The three labs’ safety body now has a working name: SAFA. It is not launched
The Information, via The Verge and ET: Standards Authority for Frontier AI, late 2026 or early 2027. Pre-deployment testing and incident reporting. Krishnan and Prabhakar have been approached. No charter, no incorporation, no chief.
Feature2 minIn-depth
Priya Shah2 min read
- SAFA
- OpenAI
- Anthropic
- Safety

Photo: The420.inSource page
The industry talks that began in July now have a provisional name. Google, OpenAI and Anthropic are discussing a body they would call the Standards Authority for Frontier AI, or SAFA, people familiar with the work told The Information. Stevie Bonifield had The Verge’s write-up on Thursday. ET EnterpriseAI and The420 carried the same reporting on Friday. The window they have been given is late 2026 or early 2027. This site already logged the July working groups. What is new is the name, the launch talk, and a short list of people the labs have approached. None of that is a launch.
Bonifield, citing The Information, said SAFA’s job would include “supporting third-party organizations that conduct testing of models before they’re deployed and laying out how AI developers should report safety and security incidents.” ET wrote that the group could turn the three companies’ public safety pledges into shared benchmarks, and that they are still deciding whether SAFA should run assessments itself or only write the rules for others. It would sit outside the US government. Industry people have told reporters that federal evaluators may not have the staff, compute or money to match frontier systems. That is a claim from the talks, not a finding.
The companies have approached Sriram Krishnan, a former White House AI adviser, and Arati Prabhakar, who ran the Office of Science and Technology Policy, about leading the organisation, ET reported, citing people familiar with the initiative. The420 was careful: no appointment, no incorporation, and the name can still change. An OpenAI spokesperson recently told CNBC the company was in discussions with Google and Anthropic on coordinated safety frameworks, ET noted. Those talks follow Demis Hassabis’s July essay for a US-backed, FINRA-style public-private body. SAFA, as described now, is more industry-led than that essay.
A body built by the firms it would judge will be asked, immediately, who pays, who sits on the board, whether outside researchers see model internals, and whether a lab must accept a finding it dislikes. Until there is a charter, treat SAFA as a reported plan. The three logos on a slide are not a regulator.


