Why Are the Biggest AI Companies Calling for a Slowdown in Artificial Intelligence Development?

Four users have sued leading AI companies, claiming that behind calls for slower development lies more than just concern about safety.

On September 18, 2026, four U.S. subscribers to ChatGPT, Claude, Gemini and Grok filed a class-action lawsuit against the companies developing these systems, accusing them of agreeing to slow down the further development of their models and thereby violating U.S. antitrust laws.

The lawsuit is unusual because the users are not complaining about higher prices, poorer service or misleading advertising. They argue that the services they pay for would be improved more quickly if the companies continued competing without reaching agreements to limit development.

They are seeking damages as well as a court order banning agreements that would restrict the development and release of new models.

So far, however, there is no proof that such an agreement existed. The court will have to determine whether the companies actually coordinated their business plans and whether this resulted in a slowdown in the development of their products.

Why doesn't any company want to slow down first?

The lawsuit was prompted by an article by Anthropic CEO Dario Amodei, titled “We Must Pace the Frontier,” published on September 12.

Amodei called on leading AI developers to slow down the development of the most powerful systems so that there would be enough time to test them more thoroughly before they are widely deployed.

He proposed involving independent experts in safety assessments, establishing common restrictions for companies and, eventually, reaching an international agreement.

His proposal was supported by OpenAI CEO Sam Altman, Elon Musk and Google DeepMind chief Demis Hassabis.

Their public support, by itself, is not evidence of an agreement. However, the plaintiffs claim that representatives of Anthropic, OpenAI and Google had been participating in a joint group since July and had discussed slowing development before Amodei's public call.

The problem for any company that decides to slow down on its own is fairly simple: its competitors do not have to do the same.

While one company spends more time testing its new model, another could release a more capable product first and attract its users.

This is particularly important in an industry that requires enormous investments in chips, data centers, electricity and highly skilled professionals.

Safety or protection of market position?

The plaintiffs see another possible motive for the major AI companies here.

OpenAI, Anthropic, Google and other leading companies already have well-known products, large numbers of users and computing infrastructure that is extremely expensive to build.

If all major players agreed to a slower pace of development, the plaintiffs argue, existing companies could preserve their market advantage for longer.

New regulations and costly safety assessments could also place a significantly greater burden on smaller companies.

A large corporation can afford additional experts, independent testing and legal teams. For a young company, the same requirements could represent a cost that is difficult to bear.

The question, therefore, is not only whether AI development should be slowed down, but also who should set the rules and whether those rules would affect large companies and their smaller competitors equally.

What do the plaintiffs have to prove?

To succeed in the case, it is not enough to show that the CEOs of several companies publicly supported the same idea.

The plaintiffs must prove that the companies actually coordinated their business decisions. Internal correspondence, meeting records, witness testimony or other documents could serve as evidence of what was discussed or agreed.

Amodei himself pointed to the possibility of antitrust concerns, which is why he suggested that the U.S. government should mediate the discussions or that legislation should allow companies to cooperate in certain ways.

Even if the plaintiffs prove that an agreement existed, they would also have to demonstrate that they suffered harm as a result.

Subscribing to an existing AI service does not guarantee that a new version of a model will be released on a specific date or that it will provide particular improvements.

They would therefore have to show that, without the alleged agreement, the companies would have released new models earlier and that users would have received a better service for the same price.

Can competitors cooperate for safety reasons?

Not every form of cooperation between competitors is automatically illegal.

Companies in the same industry can jointly develop technical standards, share information about potential risks and try to solve problems that are difficult to address individually.

AI companies could therefore argue that the purpose of any cooperation is not to restrict competition, but to reduce the risks associated with insufficiently tested systems.

New AI models are increasingly capable of writing and executing computer code, using digital tools and performing complex tasks, making the question of how they should be tested increasingly important.

At the same time, the companies that know the technology best also have a business interest in protecting their products, users and market position.

That is why the dispute raises a question that goes beyond the lawsuit itself: who should set the boundaries for AI development, and under what conditions?

If common rules are introduced, the way they are designed could affect not only the safety of AI systems but also who will be able to develop the most powerful models in the future and how much room will remain in the market for new competitors.