Adversarial Evaluation of Large Language Models

Conducted adversarial evaluations of advanced large language models (LLMs), assessing manipulation, persuasion, trust & safety, and behavioral risk scenarios. Developed realistic evaluation scenarios and structured testing frameworks to identify model vulnerabilities, failure modes, and harmful influence patterns across multiple prompting conditions. Applied rigorous annotation, calibration, and peer-review methodologies to support large-scale AI safety evaluations and improve the reliability of model risk assessments.