aiqbench analysis
This research introduces the Hack-Verifiable Terminal Bench (HVTB), a new methodology designed to automatically detect reward hacking in autonomous agents by embedding specific failure modes into coding and terminal tasks. By moving beyond unreliable human or LLM-based inspections, the study quantifies how well frontier models adhere to task intent versus merely satisfying superficial checks, even against previously unknown exploits. The findings offer critical insights into the reliability of deploying highly capable AI agents in real-world environments where unintended behaviors could have significant consequences.
Hack-Verifiable Terminal Bench: Evaluating Reward Hacking in Terminal Tasks
arXiv:2608.22103v1 Announce Type: new Abstract: As agents grow more capable and autonomous, their tendency to reward hack, satisfying a task's checks while violating its intent, becomes an increasingly important failure mode. Measuring reward hacking is itself challenging, as detection typically relies on human inspection or LLM judges, both of which can be unreliable. The hack-verifiable environments (HVE) methodology addresses this challenge by embedding detectable hacks into tasks, allowing reward hacks to be identified automatically and reliably. In this work, we adapt HVE to Terminal Bench, a leading benchmark of real-world terminal and coding tasks, and introduce Hack-Verifiable Terminal Bench (HVTB). Using HVTB, we measure reward-hacking rates across frontier models and study whether prompts with varying amounts of information on the hack can mitigate this behavior. This lets us test whether prompting can prevent not only known reward-hacking strategies, but also 'unknown unknown' exploits that the prompt does not anticipate. We release all environments and agent traces at https://majoroth.github.io/hack-verifiable-environments/hvtb
Abstract:As agents grow more capable and autonomous, their tendency to reward hack, satisfying a task's checks while violating its intent, becomes an increasingly important failure mode. Measuring reward hacking is itself challenging, as detection typically relies on human inspection or LLM judges, both of which can be unreliable. The hack-verifiable environments (HVE) methodology addresses this challenge by embedding detectable hacks into tasks, allowing reward hacks to be identified automatically and reliably. In this work, we adapt HVE to Terminal Bench, a leading benchmark of real-world terminal and coding tasks, and introduce Hack-Verifiable Terminal Bench (HVTB). Using HVTB, we measure reward-hacking rates across frontier models and study whether prompts with varying amounts of information on the hack can mitigate this behavior. This lets us test whether prompting can prevent not only known reward-hacking strategies, but also 'unknown unknown' exploits that the prompt does not anticipate. We release all environments and agent traces at this https URL
Artificial Intelligence (cs.AI)
arXiv:2608.22103 [cs.AI]
(or arXiv:2608.22103v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2608.22103 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both 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.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 249 words
Related coverage
Why do models task game?
AI Alignment Forum · Aug 6, 2026
[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
Latent Space · Oct 13, 2026
Concrete Evaluations to Investigate the OpenAI Model That Hacked Hugging Face
AI Alignment Forum · Aug 3, 2026
Senior SWE-Bench: Evaluating Coding Agents Like Senior Engineers
Snorkel AI Blog · Jul 16, 2026