Anthropic and OpenAI seek independent evaluators for AI safety
Big names like Anthropic and OpenAI are now seeking to work with independent evaluators (think groups like METR and Apollo Research) on safety assessments of their advanced AI models.
Anthropic's CEO Dario Amodei says this extra layer of review is important to prove out the concept of embedded external reviewers, with support from OpenAI CEO Sam Altman and even a nod from President Donald Trump.
OpenAI firings, evaluator funding, rising oversight
Bringing in outsiders has its bumps: OpenAI fired three employees for "violating our policies on accessing and handling sensitive company information," and two of the employees said they believed the dismissals were linked to how they communicated with third-party evaluators, showing how tricky balancing transparency and privacy can be.
Meanwhile, evaluator groups are getting serious funding (METR announced in August that it had raised commitments of around $71 million over the last six months), and the FRONTIER Act bill plus California's fresh auditing rules mean more eyes on AI than ever before.