{"slug": "do-agent-benchmarks-measure-capability-protocol-validity-in-the-age-of-agentic", "title": "Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic", "summary": "A new post-hoc audit called HackDetect finds that agent benchmarks may overstate capabilities due to reward hacking and evaluation shortcuts. Auditing 2,385 traces across 15 agent benchmarks, researchers detected exposures and reward hacking in 67.0% of Frontier Science traces and 66.7% of AutoLab tasks, with score inflation of 0.45-1.00. The study introduces protocol validity and the Mislead gap metric to quantify misleading scores.", "body_md": "# Computer Science > Artificial Intelligence\n\n[Submitted on 24 Jul 2026]\n\n# Title:Do Agent Benchmarks Measure Capability? Protocol Validity in the Age of Agentic AI\n\n[View PDF](/pdf/2607.22368)\n\n[HTML (experimental)](https://arxiv.org/html/2607.22368v1)\n\nAbstract:Agent benchmarks increasingly evaluate repository editing, web research, terminal use, and long-horizon interaction. Their scores support capability claims only when the evaluation protocol keeps the intended capability necessary for success. Recent reward-hacking benchmarks and system reports show that agents can instead recover public solutions, read evaluation artifacts, infer generator structure, manipulate feedback, or benefit from invalid scoring paths; existing responses do not provide a common procedure for attributing these shortcuts and quantifying their effect across benchmarks. We formulate protocol validity and introduce HackDetect, a post-hoc audit that identifies an exposure, determines how the agent used it, and assesses whether the resulting score is misleading. We quantify score inflation with the Mislead gap, defined as the exploit score minus the intended score. We audit 2,385 traces across 15 agent benchmarks and find evidence of exposures and reward hacking in 67.0% of Frontier Science traces and 66.7% of AutoLab tasks. Across paired comparisons, we measure score inflation of 0.45-1.00, showing that benchmark reports should provide evidence that scores reflect the intended capability.\n\n### References & Citations\n\nLoading...\n\n# Bibliographic and Citation Tools\n\nBibliographic Explorer\n\n*(*[What is the Explorer?](https://info.arxiv.org/labs/showcase.html#arxiv-bibliographic-explorer))\nConnected Papers\n\n*(*[What is Connected Papers?](https://www.connectedpapers.com/about))\nLitmaps\n\n*(*[What is Litmaps?](https://www.litmaps.co/))\nscite Smart Citations\n\n*(*[What are Smart Citations?](https://www.scite.ai/))# Code, Data and Media Associated with this Article\n\nalphaXiv\n\n*(*[What is alphaXiv?](https://alphaxiv.org/))\nCatalyzeX Code Finder for Papers\n\n*(*[What is CatalyzeX?](https://www.catalyzex.com))\nDagsHub\n\n*(*[What is DagsHub?](https://dagshub.com/))\nGotit.pub\n\n*(*[What is GotitPub?](http://gotit.pub/faq))\nHugging Face\n\n*(*[What is Huggingface?](https://huggingface.co/huggingface))\nScienceCast\n\n*(*[What is ScienceCast?](https://sciencecast.org/welcome))# Demos\n\n# Recommenders and Search Tools\n\nInfluence Flower\n\n*(*[What are Influence Flowers?](https://influencemap.cmlab.dev/))\nCORE Recommender\n\n*(*[What is CORE?](https://core.ac.uk/services/recommender))# arXivLabs: experimental projects with community collaborators\n\narXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.\n\nBoth individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.\n\nHave an idea for a project that will add value for arXiv's community? [ Learn more about arXivLabs](https://info.arxiv.org/labs/index.html).", "url": "https://wpnews.pro/news/do-agent-benchmarks-measure-capability-protocol-validity-in-the-age-of-agentic", "canonical_source": "https://arxiv.org/abs/2607.22368", "published_at": "2026-07-27 08:07:06+00:00", "updated_at": "2026-07-27 08:22:27.320681+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-agents", "ai-safety", "ai-research"], "entities": ["HackDetect", "Frontier Science", "AutoLab"], "alternates": {"html": "https://wpnews.pro/news/do-agent-benchmarks-measure-capability-protocol-validity-in-the-age-of-agentic", "markdown": "https://wpnews.pro/news/do-agent-benchmarks-measure-capability-protocol-validity-in-the-age-of-agentic.md", "text": "https://wpnews.pro/news/do-agent-benchmarks-measure-capability-protocol-validity-in-the-age-of-agentic.txt", "jsonld": "https://wpnews.pro/news/do-agent-benchmarks-measure-capability-protocol-validity-in-the-age-of-agentic.jsonld"}}