Google, OpenAI and Anthropic are drawing up plans for an industry-led body to set safety standards for the most powerful AI models, the tech news site The Information reported.
The group, tentatively called the Standards Authority for Frontier AI (SAFA), could launch by the end of this year or early in 2027.
It is modelled on the Financial Industry Regulatory Authority, the industry-funded body that polices US stockbrokers under the watch of the Securities and Exchange Commission.
Candidates approached for senior roles include former White House AI adviser Sriram Krishnan, former secretary of state Condoleezza Rice and investor David Friedberg.
A draft White House executive order would have brought federal supervision, but it stalled, so the labs pressed ahead alone.
SAFA's possible functions include supporting independent testing of models and setting standards for how developers report safety and security incidents.
The companies still disagree over the body's role, including whether it should test models itself.
That job currently falls to the Center for AI Standards and Innovation, a Commerce Department unit that some in the group say lacks the resources for thorough testing.
The plan has drawn accusations of regulatory capture, where the companies being regulated end up writing the rules.
Critics argue that standards set by the three leading labs could raise barriers for newcomers trying to build cutting-edge models.
Nvidia chief executive Jensen Huang and Meta's Mark Zuckerberg have publicly opposed the approach.
It is unclear whether other developers will be allowed to join as they catch up.
