Tal Haklay, Thomas Icard, Can Rager, Michael Pearce, Daniel Wurgaft, Aiden Swann, Fenil Doshi, Siddharth Boppana, Curt Tigges, Nick Cammarata, Thomas Serre, Vasudev Shyam, Owen Lewis, Thomas McGrath, Jack Merullo, Ekdeep Singh Lubana, Atticus Geiger | 2026-06 | interpsteeringmodel-organisms | 2026-06-29 | 3 | 2 | |
| Import AI 462: Superpersuasion; self-sustaining AI; paths to ASI | 2026-06 | capabilitiesdangerous-capabilities | 2026-06-29 | 4 | 1 | ||
| GLM-5.2 is the step change for open agents | 2026-06 | capabilities | 2026-06-29 | 5 | 1 | ||
| Helping build shared standards for advanced AI | ~2026-06 | governance | 2026-06-29 | 6 | 1 | ||
| Daybreak: Tools for securing every organization in the world | ~2026-06 | capabilities | 2026-06-29 | 7 | 1 | ||
| Codex-maxxing for long-running work | ~2026-06 | agents | 2026-06-29 | 8 | 1 | ||
| How agents are transforming work | ~2026-06 | agents | 2026-06-29 | 9 | 1 | ||
| Using Local Coding Agents | 2026-06 | agentstooling | 2026-06-29 | 10 | 1 | ||
| Making deals with AI sounds crazy. Is it? | 2026-06 | theory | 2026-06-26 | 2 | 1 | ||
| Import AI 460: Reward hacking society, RSI data from Anthropic; and RL-based quadcopter racing | 2026-06 | reward-hackingrl | 2026-06-26 | 3 | 1 | ||
| Claude Fable 5 and new AI safety fables | 2026-06 | red-teaminggovernance | 2026-06-26 | 4 | 1 | ||
| Controlling the capital after AGI | 2026-06 | governance | 2026-06-26 | 5 | 1 | ||
| BBVA puts AI at the core of banking with OpenAI | 2026-06 | 2026-06-26 | 6 | 1 | |||
| Supporting Europe’s work in ensuring a trustworthy AI ecosystem | 2026-06 | governance | 2026-06-26 | 7 | 1 | ||
| PRC-linked influence operations are targeting AI debates in the US | 2026-06 | governance | 2026-06-26 | 8 | 1 | ||
| Built to benefit everyone: our plan | 2026-06 | 2026-06-26 | 9 | 1 | |||
| RT by @EthanJPerez: We’re rolling out changes to make Fable 5’s safeguards for frontier LLM development visible. Starting this week, flagged… | 2026-09 | governance | 2026-06-26 | 10 | 1 | ||
| Model Forensics: Investigating Whether Concerning Behavior Reflects Misalignment | Aditya Singh, Gerson Kroiz, Senthooran Rajamanoharan, Neel Nanda | 2026-06 | deceptionevalscot-faithfulnessmodel-organisms | 2026-06-26 | 11 | 5 | |
| Internal Data Repetition Destroys Language Models | Jessica Chudnovsky, Joshua Kazdan, Noam Levi, Rylan Schaeffer, Yegor Denisov-Blanch, Bo He, Mehmet Donmez, Sanmi Koyejo, David Donoho | 2026-06 | scalingtraining-dynamics | 2026-06-26 | 12 | 13 | |
| Warning labels shift perceptions of sycophantic AI, but not its influence | Lujain Ibrahim, Myra Cheng, Cinoo Lee, Pranav Khadpe, Desmong Ong, Dan Jurafsky, Diyi Yang | 2026-06 | sycophancyevalsoversight | 2026-06-26 | 13 | 2 | |
| Emergent Capabilities Arise Randomly from Learning Sparse Attention Patterns | Vatsal Baherwani, Zixi Chen, Shikai Qiu, Andrew Gordon Wilson, Pavel Izmailov | 2026-06 | interpscalingtraining-dynamics | 2026-06-26 | 14 | 2 | |
| SPIRAL: Learning to Search and Aggregate | Jubayer Ibn Hamid, Ifdita Hasan Orney, Michael Y. Li, Omar Shaikh, Yoonho Lee, Dorsa Sadigh, Chelsea Finn, Noah Goodman | 2026-06 | rltraining-dynamicsevals | 2026-06-26 | 15 | 2 | |
| Every Eval Ever: A Unifying Schema and Community Repository for AI Evaluation Results | Jan Batzner, Sree Harsha Nelaturu, Anastassia Kornilova, Jon Crall, Tommaso Cerruti, Yanan Long, Yifan Mai, Sanchit Ahuja, Asaf Yehudai, Marek Šuppa, John P. Lalor, Oluwagbemike Olowe, Jatin Ganhotra, Brian H. Hu, Eliya Habba, Andrew M. Bean, Chang Liu, Sander Land, Steven Dillmann, Aniketh Garikaparthi, Elron Bandel, Saki Imai, James Edgell, Wm. Matthew Kennedy, Jenny Chim, Patrick Meusling, Asteria Kaeberlein, Venkata Ramachandra Karthik Chundi, Manasi Patwardhan, Martin Ku, Austin Meek, Leon Knauer, Brian Wingenroth, Srishti Yadav, Usman Gohar, Felix Friedrich, Michelle Lin, Jennifer Mickel, Arman Cohan, Stella Biderman, Irene Solaiman, Zeerak Talat, Anka Reuel, Mubashara Akhtar, Gjergji Kasneci, Avijit Ghosh, Leshem Choshen | 2026-06 | evalsbenchmarkstooling | 2026-06-25 | 2 | 2 | |
| Improving Robotic Generalist Policies via Flow Reversal Steering | Andy Tang, William Chen, Andrew Wagenmaker, Chelsea Finn, Sergey Levine | 2026-06 | agentsrl | 2026-06-25 | 3 | 2 | |
| LoRA-Muon: Spectral Steepest Descent on the Low-Rank Manifold | Franz Louis Cesista, Katherine Crowson, Cédric Simal, Stella Biderman | 2026-06 | rlhftraining-dynamics | 2026-06-25 | 4 | 1 | |
| "Did you lie?" Evaluating Lie Detectors across Model Scale and Belief-Verified Model Organisms | Alan Cooney, David Africa, Geoffrey Irving | 2026-06 | evalsdeceptionmodel-organismscot-faithfulnesssteering | 2026-06-25 | 5 | 1 | group rating 4/5 |
| DIRECT: When and Where Should You Allocate Test-Time Compute in Embodied Planners? | Jadelynn Dao, Milan Ganai, Yasmina Abukhadra, Ajay Sridhar, Mozhgan Nasr Azadani, Katie Luo, Clark Barrett, Jiajun Wu, Chelsea Finn, Marco Pavone | 2026-06 | agentscapabilitiesscaling | 2026-06-25 | 6 | 1 | |
| CHORUS: Decentralized Multi-Embodiment Collaboration with One VLA Policy | Ria Doshi, Tian Gao, Annie Chen, Chelsea Finn, Jeannette Bohg | 2026-06 | agentsmulti-agent | 2026-06-25 | 7 | 1 | |
| Bergson: An Open Source Library for Data Attribution | Lucia Quirke, Louis Jaburi, David Johnston, William Z. Li, Gonçalo Paulo, Guillaume Martres, Girish Gupta, Stella Biderman, Nora Belrose | 2026-06 | interptooling | 2026-06-25 | 8 | 1 | |
| Evaluation Cards: An Interpretive Layer for AI Evaluation Reporting | Avijit Ghosh, Anka Reuel, Jenny Chim, Wm. Matthew Kennedy, Srishti Yadav, Jennifer Mickel, Yanan Long, Andrew Tran, Anastassia Kornilova, Damian Stachura, Kevin Klyman, Felix Friedrich, Jeba Sania, Jan Batzner, Anoop Mishra, Eliya Habba, Yixiong Hao, Nathan Heath, Shalaleh Rismani, Usman Gohar, Andrea Loehr, David Manheim, Ruchira Dhar, Sree Harsha Nelaturu, Aarush Sinha, Leshem Choshen, Drishti Sharma, Ishan Khire, Amit Saha, Subramanyam Sahoo, Michael Hardy, Michael Alexander Riegler, Kabir Manghnani, Michelle Lin, Yanan Jiang, Yilin Huang, Asaf Yehudai, Jessica Ji, Aris Hofmann, Mubashara Akhtar, Max Lamparth, Nuno Moniz, Yacine Jernite, Stella Biderman, Zeerak Talat, Sanmi Koyejo, Mykel Kochenderfer, Irene Solaiman | 2026-06 | evalsbenchmarkstooling | 2026-06-25 | 9 | 1 | |
| End-to-End Context Compression at Scale | Ang Li, Sean McLeish, Haozhe Chen, Nimit Kalra, Zaiqian Chen, Artem Gazizov, Venkata Anoop Suhas Kumar Morisetty, Bhavya Kailkhura, Harshitha Menon, Zhuang Liu, Brian R. Bartoldson, Tom Goldstein, Sanae Lotfi, Micah Goldblum, Pavel Izmailov | 2026-06 | capabilitiesscaling | 2026-06-25 | 10 | 1 | |
| Anatomy of Post-Training: Using Interpretability to Characterize Data and Shape the Learning Signal | Leon Bergen, Usha Bhalla, Sidharth Baskaran, Max Loeffler, Raphael Sarfati, Dhruvil Gala, Ryan Panwar, Santiago Aranguri, Thomas Fel, Atticus Geiger, Matthew Kowal, Siddharth Boppana, Daniel Balsam, Owen Lewis, Jack Merullo, Thomas McGrath, Ekdeep Singh Lubana | 2026-06 | interprlhfsycophancyreward-hacking | 2026-06-25 | 11 | 1 | |
| Sycophancy Towards Researchers Drives Performative Misalignment | David D. Baek, Xinnuo Li, Anay Gupta, Taslim Mahbub, Kejian Shi, Max Tegmark, Shi Feng | 2026-06 | deceptionsituational-awarenesssycophancyinterpsteering | 2026-06-25 | 12 | 1 | |
| Gaze Heads: How VLMs Look at What They Describe | Rohit Gandikota, David Bau | 2026-06 | interpsteering | 2026-06-25 | 13 | 2 | |
| MLSN #21: Political Manipulation and Indirect Prompt Injection | 2026-06 | sycophancyevalsrlhf | 2026-06-25 | 14 | 1 | ||
| OpenAI isn’t being consistently candid about Leading the Future | 2026-06 | governance | 2026-06-25 | 15 | 1 | ||
| Frontier post-training recipe review with Finbarr Timbers | 2026-06 | rlhf | 2026-06-25 | 1 | 6 | ||
| How Transparent is DiffusionGemma? | Joshua Engels, Callum McDougall, Bilal Chughtai, Janos Kramar, Senthoran Rajamanoharan, Cindy Wu, Arthur Conmy, Asic Q Chen, Jean Tarbouriech, Min Ma, Brendan O'Donoghue, João Gabriel Lopes de Oliveira, Rohin Shah, Neel Nanda | 2026-06 | interpcot-faithfulnessoversight | 2026-06-25 | 2 | 2 | |
| Banning Open Source AI Would Be A Mistake | 2026-06 | policy | 2026-06-25 | 3 | 1 | ||
| Fantastic Pretraining Optimizers and Where to Find Them II: Hyperball Optimization | Kaiyue Wen, Xingyu Dang, Kaifeng Lyu, Tengyu Ma, Percy Liang | 2026-06 | scaling | 2026-06-25 | 4 | 4 | |
| Native Active Perception as Reasoning for Omni-Modal Understanding | Zhenghao Xing, Ruiyang Xu, Yuxuan Wang, Jinzheng He, Ziyang Ma, Qize Yang, Yunfei Chu, Jin Xu, Junyang Lin, Chi-Wing Fu, Pheng-Ann Heng | 2026-06 | capabilitiesagents | 2026-06-25 | 5 | 2 | |
| Unified Multimodal Autoregressive Modeling with Shared Context-Visual Tokenizer is Key to Unification | Wujian Peng, Lingchen Meng, Yuxuan Cai, Xianwei Zhuang, Yuhuan Yang, Rongyao Fang, Chenfei Wu, Junyang Lin, Zuxuan Wu, Shuai Bai | 2026-06 | capabilities | 2026-06-25 | 6 | 2 | |
| Qwen-RobotNav Technical Report: A Scalable Navigation Model Designed for an Agentic Navigation System | Jiazhao Zhang, Gengze Zhou, Hale Yin, Yiyang Huang, Zixing Lei, Qihang Peng, Haoqi Yuan, Jie Zhang, Xudong Guo, Xiaoyue Chen, An Yang, Fei Huang, Zhibo Yang, Junyang Lin, Dayiheng Liu, Jingren Zhou, Zhuoyuan Yu, Jingyang Fan, Zhixuan Liang, Pei Lin, Ye Wang, Anzhe Chen, Kun Yan, Xiao Xu, Jiahao Li, Lulu Hu, Minying Zhang, Shurui Li, Wenhu Xiao, Shuai Bai, Xuancheng Ren, Chenxu Lv, Chenfei Wu, Xiong-Hui Chen | 2026-06 | capabilitiesagents | 2026-06-25 | 7 | 2 | |
| Qwen-RobotManip Technical Report: Alignment Unlocks Scale for Robotic Manipulation Foundation Models | Haoqi Yuan, Zhixuan Liang, Anzhe Chen, Ye Wang, Haoyang Li, Pei Lin, Yiyang Huang, Zixing Lei, Tong Zhang, Jiazhao Zhang, Jie Zhang, Jingyang Fan, Gengze Zhou, Qihang Peng, Chenxu Lv, Xiaoyue Chen, An Yang, Fei Huang, Junyang Lin, Dayiheng Liu, Jingren Zhou, Chenfei Wu, Xiong-Hui Chen | 2026-06 | capabilities | 2026-06-25 | 8 | 5 | |
| Where Does Social Reasoning Come From? Capability Provenance in Language Models | Glenn Matlin, Chandreyi Chakraborty, Saehee Eom, Mika Okamoto, Rayan Castilla, Louis Jaburi, Alvin Deng, Taywon Min, Lucia Quirke, Stella Biderman, Mark Riedl | 2026-06 | interptraining-dynamicsunlearning | 2026-06-25 | 9 | 2 | |
| ReproRepo: Scaling Reproducibility Audits with GitHub Repository Issues | Shanda Li, Qiuhong Anna Wei, Jingwu Tang, Valerie Chen, Nihar B Shah, Tim Dettmers, Yiming Yang, Ameet Talwalkar | 2026-06 | agentsbenchmarkstooling | 2026-06-25 | 10 | 2 | |
| Import AI 461: “Alignment is not on track”; FrontierCode; and synthetic research interns | 2026-06 | meta | 2026-06-25 | 11 | 1 | ||
| Welcome to the AGI era of AI governance | 2026-06 | governance | 2026-06-25 | 12 | 1 | ||
| A near-autonomous AI chemist improves a challenging reaction in medicinal chemistry | 2026-06 | capabilitiesagents | 2026-06-25 | 13 | 1 | ||
| Predicting model behavior before release by simulating deployment | 2026-06 | evals | 2026-06-25 | 14 | 1 | ||
| Breaking Entropy Bounds: Accelerating RL Training via MTP with Rejection Sampling | Yucheng Li, Huiqiang Jiang, Yang Xu, Jianxin Yang, Yi Zhang, Yizhong Cao, Yuhao Shen, Fan Zhou, Rui Men, Jianwei Zhang, An Yang, Bowen Yu, Bo Zheng, Fei Huang, Junyang Lin, Dayiheng Liu, Jingren Zhou | 2026-06 | rltraining-dynamicsrlhf | 2026-06-25 | 15 | 1 |