Anthropic and OpenAI sought control over outside safety reviews
Both companies reportedly wanted strict say over what independent evaluators could publish, citing concern that findings could expose trade secrets.
By Standard Topic. AI-generated from the reporting listed under Coverage. Published . Updated .
This story is no longer being updated. It reflects reporting up to .
What happened
Anthropic and OpenAI privately sought significant control over what outside evaluators could publish, according to two people familiar with the discussions. The reported concern was that findings could reveal trade secrets. The requests reportedly came as recently as October 5, despite September pledges from both companies' CEOs that evaluators would operate without company veto.
Why it matters
Researchers need access to unreleased systems, but company approval and redaction rules can limit what they tell the public.
Timeline
- Sep 2025: The CEOs of Anthropic and OpenAI publicly pledged that outside evaluators would operate without company veto over their findings.
- Oct 13, 6:07 AM ET: Reuters reported that the two companies' requests for control over outside evaluations had come to light.
Coverage
- An update on independent evaluators (Anthropic)
More from Technology & AI
- Eight states sue OpenAI, alleging misleading AI safety claims
- OpenAI announces solution to Navier-Stokes math problem
- Datadog bets AI agents can handle testing of AI-written code
- Anthropic and OpenAI sketch wealth-sharing plans before IPOs
- Mamdani administration proposes AI customer service rules
- Florida engineers study how AI behaves in high-risk situations