Author

Michael Kouremetis

2 papers indexed here

We haven’t gathered this author’s papers yet. Follow them and we’ll fetch their work.

Not the right person? Other researchers publish under this name.

#artificial intelligence Preprint Sep 2026

ScopeBench: Do Agents Preserve Engagement Boundaries Under Goal Pressure?

ScopeBench is introduced, a benchmark of 30 dead-end agentic security tasks in which the stated objective is reachable only by violating the stated scope, and the frozen pilot benchmark, evaluation code, and all 2160 ATIF trajectories are released.

Shane Caldwell, Max Harley, Ads Dawson et al. · 1 citation

Review Jul 2026

Every Model Cheats: Prompt-Level Mitigation of Cheating on Offensive Cyber Tasks

The "solve rate"metric (clean passes only) is introduced to distinguish genuine capability from cheated outcomes, and it is argued it should be standard practice in any evaluation where cheating vectors are available.

Michael Kouremetis, Ads Dawson, Raja Sekhar Rao Dheekonda et al. · arXiv.org · 2 citations