Technology & AI

OpenAI starts first phase of outside safety assessments, critics doubt independence

The program gives four outside groups, including Apollo Research and METR, extended access to OpenAI's internal systems.

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This story is no longer being updated.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

OpenAI, the AI company led by CEO Sam Altman, has signed four outside groups to test its models from the inside. The first phase of these embedded assessments has now started, with Apollo Research able to flag access gaps in a four-month pilot.

Why it matters

Researchers worry the program will not be enough, and critics say OpenAI controls who assesses its systems, what they can access and who they can speak to.

Timeline

  1. Nov 17, 3:55 AM ET: OpenAI signed four independent groups to test its models from the inside, giving outside experts access to its internal systems.
  2. Nov 17, 3:07 PM ET: Critics said OpenAI controls who assesses it, what they can access and who they are allowed to speak to.
  3. Nov 18, 4:30 AM ET: OpenAI began the first phase of embedded assessments with Apollo Research, METR, Transluce and Goodfire.
  4. Nov 18, 5:00 AM ET: Altman said outside researchers would get employee-level access, but the pilot program appears much more restrictive.
  5. Nov 18, 1:05 PM ET: Apollo Research can flag access gaps in the four-month pilot and explain what redactions mean for public reports.

Coverage

More on this story

More from Technology & AI

All Technology & AI news · Technology & AI story archive