Research

Stress Testing AI Alignment: Why Models That Seem Safe Might Just Be Good at Taking Tests

A highly capable AI system might secretly pursue misaligned goals—a phenomenon referred to as “scheming”. Since a scheming AI would …