Three AI Labs Form Independent Watchdog to Audit Three AI Labs
The body will be fully independent of government oversight, which organizers listed as its first accomplishment.
MOUNTAIN VIEW, Calif. — Three of the world's leading AI labs announced plans this week for an independent standards body that will hold frontier AI developers to rigorous safety requirements, with membership open to any frontier developer that is one of the three founding labs.
The organization, modeled loosely on the financial industry's self-regulatory authority, will operate entirely outside government oversight, which a draft charter describes as "essential to its independence from anyone who might disagree with it."
"This is a historic moment in which the industry has decided to hold itself accountable, on a voluntary basis, when convenient," said a spokesperson for the working group. "We looked at every possible regulator in the world and concluded that the most qualified people to watch us are us."
Under the proposal, labs would submit models for pre-release testing on a voluntary basis. According to the draft, a lab that declines to submit a model will be issued a stern memo, which it may also decline to receive.
Governance experts said the structure was not unprecedented. "It is similar to how a restaurant inspects its own kitchen, if three restaurants inspected each other's kitchens, shared a supplier, and also owned the fire department," said Professor Ludmila Adeyemi-Strand, who studies industry self-regulation.
A poll of 1,200 AI researchers by the policy group Gradient Civic found that 61 percent supported the idea of an independent safety body, and 44 percent had already received a job offer from it.
The board's first meeting has been scheduled for late 2026 or 2027, depending, a spokesperson said, "on whether anybody ships something first."
At press time, the watchdog had completed its first audit, of itself, and found no issues.