The three companies racing hardest to build the most powerful AI on earth just admitted they’re also quietly working together to make sure none of it blows up in anyone’s face.
OpenAI confirmed Monday that it has been meeting with Anthropic and Google in a working group aimed at setting shared safety standards for advanced AI models. Here’s what that actually means in plain terms: instead of each company grading its own homework on how safe its AI is before releasing it, the three labs are talking about outside testing, independent audits and a common checklist everyone would have to pass first.
According to The Information, representatives from the three companies have been meeting regularly since July, with a session as recently as last week. That’s well before the public got wind of any of it. The talks came out into the open only after Anthropic CEO Dario Amodei published an essay over the weekend urging the industry to slow down and coordinate more closely on testing, and after a former Anthropic researcher publicly resigned, saying the leading labs weren’t acting responsibly.
Google DeepMind chief Demis Hassabis had floated the basic idea back on July 14: a U.S.-based safety body modeled on FINRA, the group that oversees Wall Street brokers. Under that model, an industry-funded but independently staffed organization would run safety checks on the most powerful new AI models roughly 30 days before they’re released to the public.
OpenAI chief scientist Jakub Pachocki told reporters the company sees shared standards and international coordination as an immediate priority, and said OpenAI has already been talking to outside groups about what those standards should look like. OpenAI CEO Sam Altman has separately said he backs building a testing-and-auditing body, though people familiar with his thinking say he wants any such group kept independent of direct government control.
That’s where the real disagreement sits. Anthropic has pushed harder for government involvement, with the company’s safety lead calling in July for standards built jointly with federal regulators. Google and OpenAI are more cautious about handing that much authority to Washington, preferring an industry-run model that keeps regulators at arm’s length.
No one has agreed yet on what the rules would actually require. But if it happens, the effect would ripple well beyond the three labs. Businesses and developers who build products on top of ChatGPT, Claude or Gemini could eventually get a single, consistent safety rating to check before deploying a model, instead of guessing at each company’s internal review process.
There’s a business calculation behind the safety talk, too. A credible industry standard could get ahead of tougher government rules before they arrive, cutting the labs’ regulatory risk. Critics counter that the high cost of meeting shared standards could end up boxing out smaller AI startups that can’t afford the compliance overhead, leaving the field to the handful of companies big enough to absorb it.
It isn’t the industry’s first attempt at self-policing. OpenAI, Google, Anthropic and Meta already belong to the Frontier Model Forum, formed in 2023 to fund safety research, and all three companies have separately met with the White House this year on cybersecurity testing for advanced models. What’s different now is the direct working relationship between the three biggest labs on standards specifically, rather than research funding alone.
JBizNews Desk | San Francisco
© JBizNews.com All Rights Reserved. Reproduction or distribution without written permission is prohibited.
