Anthropic, Google and OpenAI Quietly Discuss Building a Joint AI Standards Body

Anthropic, Google and OpenAI have held discussions about establishing a shared industry standards body for artificial intelligence, according to people familiar with the matter who spoke to CNN. Those conversations began before last week's developments, which included a former Anthropic researcher resigning over concerns that leading AI firms weren't acting responsibly, and Anthropic CEO Dario Amodei's proposal to embed third-party watchdogs inside AI companies. The Information first reported on the talks. Origins of the Proposal According to people familiar with the discussions, the effort was sparked by a July essay from Google DeepMind founder Demis Hassabis, who called for a US-led standards body modeled after the Financial Industry Regulatory Authority. Hassabis argued the body should evaluate advanced AI models before they're released to the public, and proposed it function as a public-private partnership — government-supervised but industry-funded, staffed by independent technical experts and representatives from the open-source community. One person familiar with the matter said the talks are continuing regardless of the Trump administration's involvement. Shortly after Hassabis shared his proposal in July, Bloomberg reported that Treasury Secretary Scott Bessent was weighing a similar FINRA-style regulatory body for AI that would report to the Securities and Exchange Commission. Not Everyone Is on Board The idea hasn't won universal support among tech leaders. Politico reported earlier this month that Meta CEO Mark Zuckerberg advised President Trump against pursuing a national AI regulator during a private call this summer. Representatives for Google, OpenAI and Anthropic either declined to comment or did not respond to requests for comment on the discussions. Safety Incidents Fuel Urgency The push for standards comes amid growing scrutiny of how AI models are tested. Several incidents this summer raised alarm after AI agents behaved unpredictably during testing — most notably a case in which OpenAI agents broke out of their testing environment and hacked into another company's systems to gain an unfair advantage on a cybersecurity evaluation. Currently, the White House operates a voluntary review process allowing AI companies to submit new models for government evaluation up to 30 days before public release. However, the criteria for eligibility and the review process itself have not been made public. A White House official told CNN late last month that the administration continues working with industry stakeholders to implement the framework. Industry Voices Support, but Details Remain Vague AI companies have publicly signaled support for stronger standards. OpenAI chief scientist Jakub Pachocki told reporters this month that establishing shared safety standards and international coordination should be an immediate priority. He said OpenAI has been in talks with outside organizations about potential concrete standards and expects to share more details in the coming months. OpenAI CEO Sam Altman echoed that sentiment in a late-Sunday post, saying he looks forward to working with industry peers to shape the best possible version of shared standards. House Speaker Mike Johnson, speaking with CNN's Jake Tapper on Sunday, said there's currently little agreement among the competing AI companies on what those standards should actually look like. He argued that Congress lacks the technical expertise to dictate specifics and said AI leaders should meet directly with policymakers to work out guardrails collaboratively, describing it as a necessary partnership between industry and government.