Tony Li, Hongliang Liu and Yuhao Wu
Vulnerabilities
HIGHAI Judges Exposed: Security Flaws Uncovered!
Unit 42's research reveals that AI judges can be tricked by simple formatting symbols. This vulnerability poses risks to security controls and decision-making processes. Developers are now working on patches to address these issues.