In a rare show of unity, OpenAI, Anthropic, and Google DeepMind are in active discussions to establish a self-regulatory standards body aimed at mitigating the catastrophic risks of advanced artificial intelligence (AI).
Representatives from the three leading laboratories have been convening in working groups since at least July, according to industry sources and company officials.
The talks center on creating a framework to evaluate and audit frontier AI models prior to their public release, operating as a public-private partnership like the Financial Industry Regulatory Authority (FINRA), which oversees Wall Street.
The impetus for the secret talks stems from a July proposal by Google DeepMind founder Demis Hassabis, who advocated for an independent watchdog staffed by technical experts. Momentum accelerated following a recent essay by Anthropic CEO Dario Amodei, who urged major labs to voluntarily slow the development of their most advanced systems. OpenAI CEO Sam Altman and Hassabis both publicly endorsed Amodei’s call to action.
OpenAI Chief Global Affairs Officer Chris Lehane confirmed the ongoing negotiations during a Tuesday press briefing in Washington, D.C., where he met with lawmakers regarding federal AI oversight. Lehane emphasized that the tech sector must act decisively on safety measures, with or without explicit federal backing.
“Any industry-led standards would complement — not replace — mandatory federal safeguards and democratic oversight,” Lehane said, adding that the shift in AI capabilities requires all immediate options to be on the table.
Despite the momentum, the initiative faces significant political, legal, and competitive headwinds.
The Trump administration has largely pushed back against binding AI regulations, expressing concern that slowing domestic progress could cede a technological advantage to China. A draft executive order attempting to establish a national AI regulator was recently shelved due to a lack of internal White House support.
House Speaker Mike Johnson, R-La., noted Sunday that tech leaders remain far apart on specific technical guardrails, emphasizing that any meaningful progress will require close collaboration between policy makers and corporate actors.
At the same time, industry experts and executives, including Altman, have acknowledged that private coordination on production timelines and standards could trigger antitrust scrutiny for cartel-like behavior unless granted explicit federal legal safe harbors.
Smaller competitors, meanwhile, have voiced fierce opposition. Cohere CEO Aidan Gomez publicly criticized the proposal, warning that allowing dominant players to set the rules risks locking out smaller startups. And Meta Platforms Inc. CEO Mark Zuckerberg reportedly expressed opposition to a centralized national regulator in calls with political leadership.
Greg Notch, chief technology officer at Expel, is skeptical that self-regulation, no matter how well intended, will extended pass U.S. boundaries.
“Another consideration here is that our adversaries likely won’t agree to these guardrails,” he said. “The gap between their frontier models (many of which are open weight) and ours is not that large, and even when it appears to be (see models like Fable and Astra), the gap is closed within months if not weeks. Ultimately, these restrictions are tying the hands of our own advancement.
While a universal consensus remains elusive, OpenAI announced support for provisions in the proposed federal FRONTIER Act, which would mandate that top labs grant independent third-party auditors access to monitor system development. OpenAI Executives confirmed that further details regarding voluntary industry standards will be released in the coming months.
