Research Scientist, Manipulation Evals

Design and build novel AI manipulation evaluations to inform EU AI Act enforcement and regulatory action against frontier models.
- Develop risk scenarios and build end-to-end evaluations against frontier models using coding agents, analyzing results for regulatory use.
- Design rigorous evaluation methods for AI manipulation, deception, and persuasion; red-team findings and track state-of-the-art literature.
- Write technical results for EU AI Office and academic standards; co-author papers, benchmarks, and technical reports.
- Requires hands-on evals experience, at least one co-authored top-venue paper, and strong Python with eval frameworks like Inspect.
- Requires empirical AI safety background in evals, benchmarks, or adversarial robustness; clear technical writing for regulators essential.

Apart Research is a decentralised AI safety organisation that promotes technical AI safety by fostering a global community of researchers.