Built entirely on remarks made openly by executives, the complaint does not dispute that the companies' stated safety rationale is genuine. On 18 September, four consumers brought a class action suit in the Northern District of California against Anthropic, OpenAI, SpaceXAI and Google. The case, Buist v. Anthropic PBC, contends that the four firms entered into a horizontal arrangement to decelerate their competitive progress in developing AI systems, violating section 1 of the Sherman Act.

Bloomberg Law first reported the filing. The plaintiffs seek treble damages, an injunction and a jury trial.

The antitrust waiver forms the case's foundation

In his essay, Amodei argued that government involvement would be beneficial and that a limited exemption would be necessary for certain safety discussions. The lawsuit's response is direct: no such exemption currently exists, Congress has not created one, and no regulatory body has mandated this behaviour.

Altman stated two days later that OpenAI would not await an antitrust exemption or new legislation. Chris Lehane, OpenAI's head of policy, acknowledged the next day that the company had already been collaborating with Anthropic and Google DeepMind for several weeks.

The lawsuit characterizes this sequence as evidence of awareness of antitrust violations. The complaint treats inquiring about the legality of the conduct as proof the companies recognized it as coordination, and moving forward regardless as evidence of deliberate intent.

The absence of hidden communications is significant

The second section of the complaint states plainly: the alleged agreement was "proposed in public, accepted in public, and confirmed in public". The filing contains no internal correspondence, confidential informants or records of private meetings.

Every piece of evidence cited comes from a public declaration by a named company leader or media coverage. The essay the complaint treats as the initial proposal and the subsequent response treated as acceptance were both publicly reported.

This approach is atypical for cartel litigation and presents advantages and disadvantages. The evidence can be readily verified, yet the alleged conduct is something the companies made no effort to conceal.

The alleged terms of the agreement

According to the complaint, the companies agreed to maintain a slower pace of capability development than each would pursue independently. The alleged methods include caps on computational resources for training, limits on training iterations, restrictions on leveraging AI systems to enhance AI development, and staged capability assessments.

The complaint identifies the proposed evaluator framework as the enforcement mechanism, framed as measurable pacing. This mirrors the programme Anthropic has started rolling out in partnership with Accenture.

What the complaint does not claim

The filing does not contend that the companies fabricated their safety arguments. Rather, it argues structurally that any stated safety goal can be achieved through unilateral action, independent assessment, or legislative action.

The plaintiffs assert they regard AI risks as legitimate, and section eleven describes independent safety efforts and support for government involvement as the appropriate approach. They also exclude from their challenge unilateral restraint, independent evaluators, legitimate safety research, compliance with regulatory directives, or advocacy to Congress, including requests for an exemption.

The legal argument is more limited than it initially seems. The filing contends that courts cannot accept the argument that competition poses danger as a defence, since Congress holds that authority.

Apparent vulnerabilities in the filing

The complaint itself acknowledges that the complete impact on commercial releases has not yet become apparent. Across its twenty-nine pages, it identifies no delayed launch, discontinued product or modified training process.

The assertion of an 80% share of the US subscription market is made on information and belief and presented as a minimum rather than a definitive figure. The plaintiffs also characterize the restraint as naked, which would eliminate the need for market analysis, while also presenting market data as a backup argument.

The four alleged acceptances differ significantly. Musk's consisted of four words posted online, while Google's was a statement supporting the concept with specifics left unresolved.

Why this outcome was predictable

The antitrust risk was discussed openly long before the lawsuit was filed. A July statement from researchers across the organizations noted each experienced severe market pressure against independent restraint, which the complaint now cites as evidence of both motivation and conduct contrary to their interests.

This statement carries substantial weight in the argument. Acknowledging that independent restraint is economically irrational, from an antitrust perspective, explains why an agreement would be necessary.

The business environment is not peripheral to the case. The sector has seen AI labs discussing existential risks while simultaneously racing toward public offerings.

European regulators would face the same issue

Article 101 of the EU treaty bars agreements between firms that restrict competition, and it includes no safety exemption companies can invoke themselves. A coordinated pacing scheme among leading-edge developers would receive identical scrutiny in Brussels.

No investigation has begun in Europe, and none has been announced. The significance is that the exemption challenge Amodei raised is not unique to American antitrust law.

International tensions are pushing in the opposite direction. China characterized the slowdown proposal as a Cold War tactic, creating opposition to coordination from multiple sides.

What comes next

  • Monitor whether the working group reconvenes. The complaint states it was still active in mid-September; resuming meetings would be interpreted as continuing the alleged conduct.
  • Observe the defendants' responses. The allegations were filed the day before publication, no company has yet answered, and they possess substantial legal arguments that public statements and independent safety initiatives are lawful.

Source: The Next Web