AINEWS
OpenAI, Anthropic, and Google DeepMind want the U.S. government to bless their coordination with antitrust waivers. A smaller rival says the deal is a cartel by another name.
OpenAI, Anthropic, and Google DeepMind have spent weeks coordinating on AI safety, OpenAI's global head of policy Chris Lehane confirmed to Bloomberg. They are not asking for permission to talk. They are asking for the legal authority to act on what they agree on.
The ask landed in a Senate bill in July. Sens. Jim Banks (R-IN) and Adam Schiff (D-CA) introduced the Collaboration on Adversarial Threats and Security Risks Act, a bipartisan measure that would grant antitrust carveouts to AI labs coordinating on safety and national-security threats. The bill's national-security title frames it as a response to Chinese espionage. The load-bearing mechanism is the antitrust exemption: the legal cover for the three labs to set shared limits on what their models can do.
The trigger was a public letter, not a closed-door deal. Anthropic CEO Dario Amodei published an essay in The Atlantic in September warning of catastrophic outcomes from unchecked AI development and calling on the industry to slow down. He asked the U.S. government to "mediate or at least enable" the safety coordination, warning that without a framework the cooperation could trigger antitrust issues. OpenAI CEO Sam Altman publicly backed the essay, the rare cross-lab alignment that signals this is not one CEO's view.
The legal frame is the Sherman Act, the U.S. antitrust law that has policed price-fixing and output-restricting cartels since 1890. Industry-wide coordination on what AI can output risks running afoul of its "output restriction" prohibitions, the same doctrine used against price-fixing cartels. Legal experts quoted in the Gizmodo synthesis say an industry-wide AI pause could amount to coordinated output restriction, the kind of behavior the Sherman Act was written to stop. The carveout is the workaround: a statutory safe harbor that converts the coordination from a potential violation into licensed collaboration.
Cohere, one of the better-funded AI labs outside the top three, has accused the dominant labs of "using fear under the pretext of protecting the public" to "bend competition rules and be permitted to dictate the terms for everyone else." CEO Aidan Gomez called the arrangement "a wolf in sheep's clothing, a cartel by another name." His essay, published as part of the same Gizmodo reporting layer, reads as a competitor's challenge, not a neutral market analysis. The accusation lands because Cohere sits outside the coordination and has no seat at the rule-writing table.
Attorney General Todd Blanche told Bloomberg he would not prejudge whether the labs should receive an antitrust exemption, deferring to the President's policy direction with a "legal part of that." President Trump has long opposed increased AI regulation, which puts the carveout request in front of a White House that has no incentive to grant the labs more power. The Banks/Schiff bill is bipartisan on paper, but bipartisan bills stall when the administration does not want them.
Press News Agency's re-report of the Bloomberg scoop and Crypto Briefing's coverage of the safety pact talks both treat the coordination as a fait accompli. The actual story is the request behind it: the three labs are asking the U.S. government to let them write the rules for the most consequential technology being shipped, then enforce those rules on themselves and on the smaller labs that did not get a seat. The safety case is real, and so is the concentration risk. The bill, the carveout, and the cartel accusation are all symptoms of the same unresolved question. Who decides what "safe enough" means.