Embedded Evaluators

The concept of embedded AI evaluators, granting third parties access to frontier AI models for safety assessment, is gaining traction. While proponents aim to prevent AI from becoming uncontrollable, critics question the effectiveness of oversight without real power. Experts debate whether these evaluators, akin to banking supervisors, will have genuine influence or simply offer a veneer of safety, raising concerns about conflicts of interest and “audit washing” if companies retain ultimate control.