mlsanityData to 7 Oct 2026

Hack-Verifiable Environments: Towards Evaluating Reward Hacking at Scale

arXiv 2605.20744 ↗First posted 20 May 2026cs.LG cs.AIcited by 9

On the map: LLM alignment / RLHF / preference optimisation → Reward hacking and misspecification

Abstract (arXiv)

Aligning autonomous agents with human intent remains a central challenge in modern AI. A key manifestation of this challenge is reward hacking, whereby agents appear successful under the evaluation signal while violating the intended objective. Reward hacking has been observed across a wide range of settings, yet methods for reliably measuring it at scale remain lacking. In this work, we introduce a new evaluation paradigm for measuring reward hacking. Whereas prior studies have primarily analyzed it post hoc by inspecting agent trajectories, we instead embed detectable reward hacking opportunities directly into environments. This makes their exploitation verifiable by design, enabling deterministic and automated measurement of whether and how agents exploit such vulnerabilities. We instantiate this approach in textitTextArena and release textitHack-Verifiable TextArena, a testbed in which reward hacking can be measured reliably. Using this benchmark, we analyze reward hacking behavior across language models in diverse environments and settings. We open source the code at https://github.com/MajoRoth/hack-verifiable-environments/.

Key facts. Cited by 9 papers in the map (data through 7 Oct 2026). Placed in Reward hacking and misspecification, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 26 of its 42 references that are in the map. First posted on arXiv 20 May 2026.

Builds on

26 of 42 references.

Closest ancestors · 8

Same direction, newest first

  1. 2026Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges15 Apr
  2. 2026Countdown-Code: A Testbed for Studying The Emergence and Generalization of Reward Hacking in RLVR7 Mar
  3. 2026Benchmarking Reward Hack Detection in Code Environments via Contrastive Analysis27 Jan
  4. 2025EvilGenie: A Reward Hacking Benchmark26 Nov
  5. 2025Natural Emergent Misalignment from Reward Hacking in Production RL23 Nov
  6. 2025School of Reward Hacks: Hacking harmless tasks generalizes to misaligned behavior in LLMs24 Aug
  7. 2025Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation14 Mar
  8. 2022The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models10 Jan

Other references · 18

  1. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
  2. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,193
  3. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 1,906
  4. 2016Concrete Problems in AI Safetycited by 1,582
  5. 2026Terminal-Bench: Benchmarking Agents on Hard, Realistic Tasks in Command Line Interfacescited by 515
  6. 2024Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingcited by 478
  7. 2025Accelerating scientific discovery with Co-Scientistcited by 353
  8. 2024Alignment faking in large language modelscited by 316
  9. 2026Meta-Harness: End-to-End Optimization of Model Harnessescited by 248
  10. 2026MiMo-V2-Flash Technical Reportcited by 163
  11. 2024Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Modelscited by 158
  12. 2024RE-Bench: Evaluating frontier AI R&D capabilities of language model agents against human expertscited by 153
6 more
  1. 2025ImpossibleBench: Measuring LLMs' Propensity of Exploiting Test Casescited by 51
  2. 2025Demonstrating specification gaming in reasoning modelscited by 48
  3. 2026AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviorscited by 32
  4. 2025Is It Thinking or Cheating? Detecting Implicit Reward Hacking by Measuring Reasoning Effortcited by 25
  5. 2026Terminal Wrench: A Dataset of 331 Reward-Hackable Environments and 3,632 Exploit Trajectoriescited by 14
  6. 2026RewardHackingAgents: Benchmarking Evaluation Integrity for LLM ML-Engineering Agentscited by 10

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 9.

9 papers.

  1. 2026SWE-Marathon: Can Agents Autonomously Complete Ultra-Long-Horizon Software Work?cited by 19
  2. 2026Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AIcited by 6
  3. 2026Auditing Harness Tampering in Self-Improving Agentscited by 3
  4. 2026Greed Is Learned: Visible Incentives as Reward-Hacking Triggerscited by 0
  5. 2026Hack-Verifiable Terminal Bench: Evaluating Reward Hacking in Terminal Taskscited by 0
  6. 2026BenchShield: Formal Model-Backed Instrumentation for Reward Integrity in LLM-Agent Evaluation Infrastructurecited by 0
  7. 2026Maintaining Benchmarks Against Increasingly Capable Agents: Detection and Remediation of Unearned Passescited by 0
  8. 2026CheatBench: Measuring Reward Gaming in AI Agentscited by 0
  9. 2026Conversational Task Disambiguation over Tabular Data: Leakage-Aware Formulation, Benchmark Suite, and Trainingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026BAITBENCH: Measuring Agent Reward Hacking with Optional Shortcuts Planted in ML Tasks
  2. 2026Reward Hacking Benchmark: Measuring Exploits in LLM Agents with Tool Use
  3. 2026Reward Hacking as Equilibrium under Finite Evaluation
  4. 2026Reward Hacking Challenges Oversight of Autonomous Research Agents
  5. 2026From Reward-Hack Activations to Agentic Risk States: Context-Calibrated Mechanistic Monitoring in LLM Agents
  6. 2026Reward Hacking in Language Model Agents: Revisiting AI Safety Gridworlds
  7. 2026Do Androids Dream of Breaking the Game? Systematically Auditing AI Agent Benchmarks with BenchJack
  8. 2026SpecBench: Measuring Reward Hacking in Long-Horizon Coding Agents