OpenAI's New Priorities for Third-Party Assessments Are a Fine Start, but They Lack Teeth
Computerworld, Wednesday, September 23rd, 2026
Analysts say OpenAI's new principles for third-party model assessors sound good but include no enforcement mechanism.
OpenAI published a set of principles meant to govern how third-party assessors evaluate its models, which industry observers called conceptually sound but toothless in practice.
Experts point out OpenAI still controls the scope, access, reporting and redaction of any assessment, prompting one analyst to call it "a code of conduct for the assessors, not for OpenAI."
Some suggest OpenAI's real goal is pressuring competitors into matching restrictions rather than strengthening its own accountability, and that the framework's value will only be proven once tested against a real risk finding.