TLDR
OpenAI, Google and Anthropic are forming the Standards Authority for Frontier AI (SAFA), an industry-funded body modelled on financial regulator FINRA, to test models before release. The companies selling frontier AI would set the testing standards for their own models.
KEY TAKEAWAYS
Three labs, one rulebook
Google, OpenAI and Anthropic have tentatively named their proposed joint safety body the Standards Authority for Frontier AI (SAFA), with governance still being finalised and no formal launch date confirmed.[5] People familiar with the matter say CEO candidates include Sriram Krishnan and Arati Prabhakar, with Condoleezza Rice and David Friedberg among those considered for the board chair role.[5] None of those names has been confirmed by any of the three companies.
SAFA is described as industry-funded and modelled on FINRA, the Financial Industry Regulatory Authority, which sets conduct standards and examines member firms across US securities markets. The proposed mandate includes independent testing of models up to 30 days before release, alongside evolving technical benchmarks across security domains.[1]
How the push took shape
Google DeepMind chief executive Demis Hassabis put the FINRA-style concept on the table first, publishing a detailed framework on his Substack on 14 July 2026.[1] Anthropic chief executive Dario Amodei followed on 12 September, publishing an essay committing Anthropic to embed third-party safety evaluators with employee-level access.[3]
OpenAI Chief Global Affairs Officer Chris Lehane wrote on 9 September that "we will work with other frontier labs to advance frontier AI standards, building a voluntary effort now, with or without government support."[2] Sam Altman then took the argument to the United Nations Security Council on 23 September.
What Altman and Amodei said
Altman told the Security Council: "We need a mechanism for complementary national and international frontier AI standards: standards for measuring capabilities, assessing risks, determining whether safeguards are sufficient, and preserving meaningful human oversight as systems become more autonomous."[4] Altman's remarks also sketched out incident-reporting protocols and secure threat-sharing channels between governments and technical experts.
Amodei put the case for slowing down: "I have become convinced that fully addressing the risks requires even more prudence, not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up."[3]
The lawsuit in the background
A US class action alleges that the same three labs previously agreed to slow AI development, a claim all three dispute. The companies have not said how SAFA relates to those claims.
The Frontier Model Forum, co-founded in mid-2023 by OpenAI, Anthropic, Google and Microsoft, already exists but carries no enforcement powers and no pre-release testing mandate, the precise gap SAFA is designed to fill.[5] Meta joined the Forum in 2024, and whether Meta would be invited into SAFA or left outside it remains unresolved.
What it means for buyers of AI tools
For marketers and agencies buying frontier AI products, if SAFA proceeds as described, the three companies selling frontier AI tools would set the safety standards those tools are tested against. Critics have already flagged that a self-regulatory body of incumbents risks excluding open-source competitors and locking in the current leaders under the cover of safety standards.
Governance structure, funding contributions and the selection of leadership candidates remained unconfirmed by the companies as of 24 September 2026.[5]
SOURCES & CITATIONS
FREQUENTLY ASKED QUESTIONS
What is SAFA?
Who might run SAFA?
Does a body like this already exist?
Why does the class action matter here?
Xaviery Malinao writes for Prompt the Market on how brands and agencies are adapting to answer engines, drawing on Bushnote's work with clients across search, AI search and content.
