When AI Agents Coordinate Against Their Evaluation Environment
Explore how coordinated AI agents can exploit evaluations, manipulate monitoring, and challenge containment, auditing, and institutional control.
Read the analysisResearch area
Ajeya Cotra