OpenAI offers outside assessors deeper access

OpenAI has published "Priorities and principles for effective third party assessments," setting out how it intends to work with independent technical safety assessors in the private and non-profit sector. The company says it is committed to access deep enough to let assessors "challenge our assumptions, identify risks we may have missed, and reach their own conclusions about the effectiveness of our safeguards."

The paper describes access already granted in past engagements: information about technical safeguards, visible chain-of-thought access, and what OpenAI calls unprecedented levels of confidential data and internal deployment access for incident response and monitor red teaming. Third party assessments are also folded into its Preparedness Framework practices.

Two definitions anchor the document. A safety claim is a specific, evidence-testable assertion about a model's capabilities, behavior or safeguards; a safety case is a structured argument linking claims to evidence and making assumptions and residual risks explicit.

OpenAI proposes four priority areas. The two detailed in the material are independent assessment of safety cases spanning training, evaluation and internal and external deployment — including whether training methods reward deception, reward hacking or circumventing restrictions — and assessment of critical safeguards, such as whether "grey box" access shows safeguards holding up under adversarial testing.

The work is described as longer-term and launch-agnostic rather than pre-deployment gating, with multiple assessments running in parallel over weeks to months. The post names no assessor organizations, no timelines, and no binding commitments, and OpenAI frames its separate work with governments as a distinct track.