After the Joint Commitment on Frontier Responsibilities signed with the White House, Fortune and The Verge reported in October 2026 that Google, OpenAI and Anthropic are accelerating plans for an industry-run AI safety standards body. The Information detailed that the project is taking shape under the tentative name Standards Authority for Frontier AI (SAFA), aiming to launch by late 2026 or early 2027.
The proposal follows a FINRA-style body effort that lost steam inside the administration. Instead, SAFA would proceed independently at first, with optimism for a public-private partnership later. It could work alongside the Frontier Model Forum, the earlier consortium of frontier labs.
Functions under discussion include supporting third-party pre-deployment testing, standardizing incident reporting and qualifying auditors. There is internal debate over whether the body should run evaluations itself, amid concerns about limited resources at public safety-evaluation entities.
Leo Schwartz and Steph Palazzolo of The Information noted that former White House AI adviser Sriram Krishnan was approached for a CEO role, and that names such as David Friedberg and Condoleezza Rice were considered for chair. Those details illustrate a search for political and technical legitimacy at once.
Outside the founding trio, The Information recorded widespread industry skepticism. Competitors and infrastructure vendors fear a three-lab club will set rules that are costly for them or that serve more as reputational cover than real control.
The timing is not accidental: it coincides with Trump’s “morally binding” pledge, with OpenAI’s wave of agent-activity notifications, and with the FTC probe opened in late September into lab safety practices. An industry body gives firms a “we are doing something” narrative without yet accepting hard federal mandates.
For European regulators, who already enforce general-purpose AI obligations from August 2026, a U.S. SAFA would be a private interlocutor, not a substitute for the Commission’s AI Office. Fragmentation of standards across Washington, Brussels and multilateral forums remains the baseline scenario.
If SAFA is limited to soft guidance and audits hired by the labs themselves, critics will say it replicates the self-policing problem. If it delivers public notification criteria, pre-release test thresholds and a comparable incident register across firms, it could shrink the information asymmetry that today leaves governments and customers in the dark.
The link to the Joint Commitment is direct: that document already called for regular meetings on shared standards. SAFA would be the institutional vehicle for that promise. Whether Meta, xAI, Nvidia or others join — or the club stays closed to the three labs driving it — remains open.
Enterprises that deploy third-party models should watch which audit and reporting requirements stick: they could become de facto purchase conditions in business contracts even without force of law.
Neural Pulse will track formal incorporation, leadership selection and the first published standards set. Until then, SAFA is a project in formation documented by The Information, not yet an operating regulator.
In October 2026, industry’s message is dual: sign White House pledges and, in parallel, build an authority of its own. The test will be whether that authority adds real friction to release calendars or only certifies what the labs already planned to do.
