OpenAI published new priorities and principles for independent assessments of advanced AI systems.

The framework covers third-party reviews of safety claims, safeguards, dangerous-capability evaluations and critical misalignment incidents. OpenAI proposes giving assessors proportionate access across model training, evaluation and deployment.