OpenAI and Anthropic Expand Access for Independent AI Safety Evaluators as Apollo Presses for Binding Oversight

OpenAI and Anthropic are turning external AI safety reviews into a deeper, more continuous form of oversight, with both companies backing access for independent assessors across training, evaluation and deployment. The push follows high-profile agent-misalignment incidents and has prompted Apollo Research to argue that voluntary promises need enforceable access, clear publication rights and employee-equivalent visibility to be meaningful.