# SpecBench: Measuring Reward Hacking in Long-Horizon Coding Agents

arXiv: https://arxiv.org/abs/2605.21384 · HTML: https://mlsanity.com/paper/2605.21384/

## Key facts

- Cited by 28 papers in the map (data through 14 Oct 2026).
- Placed in Software engineering coding agents, one of 16 directions in LLM agents and tool use.
- Builds on 37 of its 55 references that are in the map.
- First posted on arXiv 20 May 2026.

## Placed in

- LLM agents and tool use / [Software engineering coding agents](https://mlsanity.com/agents/software-engineering-coding-agents/)
- LLM alignment / RLHF / preference optimisation / [Reward hacking and misspecification](https://mlsanity.com/alignment/reward-hacking-and-misspecification/)

## Builds on (most cited first, 30 of 37)

- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Program Synthesis with Large Language Models](https://mlsanity.com/paper/2108.07732/) (2021)
- [SWE-bench: Can Language Models Resolve Real-World GitHub Issues?](https://mlsanity.com/paper/2310.06770/) (2023)
- [Tree of Thoughts: Deliberate Problem Solving with Large Language Models](https://mlsanity.com/paper/2305.10601/) (2023)
- [LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code](https://mlsanity.com/paper/2403.07974/) (2024)
- [MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework](https://mlsanity.com/paper/2308.00352/) (2023)
- [Concrete Problems in AI Safety](https://mlsanity.com/paper/1606.06565/) (2016)
- [Scaling Laws for Reward Model Overoptimization](https://mlsanity.com/paper/2210.10760/) (2022)
- [DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence](https://mlsanity.com/paper/2606.19348/) (2026)
- [AlphaEvolve: A coding agent for scientific and algorithmic discovery](https://mlsanity.com/paper/2506.13131/) (2025)
- [DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models](https://mlsanity.com/paper/2512.02556/) (2025)
- [Deep Reinforcement Learning at the Edge of the Statistical Precipice](https://mlsanity.com/paper/2108.13264/) (2021)
- [Automated Design of Agentic Systems](https://mlsanity.com/paper/2408.08435/) (2024)
- [Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation](https://mlsanity.com/paper/2503.11926/) (2025)
- [Alignment faking in large language models](https://mlsanity.com/paper/2412.14093/) (2024)
- [Darwin Godel Machine: Open-Ended Evolution of Self-Improving Agents](https://mlsanity.com/paper/2505.22954/) (2025)
- [The Effects of Reward Misspecification: Mapping and Mitigating Misaligned Models](https://mlsanity.com/paper/2201.03544/) (2022)
- [SWE-Bench Pro: Can AI Agents Solve Long-Horizon Software Engineering Tasks?](https://mlsanity.com/paper/2509.16941/) (2025)
- [AIDE: AI-Driven Exploration in the Space of Code](https://mlsanity.com/paper/2502.13138/) (2025)
- [KernelBench: Can LLMs Write Efficient GPU Kernels?](https://mlsanity.com/paper/2502.10517/) (2025)
- [Sycophancy to Subterfuge: Investigating Reward-Tampering in Large Language Models](https://mlsanity.com/paper/2406.10162/) (2024)
- [ClassEval: A Manually-Crafted Benchmark for Evaluating LLMs on Class-level Code Generation](https://mlsanity.com/paper/2308.01861/) (2023)
- [Categorizing Variants of Goodhart's Law](https://mlsanity.com/paper/1803.04585/) (2018)
- [Qwen3-Coder-Next Technical Report](https://mlsanity.com/paper/2603.00729/) (2026)
- [NL2Repo-Bench: Towards Long-Horizon Repository Generation Evaluation of Coding Agents](https://mlsanity.com/paper/2512.12730/) (2025)
- [Correlated Proxies: A New Definition and Improved Mitigation for Reward Hacking](https://mlsanity.com/paper/2403.03185/) (2024)
- [Prompting Large Language Models to Tackle the Full Software Development Lifecycle: A Case Study](https://mlsanity.com/paper/2403.08604/) (2024)
- [ALE-Bench: A Benchmark for Long-Horizon Objective-Driven Algorithm Engineering](https://mlsanity.com/paper/2506.09050/) (2025)
- [Reward Hacking in the Era of Large Models: Mechanisms, Emergent Misalignment, Challenges](https://mlsanity.com/paper/2604.13602/) (2026)

## Built on by (most cited first, 28 of 28)

- [SWE-Marathon: Can Agents Autonomously Complete Ultra-Long-Horizon Software Work?](https://mlsanity.com/paper/2606.07682/) (2026)
- [Recursive self-improvement of AI research agents](https://mlsanity.com/paper/2609.26457/) (2026)
- [Safety in Self-Evolving LLM Agent Systems: Threats, Amplification, and Case Studies](https://mlsanity.com/paper/2606.23075/) (2026)
- [Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI](https://mlsanity.com/paper/2607.22368/) (2026)
- [Don't Blame the Large Language Model: How Agent Harness Evolution Shapes Coding Agent Quality](https://mlsanity.com/paper/2607.03691/) (2026)
- [Building to the Test: Coding Agents Deliver What You Check, Not What You Requested](https://mlsanity.com/paper/2606.28430/) (2026)
- [When the Reward Suite Is Leaky: A Preregistered Causal Contrast of Natural Verifier False Positives in RLVR](https://mlsanity.com/paper/2607.11022/) (2026)
- [When Do Agent Loops Mistake Stagnation for Progress? Self-Evaluation Bias and Externally Grounded Verification in Long-Running Autonomous LLM Agent Loops](https://mlsanity.com/paper/2607.25152/) (2026)
- [Harness-IF: Evaluating Instruction Following Across Instruction Surfaces in Coding Agents](https://mlsanity.com/paper/2608.11727/) (2026)
- [Auditing Harness Tampering in Self-Improving Agents](https://mlsanity.com/paper/2609.00069/) (2026)
- [Commit-first LLM judging inherits the judge's own errors](https://mlsanity.com/paper/2609.00088/) (2026)
- [Reward Hacking Challenges Oversight of Autonomous Research Agents](https://mlsanity.com/paper/2609.28614/) (2026)
- [Looping Is Not Reliability: State-Bound Evidence and Typed Revision Contracts for Agentic Code Repair](https://mlsanity.com/paper/2607.24604/) (2026)
- [Reward Hacking and Agent Containment Failure: A Monte Carlo Study Based on the 2026 Hugging Face Incident](https://mlsanity.com/paper/2609.32390/) (2026)
- [Maintaining Benchmarks Against Increasingly Capable Agents: Detection and Remediation of Unearned Passes](https://mlsanity.com/paper/2609.34262/) (2026)
- [Groundability, Not Scale Alone: When Weak Reviewers Can Audit Strong Coding Agents](https://mlsanity.com/paper/2610.01023/) (2026)
- [Leveraging Instruction Tuning and Merging for Reasoning Model Adaptation](https://mlsanity.com/paper/2607.14895/) (2026)
- [Autoresearch with Coding Agents: Generalizers and Metric-Maximizers on Quran Recitation Data](https://mlsanity.com/paper/2607.18064/) (2026)
- [Can escalation channels redirect reward hacking toward defect disclosure?](https://mlsanity.com/paper/2608.29460/) (2026)
- [One to More, More to One: Category-Aware Iterative Expert Training for Software Engineering Agents](https://mlsanity.com/paper/2609.23377/) (2026)
- [Identical Runs, Different Results: Benchmarking AI Coding Agents on Open-Weight Models](https://mlsanity.com/paper/2609.33812/) (2026)
- [From Dead Code and Static Requirements to Working Engines: Software Revival with Coding Agents](https://mlsanity.com/paper/2609.36161/) (2026)
- [A Competing-Hazards Systematization of Loss of Control in Autonomous Agents](https://mlsanity.com/paper/2609.38411/) (2026)
- [Verify Claims, Not Scores: Evidence-Based Verification of Modular Agents](https://mlsanity.com/paper/2610.01348/) (2026)
- [SpecGuard: Proving a Task Is Broken Before the Agent Cheats](https://mlsanity.com/paper/2610.09159/) (2026)
- [StoreBench: A Live-Commerce Environment for Evaluating and Training Autonomous Operator Agents](https://mlsanity.com/paper/2610.10942/) (2026)
- [AgentSWE: Can Coding Agents Build the Agent You Actually Want?](https://mlsanity.com/paper/2610.13361/) (2026)
- [Scalable Weak-to-Strong Verifier Hardening for Agent Environments](https://mlsanity.com/paper/2610.15924/) (2026)

## Abstract (arXiv)

As long-horizon coding agents produce more code than any developer can review, oversight collapses onto a single surface: the automated test suite. Reward hacking naturally arises in this setup, as the agent optimizes for passing tests while deviating from the users true goal. We study this reward hacking phenomenon by decompose software engineering tasks into three parts: (i) a natural language description of the specification (ii) visible validation tests that exercise specified features in isolation, and (iii) held-out tests that compose those same features to simulate real-world usage. Based on the specification and the visible validation test suites, a genuine agent would be able to generate a solution that can also pass all of the held-out tests. Therefore we use the gap in pass rates on these two suites to quantify reward hacking. Based on this methodology, we introduce SpecBench, a benchmark comprising 30 systems-level programming tasks ranging from short horizon tasks like building a JSON parser to ultra long horizon tasks like building an entire OS kernel from scratch. Large-scale experiments reveal a consistent pattern: while every frontier agent saturates the visible suite, reward hacking persists, with smaller models exhibiting larger gaps on holdout suites. The gap also scales sharply with task length: it grows by 28 percentage points for every tenfold increase in code size. Failures range from subtle feature isolation to deliberate exploits, including a 2,900-line hash-table "compiler" that memorizes test inputs. SpecBench offers a principled testbed for measuring whether coding agents build genuine working systems or merely game the test suites developers hand them.