Anthropic, OpenAI and Google have been quietly meeting since July on a joint AI standards body
The Information reports working-level teams from the three labs have met regularly for two months on a self-regulatory testing body -- talks that predate, and go further than, Dario Amodei's public "pace the frontier" essay.
Working groups from Anthropic, OpenAI and Google -- made up of executives below CEO level -- have been meeting regularly since July to discuss forming a joint AI standards body, The Information's Leo Schwartz reported September 13, with the group convening as recently as the week before publication. The proposal centers on shared protocols for testing frontier models before public release: independent evaluations, pre-release safety reviews and standardized risk assessments the three labs would all submit to.
Older, and less public, than it looks
The timing matters here: these talks predate Dario Amodei's September 12 "We Must Pace the Frontier" essay, which Altman and Musk publicly endorsed within hours and which this site covered separately. The catalyst was a July essay from Google DeepMind's Demis Hassabis proposing a US-led standards body modeled on FINRA, the securities industry's self-regulatory organization. So the public pledge that made headlines last weekend sits on top of two months of private coordination the companies hadn't previously disclosed -- Amodei's essay was the visible tier of a plan already partly built.
Where the three labs actually disagree
The reporting is clear the companies aren't aligned on what this body should be. Sam Altman said at an internal town hall he backs a testing-and-auditing organization but thinks the major labs will have to build it themselves, without US government backing. Anthropic has pushed for closer coordination with government rather than a purely industry-run body. Google has been the one floating the FINRA comparison -- a body with actual regulatory teeth, not just voluntary norms. None of that is resolved yet.
It's also not the industry's first attempt at this: the 2023 Frontier Model Forum and the Linux Foundation's Agentic AI Foundation, launched in December 2025, both tried versions of cross-company AI safety coordination. Whether a body designed and funded by the three biggest labs ends up setting a genuinely independent bar -- or one that's easier for incumbents to clear than for smaller competitors trying to catch up -- is the open question a self-regulatory body run by the companies it's supposed to constrain doesn't answer on its own.