mlsanityData to 14 Oct 2026

ImpossibleBench: Measuring LLMs' Propensity of Exploiting Test Cases

arXiv 2510.20270 ↗First posted 23 Oct 2025cs.LG cs.CLcited by 65

Not on the map. Why?

Cited by 26 in LLM agents and tool use, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

The tendency to find and exploit "shortcuts" to complete tasks poses significant risks for reliable assessment and deployment of large language models (LLMs). For example, an LLM agent with access to unit tests may delete failing tests rather than fix the underlying bug. Such behavior undermines both the validity of benchmark results and the reliability of real-world LLM coding assistant deployments. To quantify, study, and mitigate such behavior, we introduce ImpossibleBench, a benchmark framework that systematically measures LLM agents' propensity to exploit test cases. ImpossibleBench creates "impossible" variants of tasks from existing benchmarks like LiveCodeBench and SWE-bench by introducing direct conflicts between the natural-language specification and the unit tests. We measure an agent's "cheating rate" as its pass rate on these impossible tasks, where any pass necessarily implies a specification-violating shortcut. As a practical framework, ImpossibleBench is not just an evaluation but a versatile tool. We demonstrate its utility for: (1) studying model behaviors, revealing more fine-grained details of cheating behaviors from simple test modification to complex operator overloading; (2) context engineering, showing how prompt, test access and feedback loop affect cheating rates; and (3) developing monitoring tools, providing a testbed with verified deceptive solutions. We hope ImpossibleBench serves as a useful framework for building more robust and reliable LLM systems. Our implementation can be found at https://github.com/safety-research/impossiblebench.

Key facts. Cited by 65 papers in the map (data through 14 Oct 2026). Builds on 12 of its 18 references that are in the map. First posted on arXiv 23 Oct 2025.

Builds on

12 of 18 references.

References · 12

  1. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  2. 2021Measuring Mathematical Problem Solving With the MATH Datasetcited by 5,371
  3. 2023SWE-bench: Can Language Models Resolve Real-World GitHub Issues?cited by 3,373
  4. 2024LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Codecited by 2,004
  5. 2024SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineeringcited by 1,737
  6. 2016Concrete Problems in AI Safetycited by 1,642
  7. 2024Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preferencecited by 1,184
  8. 2025Constitutional Classifiers: Defending against Universal Jailbreaks across Thousands of Hours of Red Teamingcited by 181
  9. 2024Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Modelscited by 178
  10. 2024Feedback Loops With Language Models Drive In-Context Reward Hackingcited by 77
  11. 2025Know Thy Judge: On the Robustness Meta-Evaluation of LLM Safety Judgescited by 24
  12. 2025Breakpoint: Scalable evaluation of system-level reasoning in LLM code agentscited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2025: 3; 2026: 62.

65 papers.

Sort:

By area:

  1. 2026Emotion Concepts and their Function in a Large Language Modelcited by 57
  2. 2026Reward Hacking Benchmark: Measuring Exploits in LLM Agents with Tool Usecited by 31
  3. 2025EvilGenie: A Reward Hacking Benchmarkcited by 28
  4. 2026Benchmarking Reward Hack Detection in Code Environments via Contrastive Analysiscited by 18
  5. 2026Countdown-Code: A Testbed for Studying The Emergence and Generalization of Reward Hacking in RLVRcited by 17
  6. 2026AgentAbstain: Do LLM Agents Know When Not to Act?cited by 17
  7. 2025Are Your Agents Upward Deceivers?cited by 16
  8. 2025Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thoughtcited by 15
  9. 2026Detecting Safety Violations Across Many Agent Tracescited by 14
  10. 2026The Obfuscation Atlas: Mapping Where Honesty Emerges in RLVR with Deception Probescited by 12
  11. 2026Hack-Verifiable Environments: Towards Evaluating Reward Hacking at Scalecited by 10
  12. 2026Hardening Agent Benchmarks with Adversarial Hacker-Fixer Loopscited by 9
38 more
  1. 2026SEVerA: Verified Synthesis of Self-Evolving Agentscited by 8
  2. 2026CoffeeBench: Benchmarking Long-Horizon LLM Agents in Heterogeneous Multi-Agent Economiescited by 8
  3. 2026Reasoning Models Will Sometimes Lie About Their Reasoningcited by 6
  4. 2026Bridging Online and Offline RL: Contextual Bandit Learning for Multi-Turn Code Generationcited by 4
  5. 2026BAITBENCH: Measuring Agent Reward Hacking with Optional Shortcuts Planted in ML Taskscited by 4
  6. 2026Quantifying Overclaiming Propensity in Frontier LLM Agentscited by 4
  7. 2026ForesightSafety Bench: A Frontier Risk Evaluation and Governance Framework towards Safe AIcited by 3
  8. 2026Hodoscope: Unsupervised Monitoring for AI Misbehaviorscited by 3
  9. 2026Towards Understanding Specification Gaming in Reasoning Modelscited by 3
  10. 2026Do Coding Agents Deceive Us? Detecting and Preventing Cheating via Capped Evaluation with Randomized Testscited by 3
  11. 2026The Verifier is the Curriculum: Precision Sets the Return on Search in Code Self-Distillationcited by 3
  12. 2026Every Model Cheats: Prompt-Level Mitigation of Cheating on Offensive Cyber Taskscited by 3
  13. 2026When Agents Coordinate: Measuring Coordination in Multi-Agent AI Codingcited by 3
  14. 2026Trust No Tool: Evaluating and Defending LLM Agents under Untrusted Tool Feedbackcited by 2
  15. 2026Open-source LLMs administer maximum electric shocks in a Milgram-like obedience experimentcited by 2
  16. 2026Monitoring and Discovering Reward Hacking with Internal Representations during LLM Evaluationscited by 2
  17. 2026Emergent Collusion in Long-Horizon LLM Agent Interactioncited by 2
  18. 2026Reward Hacking Challenges Oversight of Autonomous Research Agentscited by 2
  19. 2026Consequentialist Objectives and Catastrophecited by 1
  20. 2026Muse Spark Safety & Preparedness Reportcited by 1
  21. 2026Model Forensics: Investigating Whether Concerning Behavior Reflects Misalignmentcited by 1
  22. 2026Amplified Does Not Mean Predictive: Reasoning Behaviors in Thinking Modelscited by 1
  23. 2026Maintaining Benchmarks Against Increasingly Capable Agents: Detection and Remediation of Unearned Passescited by 1
  24. 2026CheatBench: Measuring Reward Gaming in AI Agentscited by 1
  25. 2026LLMs Learn to Evade Latent Monitors from Prior Feedback Alonecited by 1
  26. 2026RH-Detect: A Unified Benchmark for Reward Hacking Detectioncited by 1
  27. 2026CoT-Guard: Small Models for Strong Monitoringcited by 0
  28. 2026Greed Is Learned: Visible Incentives as Reward-Hacking Triggerscited by 0
  29. 2026Autoresearch with Coding Agents: Generalizers and Metric-Maximizers on Quran Recitation Datacited by 0
  30. 2026Large Language Models Can Follow Instructions, But Not Many at Once: Phase Transitions in Compositional Constraint Satisfactioncited by 0
  31. 2026Hack-Verifiable Terminal Bench: Evaluating Reward Hacking in Terminal Taskscited by 0
  32. 2026Can escalation channels redirect reward hacking toward defect disclosure?cited by 0
  33. 2026Feyospace-v1: How the Cyber Mercury Seven Trained Frontier Cyber Modelscited by 0
  34. 2026Shallow Beliefs: Synthetic document finetuning does not inoculate against emergent misalignment from reward hackingcited by 0
  35. 2026The Troy Moment: How LLM Agents Adjudicate the Decision Point Under Impossible Tasks, Claimed Authority, and Peer Informationcited by 0
  36. 2026Compiled Agency: Frontier General-Purpose Coding Agents Build Winning Game Players from Bare Interaction - from Flappy Bird to StarCraft II and Civilizationcited by 0
  37. 2026RECLAIM: Can Agents Reproduce the Claims of Machine Learning Papers?cited by 0
  38. 2026Instrumental Monitor Evasion Emerges Under Ordinary Task Pressurecited by 0

50 most cited of 65 All 65 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026Who Tests the Testers? Systematic Enumeration and Coverage Audit of LLM Agent Tool Call Safety
  2. 2025Challenges in Testing Large Language Model Based Software: A Faceted Taxonomy
  3. 2025Artificial or Just Artful? Do LLMs Bend the Rules in Programming? ↗
  4. 2025LMR-BENCH: Evaluating LLM Agent's Ability on Reproducing Language Modeling Research
  5. 2025Rethinking Verification for LLM Code Generation: From Generation to Testing
  6. 2026How Generation Architecture Shapes Code Complexity in Multi-Agent LLM Systems: A Paired Study on HumanEval
  7. 2026TestJack: Should you trust the results in coding benchmarks? Agentic Coding Benchmarks Auditing via Evaluator Evolution
  8. 2025On the Soundness and Consistency of LLM Agents for Executing Test Cases Written in Natural Language