OpenAI Sets Rules for Outside Safety Audits

OpenAI published a document on Tuesday naming four areas for independent safety assessors: safety cases, the safeguard stack, capability evaluations under its Preparedness Framework, and misalignment incident investigations. It sets seven principles covering pre-registered claims, grey box access, conflict disclosure and redaction rights. OpenAI says it is in talks with multiple third parties but names none, and funding goes unmentioned.