Paying for an AI assistant means weighing what it can do today against the improvements you expect next. A new proposed class action puts that second part of the purchase under scrutiny: subscribers allege that Anthropic, OpenAI, SpaceXAI and Google unlawfully coordinated to slow AI development, reducing the value of their subscriptions. AP reported the case on September 19.
The dispute concerns how quickly future models become more capable. A chatbot taking longer to answer your latest prompt would not, by itself, establish the conduct alleged here.
The public docket reviewed for this article shows a newly filed case, with no class-certification decision, settlement or refund award. The accusations remain unproven. The available docket snapshot was last retrieved September 18, 2026.
What subscribers claim they lost
In their September 18 complaint, the plaintiffs argue that coordinated limits on improvements make subscriptions worth less at the same price. They challenge an alleged agreement between competitors while expressly allowing companies to make independent safety decisions.
Think about the buying decision. You might keep an AI subscription because it already helps with your work. You might also renew because you expect better research, fewer errors or more reliable coding. Those expectations have value to you, even when they never appear as separate items on your receipt.
That makes the economic question harder than checking whether a monthly bill increased. Consider a hypothetical software service that costs $20 in both January and June. If its capabilities improve substantially, June’s service may be a better deal. If a promised feature never arrives, the customer may value it less. Neither observation, on its own, tells us why the product changed or establishes an antitrust violation.
For this case, our view is that the difficult work will be separating any effect of competitor coordination from ordinary development setbacks, safety testing and independent business choices. A disappointing model release cannot answer that question on its own.
Who brought the case, and who it could cover
Buist et al. v. Anthropic PBC et al. was filed in the Northern District of California by Charles Buist, Cheyenne Hunt, Christine Bullock and Nick Spetsas. The defendants are Anthropic PBC, OpenAI OpCo LLC, SpaceXAI LLC and Google LLC. The public docket lists case 5:26-cv-10693 and assignment to Magistrate Judge Nathanael M. Cousins.
The proposed class covers people in the United States who bought paid individual consumer subscriptions directly from the defendants for premium ChatGPT, Claude, Grok or Gemini access, beginning September 12, 2026 and continuing until the alleged effects cease. That wording does not automatically include every global user, business account or API customer. Purchases through intermediaries require particular care because the definition specifies direct purchases.
“Proposed” matters. Under Federal Rule of Civil Procedure 23, a court must decide whether the requirements for class treatment are satisfied. These include common legal or factual questions, suitable representatives and, for the damages class sought here, whether common questions predominate. Filing a complaint on behalf of subscribers does not complete that process.
Why the AI companies are talking about slowing down
The policy argument deserves to be read on its own terms. In “We Must Pace the Frontier,” Dario Amodei argues that capability gains are outpacing the industry’s ability to understand and safeguard advanced AI. He proposes three steps: embedded outside evaluators, coordination among labs in democratic countries and eventual coordination between governments internationally.
Anthropic’s immediate unilateral commitment in that essay is to give independent evaluators ongoing access comparable to employees. Amodei also acknowledges legal obstacles to some cross-company coordination and suggests government mediation or a narrow antitrust waiver. His stated aim is to give safety work time to catch up while development continues.
OpenAI had already described its own slowdown in an August 18 statement. It said it temporarily reduced the pace of scaling, including a two-week pause in reinforcement-learning training for its latest models intended for deployment, while strengthening its research environments and monitoring. At that time, its largest planned frontier reinforcement-learning run remained on hold.
That statement supplies a concrete account of OpenAI’s stated safety rationale. It does not establish that rivals agreed to restrain competition. It also predates this lawsuit, so treating it as a response to the plaintiffs would misrepresent the chronology.
The plaintiffs cite September 12 public endorsements by Sam Altman, Elon Musk and Demis Hassabis as evidence of agreement. Whether those statements establish an unlawful arrangement has not been decided by the court. Complaint, introduction.
CBS/AP reported September 19 that all four companies had not immediately responded to its comment request.
Kingy.ai has separately examined possible explanations for the industry’s slowdown push. Those competing interpretations are context; none substitutes for evidence about what the defendants agreed to do.
What the plaintiffs want, and what “triple damages” means
The complaint seeks class certification, treble damages, preliminary and permanent injunctions against the alleged coordination, interest, and legal fees and costs.
The damages request comes from Section 4 of the Clayton Act, which provides for three times the damages sustained by qualifying plaintiffs injured by an antitrust violation, plus costs and reasonable attorney’s fees. It does not mean that every subscriber is now entitled to three times every subscription payment.
For a purely illustrative calculation, if legally recoverable harm were established at $5, tripling that harm would produce $15. The exercise starts with proven damages, not an assumption that the entire purchase price was lost. Neither number is an estimate of a payment in this lawsuit, and this article identifies no court-approved payout amount.
An injunction is a different kind of remedy: a court order addressing conduct. Section 16 of the Clayton Act permits private plaintiffs to seek such relief against threatened antitrust loss or damage, subject to the governing requirements. Asking for that order and obtaining it are separate events.
The evidence that will matter next
Section 1 of the Sherman Act addresses agreements that restrain trade. The existence, terms and legal treatment of an agreement therefore matter. Similar public statements do not let readers skip those questions, and a company’s stated safety motive does not resolve them either.
Our assessment is that the most useful future reporting will examine specific decisions. Did a release schedule change? Who made that decision? Was it tied to another company’s commitment? What could the product reasonably have delivered otherwise? These questions are more informative than treating every announcement of caution as proof of collusion.
The defendants can respond through answers and potentially motions challenging the complaint. Rule 12 allows specified defenses to be raised by motion, including failure to state a claim. A decision allowing allegations to proceed would not itself prove them; it would determine whether the case can continue past that challenge.
If the case advances into discovery, the parties can seek relevant, nonprivileged evidence within the limits of Rule 26. In a dispute about coordination, communications and contemporaneous business records could be more revealing than public slogans. Whether particular documents exist, can be obtained or support either side is still unknown.
Class certification is another decision to watch. A later class settlement would also need court approval under Rule 23, with applicable notice procedures. None of those outcomes follows automatically from the September 18 filing.
What paying subscribers should take from this
For a renewal decision, judge the service you can use now. Does it complete the work you need? How much checking does it require? Do the available features justify the price for your usage? Treat hoped-for future capabilities as uncertain when deciding what a subscription is worth to you.
Keep ordinary billing records if you want to follow the case, and check any future eligibility or payment announcement against the court record. A headline about requested damages is insufficient evidence that money is available.
The next useful developments are the defendants’ formal responses, substantive court orders and any ruling on class treatment. Until then, the accurate description is a proposed subscriber lawsuit over allegedly restricted AI improvement, with liability and compensation unresolved.
Reporting checked September 20, 2026 (Pacific time). This article draws on the filed complaint, the available September 18 public docket snapshot, company policy statements and cited legal provisions. Kingy.ai has not independently established the alleged agreement or subscriber losses. Featured artwork is an original editorial illustration, not a court document.