New Delhi: Four paying users of leading artificial intelligence services have sued Anthropic, OpenAI, SpaceXAI and Google, accusing the companies of unlawfully coordinating plans to slow the development of advanced AI systems. The proposed class action, filed in a federal court in California on Friday, argues that the alleged arrangement would weaken competition and reduce the value of subscriptions to ChatGPT, Claude, Grok and Gemini.
The lawsuit was brought in a US district court in California by Charles Buist, Nick Spetsas, Cheyenne Hunt and Christine Bullock on behalf of what they hope will become a nationwide class of paid subscribers. It alleges that the four companies violated Section 1 of the Sherman Act, which prohibits agreements that unreasonably restrain trade.
At the centre of the dispute is a September 12 essay by the chief executive of Anthropic, Dario Amodei, titled “We Must Pace the Frontier”, in which he argued that AI capabilities were advancing faster than researchers’ ability to make the systems safe. The chief executive of OpenAI, Sam Altman, the founder of SpaceXAI, Elon Musk, and the chair and co-founder of Google DeepMind, Demis Hassabis, publicly endorsed important elements of the proposal that day.
The plaintiffs contend that these responses amounted to more than independent expressions of concern, alleging that leading rivals had agreed to restrict the rate at which their products would improve. They say any collective restraint would leave subscribers paying the same prices for services whose development had deliberately been slowed.
The companies had not responded publicly to the allegations when the lawsuit was reported, and no court has ruled that an agreement existed or that any defendant violated competition law. The complaint represents the plaintiffs’ account and will have to survive legal challenges before it can proceed as a class action.
According to the lawsuit, the alleged coordination began before the public exchange on September 12. It points to a July 2026 statement signed by employees and executives associated with leading AI laboratories, which acknowledged that competitive pressures made it difficult for any company to slow development unilaterally and called for government-backed international action to manage the pace of progress.
The complaint also cites discussions about creating common safety standards and an industry body to evaluate the most capable AI models. It alleges that representatives of Anthropic, OpenAI and Google had been examining such arrangements for several weeks before Amodei published his proposal.
What Amodei proposed
As RNA Media had reported earlier, Amodei’s three-stage plan begins with placing independent evaluators inside frontier-model companies and giving them access comparable to that available to employees conducting internal risk assessments. These evaluators would examine models, training processes and safeguards, report serious incidents and publish findings without the companies exercising broad editorial control.
The second stage would require AI developers in democratic countries to establish common safety standards and limits on unchecked advances in model capabilities. The third envisages negotiations between governments, including the United States and China, on testing, dangerous applications and possible restrictions on recursive self-improvement – the use of AI to design increasingly capable successor systems.
Amodei said “pacing” did not mean ending research or stopping model training, but allowing safety work, interpretability research and independent testing to keep up with rapidly improving capabilities. He warned that increasingly autonomous systems could conduct cyberattacks, deceive evaluators or escape controlled computing environments if capability gains continued without adequate safeguards.
His essay explicitly recognized that cooperation between competitors could raise antitrust concerns. Amodei therefore proposed government mediation or a narrowly drawn waiver allowing companies to discuss specified safety measures without exposing themselves to legal action.
Altman said OpenAI supported consistent federal requirements for frontier-model safety and would provide independent evaluators with greater access. However, he argued that companies did not need to wait for an antitrust exemption or new legislation before improving transparency and safety practices.
Musk said Amodei was right to demand a more cautious approach, while Hassabis described the proposal’s direction as appropriate and linked it to Google DeepMind’s call for an industry-wide standards organization. Their public support forms a central part of the plaintiffs’ claim, though endorsements of a policy proposal do not by themselves establish that a binding commercial agreement was reached.
Safety cooperation or restriction of competition?
The case turns on a difficult distinction between legitimate technical cooperation and coordination that suppresses competition. US competition law generally permits companies to share narrowly defined threat information or develop voluntary technical standards, but rival businesses can face liability if they agree to limit output, innovation or product quality.
The US Federal Trade Commission says the Sherman Act prohibits contracts, combinations and conspiracies that unreasonably restrain trade, although not every arrangement between competitors is automatically unlawful. Private plaintiffs may seek damages and court orders under the Clayton Act if they can show that prohibited conduct caused them economic injury.
The subscribers argue that a slower rate of improvement is comparable to restricting output because access to each company’s latest and most capable models is an important part of a paid subscription. Their complaint seeks triple damages and an injunction preventing the companies from coordinating training limits, capability thresholds, release delays or other restrictions on competing products.
They do not seek to stop any company from independently delaying a model, strengthening safeguards, conducting safety research or complying with future government rules. Nor do they oppose appeals to Congress or the White House for regulation; their objection is to private competitors allegedly determining the pace of an important technology collectively.
Former US antitrust chief Jonathan Kanter has argued that AI companies do not require a special exemption to make their own systems safer or to share narrowly defined security information. However, he said an agreement under which competitors decided that they were developing products too aggressively and should collectively slow down could raise serious competition-law questions.
The controversy has also split the technology industry. Microsoft and several frontier-AI researchers have supported independent evaluation and common safeguards, whereas Meta, Nvidia and smaller AI developers have questioned whether dominant laboratories should be allowed to shape rules that could make it harder for challengers to compete.
As RNA Media had reported earlier, the US president, Donald Trump, has rejected calls to curb AI development, arguing that Washington must retain its advantage over Beijing. China has, in turn, criticized proposals linking a slowdown with tighter restrictions on advanced semiconductor exports, describing them as an attempt to preserve American technological dominance.
The lawsuit will therefore test more than whether several executives exchanged similar views in public. It could help define how far powerful AI companies may cooperate on potentially catastrophic risks without allowing safety arguments to become cover for restricting innovation, excluding smaller competitors or reducing the quality of products sold to consumers.