The working group has been meeting regularly since July, at the level just below the three CEOs. According to people familiar with the talks, the discussions were already underway when Dario Amodei published his essay calling for coordination on September 12. Sam Altman told an employee meeting that he supports an industry testing and auditing body, but argued the major labs must build one themselves if the US government won't back it.
The structure being discussed is straightforward: a self-regulatory body where Anthropic, OpenAI, and Google write the testing and auditing standards that smaller labs would have to meet. Demis Hassabis endorsed the direction, pointing to Google DeepMind's own proposal for a frontier AI standards body modeled on FINRA, the financial industry's self-regulator.
What's at stake is who sets the compliance bar in a market the three participants already control.
The Antitrust Wall
Amodei's plan asks Washington for “a narrow antitrust waiver” so competitors can hold safety conversations without the conversations themselves becoming the legal problem. The government has not responded publicly.
The legal obstacle is not hypothetical. Section 1 of the Sherman Antitrust Act prohibits agreements that restrain trade. A coordinated pause in AI development—labs agreeing to halt certain model types until safety thresholds are met—could be construed as an output restriction, “one of the paradigmatic examples of restraints of trade the Sherman Act was intended to prohibit”.
OpenAI has already asked Congress for guidance on whether orchestrating an industry-wide slowdown would be legal. The bipartisan Collaboration on Adversarial Threats and Security Risks Act, introduced in July, would permit AI labs to coordinate on safety without antitrust risk. It was referred to the Judiciary Committee and has not been taken up.
The legal uncertainty itself is a deterrent. As the DOJ and FTC once recognized, “a perception that antitrust laws are skeptical about agreements among actual or potential competitors may deter the development of procompetitive collaborations”. Labs facing commercial pressure may conclude the legal and reputational costs of coordination outweigh the benefits.
Europe offers no alternative. The European Commission stopped granting individual antitrust exemptions in 2004. Its horizontal cooperation guidelines include a soft safe harbor for sustainability standards, but no equivalent for safety standards, and nobody has asked for one.
The Compliance Bill Smaller Labs Will Pay
A formalized standards body would cut compliance uncertainty for the three incumbents. It would also raise fixed costs for labs that cannot afford embedded evaluators, in-house red teams, and independent audits.
Altman's own framing acknowledges this. He supports a testing body, but says the major labs must build it because the government won't. The labs that build it also write the rules.
One analyst told Fortune that OpenAI's reversal on California's SB 53 frontier AI safety law—from opposition to calling for strengthening after the Hugging Face incident—might be genuine, but could also help OpenAI cement its lead by raising the bar for smaller rivals. The same logic applies to the standards body.
Aiden Gomez, CEO of enterprise AI startup Cohere, posted on X: “Great ideas from a cartel”. The concern is that the largest AI companies are coordinating rules that could squeeze smaller competitors.
David Sacks, the White House AI czar, has questioned whether the slowdown calls are “purely altruistic,” pointing to product liability risk as a financial incentive to prevent AI harms. He opposes government creating the body, but does not oppose companies coordinating safety efforts themselves.
The Talks Started Before the Essay
The timing of the working group reveals something the public letter obscured.
The discussions began before Amodei published “We Must Pace the Frontier”. Altman's support for an industry body was expressed at an employee meeting earlier that week, not in response to the essay. When Amodei asked the industry to coordinate, the coordination was already happening.
What Amodei's essay added was the public layer: a call for government to enable the talks with an antitrust waiver, and a unilateral commitment from Anthropic to embed METR evaluators with employee-like access. The working group added the private layer: three competitors negotiating the architecture of self-regulation.
Both layers face the same unresolved question. The labs want to coordinate. The law says they can't. The government says figure it out yourselves. And the smaller labs that would have to meet whatever standards emerge have no seat at the table.
P.S. The working group has been meeting since July, according to people familiar with the talks. That means the three largest AI companies in the US were negotiating a self-regulatory framework for at least two months before Amodei's essay made it public—and before the Hugging Face breach that he cited as the triggering incident.
Frequently Asked Questions
Q: What is the industry standards body being discussed?
A: Anthropic, OpenAI and Google have been negotiating a self-regulatory body since July where the three largest AI labs would write testing and auditing standards that smaller labs would have to meet. It is modeled in part on FINRA, the financial industry's self-regulator.
Q: Why is antitrust law a problem?
A: Section 1 of the Sherman Antitrust Act prohibits agreements that restrain trade. Competitors coordinating on safety standards—especially a coordinated pause in development—could be construed as an output restriction, which is illegal without a government exemption.
Q: What did Amodei ask for?
A: In his September 12 essay, Amodei asked Washington for “a narrow antitrust waiver” so competitors could hold safety conversations without legal risk. The government has not responded publicly.
Q: What legislation could address this?
A: The bipartisan Collaboration on Adversarial Threats and Security Risks Act, introduced in July, would permit AI labs to coordinate on safety without antitrust risk. It was referred to the Judiciary Committee and has not been taken up.
Q: Does Europe offer an alternative?
A: No. The European Commission stopped granting individual antitrust exemptions in 2004. Its horizontal cooperation guidelines include a soft safe harbor for sustainability standards, but no equivalent for AI safety standards.
Q: How does this affect smaller labs?
A: A formalized standards body would raise fixed compliance costs for labs that cannot afford embedded evaluators, in-house red teams, and independent audits. The three incumbents would write the rules that smaller competitors must meet.
Q: What has Cohere's CEO said?
A: Aiden Gomez posted on X: “Great ideas from a cartel”. The concern is that the largest AI companies are coordinating rules that could squeeze smaller competitors.
Q: When did the talks start?
A: The working group has been meeting regularly since July, at the level just below the three CEOs—at least two months before Amodei's essay made the effort public.
Q: What is the White House position?
A: David Sacks, the White House AI czar, opposes the government creating the body but does not oppose companies coordinating safety efforts themselves. He has questioned whether slowdown calls are “purely altruistic”.
Q: Why does the timing matter?
A: Amodei's essay presented the coordination as a new proposal. In fact, the talks were already underway. The essay added a public layer—a call for an antitrust waiver—to a private process that had been running since July.
