AI's quiet safety gatekeepers step into the spotlight

With Washington declining to regulate AI, small nonprofit evaluators like METR, Apollo Research and Transluce are becoming key arbiters of model safety. Anthropic and OpenAI have endorsed embedding external auditors, but funding, access and reporting structures remain unresolved; METR raised about $71 million in six months versus $13.6 million in all of 2024.

• METR raised about $71M in six months, up from $13.6M in all of 2024
• OpenAI fired three employees over contacts with third-party evaluators
• Vals AI grew from 8 to 30 staff and raised $40 million
• Trump proposed voluntary use of independent external auditors