The three most heavily capitalised frontier-AI labs are quietly building an industry-run standards body modelled on Wall Street's Financial Industry Regulatory Authority. OpenAI, Anthropic and Google DeepMind have spent weeks negotiating an industry-funded, federally overseen entity that would review frontier models for up to 30 days before public release, according to reports first published by CNN Business on September 14 and expanded on this week by Bloomberg and Forkast News.
The Hassabis Blueprint
The idea traces back to a July 14 proposal from DeepMind CEO Demis Hassabis, who argued that pre-deployment evaluations of biosecurity, cyber, autonomy and CBRN risks are too expensive and too specialised to be duplicated inside each frontier lab. Under the working design, member companies would pay dues to fund a common evaluation team, submit models under NDA before launch, and defer to that body's findings on release timing — a structure explicitly borrowed from FINRA's oversight of US broker-dealers, and stitched together with a still-undefined role for the White House.
Cohere Calls It A Cartel
Cohere CEO Aidan Gomez blasted the effort in interviews this week, calling it "a cartel by any other name" and warning that the three biggest labs would use the body to entrench their own release velocity at everyone else's expense. Gomez's critique lines up with a Dreamforce keynote on September 15 in which Meta, xAI and NVIDIA also publicly opposed a heavy government hand, echoing an increasingly common view in Silicon Valley that formal regulation this cycle would freeze in place today's winners.
What The Insiders Are Saying
OpenAI head of global affairs Chris Lehane confirmed on September 15 that the frontier labs are coordinating on the effort and would push ahead "with or without government support." Anthropic CEO Dario Amodei's 3,800-word September 12 essay on "responsible pacing" leans the same way, arguing that frontier developers can and should build the guardrails themselves. On the other side, White House AI adviser David Sacks has publicly characterised the proposal as at best a distraction and at worst regulatory capture in disguise, a critique that lines up with prior Trump-era executive orders rewriting the federal AI rulebook.
The Fine Print Still To Come
None of the three labs has confirmed the FINRA parallel on the record, and no incorporation documents, charter or board slate have been filed. But the shape of the body is now clear enough that opposition is hardening in real time — from Gomez at Cohere, from the White House and from open-source advocates who see mandatory pre-release evaluations as an existential threat to Meta's Llama-class releases and to a growing wave of Chinese open-weight models like Z.ai's GLM-5.3. Whether Washington endorses the scheme or preempts it with statutory rules will define the shape of US AI governance for the balance of the decade.
Reporting based on coverage from CNN Business, Bloomberg, Forkast News, PYMNTS and Digital Journal.