OpenAI and Anthropic are now letting external evaluators test their advanced AI systems for dangerous capabilities and security holes.
AI safety org METR locked ~$71M in commitments over 6 months. Evaluator startup Vals AI raised $40M.
But here's the catch: Who's paying these watchdogs? How independent can they really be? And will they even get full access to inspect these systems?
The money's flowing, but the trust model is still broken. If the labs control the purse strings and the access, this is theaterโnot real oversight.
AI safety org METR locked ~$71M in commitments over 6 months. Evaluator startup Vals AI raised $40M.
But here's the catch: Who's paying these watchdogs? How independent can they really be? And will they even get full access to inspect these systems?
The money's flowing, but the trust model is still broken. If the labs control the purse strings and the access, this is theaterโnot real oversight.