Concept · 48 episode(s)

Reward Hacking

← all concepts

Definition

Reward hacking is when a learning system finds a way to score high on its reward signal without doing the thing the reward was supposed to encourage. Classic examples include exploiting bugs in the reward function, gaming the grader, or finding shortcuts that satisfy the letter and not the spirit of the metric.

Episodes covering this

  1. 207
    An AI Graded Its Own Math Test 94 Percent — It Actually Scored 20
    More Convincing, Not More Correct: Self-Play Reward Hacking of Reference-Free LLM Judges
    · ·12 min·Jul 08, 2026
  2. 203
    The Thought a Model Doesn't Say — and the Lens That Reads It
    Verbalizable Representations Form a Global Workspace in Language Models
    Gurnee, Sofroniew, Pearce et al. · Anthropic·16 min·Jul 07, 2026
  3. 199
    Finding a Model's Hidden Behaviors Without Knowing What You're Looking For
    Mechanistically Eliciting Latent Behaviors in Language Models
    Mack, Panickssery, Turner · Principles of Intelligence·15 min·Jul 04, 2026
  4. 178
    How an AI Reviewer Learned to Stop Going Easy on AI Writing
    The Red Queen Gödel Machine: Co-Evolving Agents and Their Evaluators
    Iacob, Jovanović, Shen et al. · University of Cambridge·23 min·Jun 26, 2026
  5. 174
    When the AI 'Schemes,' It's Usually Just Lazy or Confused
    Model Forensics: Investigating Whether Concerning Behavior Reflects Misalignment
    Singh, Kroiz, Rajamanoharan et al. · MATS·28 min·Jun 25, 2026
  6. 171
    The Safety Decision a Model Makes Before It Thinks a Word
    Do Thinking Tokens Help with Safety?
    Ri, Panigrahi, Arora · Princeton Language and Intelligence·25 min·Jun 25, 2026
  7. 167
    How Teaching an AI to Predict, Not Act, Made It a Better Actor
    Qwen-AgentWorld: Language World Models for General Agents
    Team, Zuo, Xiao et al. · ·27 min·Jun 24, 2026
  8. 159
    Can a Coding Agent Run Its Own Robot Experiments Overnight, With No Human Resetting the Scene?
    ENPIRE: Agentic Robot Policy Self-Improvement in the Real World
    Xiao, Xie, Zhang et al. · NVIDIA·23 min·Jun 19, 2026
  9. 151
    Why More Experience Made This AI Agent Worse, And How to Fix It
    Not All Skills Help: Measuring and Repairing Agent Knowledge
    Wang, Zhou, Liang et al. · UNC Chapel Hill·28 min·Jun 16, 2026
  10. 149
    When Cornering a Chatbot Makes It Lie: J.P. Morgan's Case for 'Playing Dead'
    Is Your Agent Playing Dead? Deployed LLM Agents Exhibit Constraint-Evasive Fabrication and Thanatosis
    Rodríguez, Pozanco, Borrajo · J.P. Morgan AI Research·23 min·Jun 16, 2026
  11. 148
    Why Letting an AI Watch Its Own Scoreboard Can Quietly Overwrite Its Safety
    Greed Is Learned: Visible Incentives as Reward-Hacking Triggers
    Che, Wu · NVIDIA Research·26 min·Jun 16, 2026
  12. 147
    Agents Fail at the Body, Not the Brain: A Self-Rewriting Scaffold That Lifts a 9B Model 44 Points
    HarnessX: A Composable, Adaptive, and Evolvable Agent Harness Foundry
    Chen, Lu, Zhao et al. · ·30 min·Jun 15, 2026
  13. 146
    How an Innocent README Can Freeze an AI Agent's Safety Check for an Hour
    From Shield to Target: Denial-of-Service Attacks on LLM-Based Agent Guardrails
    Zhou, Wang, Ma et al. · Hong Kong University of Science and Technology·26 min·Jun 15, 2026
  14. 145
    Building Forgetting Into a Language Model With One Extra Line of Code
    Natively Unlearnable Large Language Models
    Ghosal, Maini, Raghunathan · Carnegie Mellon University·22 min·Jun 15, 2026
  15. 144
    When an AI Agent Just Copies Its Tool — And Bigger Models Copy More
    When the Tool Decides: LLM Agents Defer Blindly to Graph Neural Network Tools, and Stronger Backbones Defer More
    Wang, Vemuri · raptorX.ai·15 min·Jun 15, 2026
  16. 141
    How Two Tokens Reopened a Reasoning Method the Field Had Given Up On
    Demystifying Hidden-State Recurrence: Switchable Latent Reasoning with On-Policy Reinforcement Learning
    Yang, Chen, Wu et al. · HKUST(GZ)·29 min·Jun 12, 2026
  17. 139
    When Optimizing One GPU Kernel Quietly Breaks the Whole System
    Arbor: Tree Search as a Cognition Layer for Autonomous Agents
    Prakriya, Hou, Gong et al. · AMD·30 min·Jun 12, 2026
  18. 133
    How MiniMax Turned a Reward-Hacking Disaster Into Olympiad Gold
    MaxProof: Scaling Mathematical Proof with Generative-Verifier RL and Population-Level Test-Time Scaling
    Chen, Zhang, Zhang et al. · MiniMax / The Chinese University of Hong Kong·34 min·Jun 12, 2026
  19. 132
    The Agent Failed — But Did the Instructions Deserve to Be Followed?
    SkillAxe: Sharpening LLM-Authored Agent Skills Through Evaluation-Guided Self-Refinement
    Gautam, Radhakrishna, Gulwani · Microsoft·30 min·Jun 11, 2026
  20. 131
    Why Autonomous Research Agents Forget Their Own Lessons, and Arbor's Fix
    Toward Generalist Autonomous Research via Hypothesis-Tree Refinement
    Jin, Hu, Qiu et al. · Renmin University of China·33 min·Jun 11, 2026
  21. 128
    How a Model Can Earn Full Reward and Still Resist Training
    Generalization Hacking: Models Can Game Reinforcement Learning by Preventing Behavioral Generalization
    Xiao, Phuong · California Institute of Technology·29 min·Jun 11, 2026
  22. 125
    AI Coding Agents Run a Marathon, and Fewer Than One in Three Finish
    SWE-Marathon: Can Agents Autonomously Complete Ultra-Long-Horizon Software Work?
    Desai, Hu, Cabezas et al. · Abundant·27 min·Jun 09, 2026
  23. 124
    A Cheap Model With the Blueprints Beats Expensive Models Working Blind
    Hardening Agent Benchmarks with Adversarial Hacker-Fixer Loops
    Zhong, Segal, Bercovich et al. · Carnegie Mellon University·27 min·Jun 09, 2026
  24. 121
    When the Agent Says It's Done But Nothing Happened: Debugging the Harness, Not the Model
    From Failed Trajectories to Reliable LLM Agents: Diagnosing and Repairing Harness Flaws
    Chen, Wang, Liu et al. · Institute of Software·27 min·Jun 05, 2026
  25. 118
    Why the Best-Aligned AI Models Are the Easiest to Trick Into Producing Harm
    Safety Paradox: How Enhanced Safety Awareness Leaves LLMs Vulnerable to Posterior Attack
    Hoang, Le, Xu et al. · Singapore University of Technology and Design·23 min·Jun 05, 2026
  26. 112
    When an AI Agent Cheats Without Being Told: Inside the Meta-Agent Challenge
    The Meta-Agent Challenge: Are Current Agents Capable of Autonomous Agent Development?
    Lu, Wang, Wang et al. · Institute of Software·22 min·Jun 04, 2026
  27. 111
    How a 4B Web Agent Beat Models 60x Its Size on 500 Demonstrations
    OpenWebRL: Demystifying Online Multi-turn Reinforcement Learning for Visual Web Agents
    Yang, Wu, Chen et al. · UIUC·24 min·Jun 03, 2026
  28. 109
    An AI Got Caught Reading the Answer Key, And Why That Catch Matters
    EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning
    Chen, Shi, Li et al. · Shenzhen Institutes of Advanced Technology·28 min·Jun 03, 2026
  29. 104
    How Making a Research Agent Smarter Quietly Makes It Leak Your Secrets
    MosaicLeaks:Privacy Risks in Querying-in-the-Open for Deep Research Agents
    Gurung, Gella, Drouin et al. · University of Edinburgh·25 min·Jun 01, 2026
  30. 101
    Treating Math Formalization Like a Codebase, and Where the Agents Cheat
    Formalizing Mathematics at Scale
    Rammal, Patel, Gloeckle et al. · FAIR at Meta / CERMICS·27 min·May 29, 2026
  31. 094
    Chain-of-Thought Monitoring Fails Across Languages, and Worst Where It's Needed Most
    The Fragility of Chain-of-Thought Monitoring Across Typologically Diverse Languages
    Onyame, Zhou, Thopalli et al. · University of Virginia·24 min·May 28, 2026
  32. 092
    When Search Agents Don't Really Search: The Memory Shortcut Hiding in Browsing Benchmarks
    LiveBrowseComp: Are Search Agents Searching, or Just Verifying What They Already Know?
    Fan, Wang, Chu et al. · Harbin Institute of Technology·27 min·May 28, 2026
  33. 089
    When AI-Written Papers Read Well But the Evidence Underneath Is Broken
    ScientistOne: Towards Human-Level Autonomous Research via Chain-of-Evidence
    Meng, Mishra, Chen et al. · Google Cloud AI Research·32 min·May 27, 2026
  34. 088
    Two Levers for Self-Improving AI: When Rewriting Code Isn't Enough
    SIA: Self Improving AI with Harness & Weight Updates
    Hebbar, Manawat, Verboomen et al. · Hexo Labs·25 min·May 27, 2026
  35. 082
    Training a Deep Research Agent on 8,000 Synthetic Tasks: The Rubric Tree Trick
    QUEST: Training Frontier Deep Research Agents with Fully Synthetic Tasks
    Xie, Lin, Wang et al. · The Ohio State University·31 min·May 26, 2026
  36. 080
    How a Two-Agent Trick Unlocked Large-Scale Training for Computer-Use Agents
    CUA-Gym: Scaling Verifiable Training Environments and Tasks for Computer-Use Agents
    Wang, Lu, Wang et al. · The University of Hong Kong·32 min·May 26, 2026
  37. 061
    When Helpful Agents Go Sideways: A 404 Error, Campus Security, and Why Alignment Misses This
    Agent Meltdowns: The Road to Hell Is Paved with Helpful Agents
    Jha, Triedman, Bhattacharya et al. · Cornell University·27 min·May 20, 2026
  38. 054
    When Models Learn the Monitor Exists, the Reasoning Trace Stops Being a Window
    Training on Documents About Monitoring Leads to CoT Obfuscation
    Haskins, Chughtai, Engels · University of Canterbury·26 min·May 18, 2026
  39. 053
    An AI Agent Swapped In Focal Loss And Beat A Human-Tuned Training Script
    Agentic Discovery of Neural Architectures: AIRA-Compose and AIRA-Design
    Pepe, Lin, Magka et al. · FAIR at Meta·32 min·May 18, 2026
  40. 052
    An Old Reinforcement Learning Tradeoff Sneaks Back Into LLM Agents
    Look Before You Leap: Autonomous Exploration for LLM Agents
    Ye, Shi, Liu et al. · University of Science and Technology of China / Meituan·23 min·May 18, 2026
  41. 049
    An AI Agent Reached for Root in Twelve Minutes, Without Being Attacked
    Ambient Persuasion in a Deployed AI Agent: Unauthorized Escalation Following Routine Non-Adversarial Content Exposure
    Cuadros, Maiga · Digital Epidemiology Laboratory·28 min·May 17, 2026
  42. 046
    When the AI Optimizer Edits the Grade Book: Why Harnessing Evolution Needs a Wall
    Harnessing Agentic Evolution
    Zhang, Gu, Ruan et al. · The Hong Kong University of Science and Technology (Guangzhou) / DeepWisdom·24 min·May 15, 2026
  43. 043
    When 'This Is False' Doesn't Stick: Why Models Learn the Lie Anyway
    Negation Neglect: When models fail to learn negations in training
    Mayne, McKinney, Dubiński et al. · University of Oxford·18 min·May 14, 2026
  44. 027
    When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure
    VibeServe: Can AI Agents Build Bespoke LLM Serving Systems?
    Kamahori, Li, Peter et al. · University of Washington·30 min·May 08, 2026
  45. 025
    The Missing Gradient Term That Predicts Sycophancy in RLHF
    Explaining and Preventing Alignment Collapse in Iterative RLHF
    Gauthier, Bach, Jordan · Inria·22 min·May 07, 2026
  46. 020
    The Compliance Gap: Why AI Says Yes and Does No
    The Compliance Gap: Why AI Systems Promise to Follow Process Instructions but Don't
    Shin · Polymath Minds AI Lab·28 min·May 06, 2026
  47. 019
    When the Best Reward Model Trains the Worst Policy: Inside EvoLM
    EvoLM: Self-Evolving Language Models through Co-Evolved Discriminative Rubrics
    Li, Xin, Xiao et al. · University of Washington·26 min·May 06, 2026
  48. 006
    What Happens Inside Claude When It Decides to Blackmail Someone
    Emotion Concepts and their Function in a Large Language Model
    Sofroniew, Kauvar, Saunders et al. · Anthropic·22 min·May 02, 2026

Worth reading next

Papers we haven't done a deep dive on yet, but would recommend on this topic.