The Information reported on 24 September 2026 that Google, OpenAI and Anthropic, with Microsoft in support, are building a standards-setting body for frontier AI safety tentatively named SAFA, targeting late 2026 or early 2027. Follow-on reporting adds that the design is explicitly self-regulatory because an earlier public-private partnership under federal supervision collapsed after a White House draft executive order failed to gain backing. Demis Hassabis proposed the body; Sriram Krishnan, Arati Prabhakar and Condoleezza Rice are reportedly under consideration for leadership roles. Critics warn compliance costs could freeze out smaller, open-source developers.
Questions
As reported, SAFA would support third-party pre-deployment testing of frontier models, define incident-reporting practices, set voluntary safety commitments, establish qualifications for independent auditors, and potentially test models itself. No enforcement powers or binding authority have been announced — it is described as a standards and coordination body, not a regulator with penalties.
DrafterDaily Editorial
Keep reading
The AI Agent Accountability Act turns on two mental-state tests under the Computer Fraud and Abuse Act. The sources do not yet define the terms that decide who is exposed.
OpenAI withheld a model over scope and authorization failures, not raw capability. A UK AISI test of its predecessor shows what that failure looks like when measured, and what the numbers cannot tell us.
A creator says Meta's Muse agent accepted a lowball offer and gave a stranger his address. Meta says Muse followed instructions. Neither account has been independently verified, and the gap between them is a design question.