Slowing Down AI: Shared Caution or a Cartel?

Slowing down artificial intelligence for safety reasons may protect the public, but if the brake comes from an agreement between competitors, it risks becoming a restriction of competition. This is the dilemma that Agenda Digitale describes after the appeal from the leading American labs, which has reached the courts and the Senate.¹ A cartel means an agreement between companies that limits competition, investment and innovation. The legal question is whether coordinated caution can fall outside that definition.¹

What happened?

In an essay titled “We Must Pace the Frontier”, published on September 12, Dario Amodei, head of Anthropic, urged the industry to slow the speed at which model capabilities grow. A few hours later, Sam Altman (OpenAI), Elon Musk (SpaceXAI) and Demis Hassabis (Google DeepMind) publicly endorsed it.¹ According to Wired, cited by Agenda Digitale, two days earlier OpenAI had asked members of Congress whether a slowdown agreed with Anthropic and Google would conflict with antitrust rules. Amodei, the article writes, asked that the state act as mediator or supervisor.¹

Why does the antitrust request matter?

Consulting the legislature before cooperating with rivals does not prove that a cartel exists, because many forms of cooperation between competitors are legitimate. For Agenda Digitale, however, the move indicates that those making it do not believe it is advantageous to slow down alone, and that a joint action falls within what the Sherman Act, the American antitrust law of 1890, treats as a restraint of trade.¹ According to critics of the initiative, safety risks becoming the most convincing justification for a coordinated slowdown in the pace of innovation. Agenda Digitale specifies that this does not prove a cartel, but explains why many jurists are beginning to consider it a concrete possibility.¹

What does the class action argue?

On September 18, four paying customers of ChatGPT, Claude, Grok and Gemini filed a class action, Buist v. Anthropic, in the federal court for the Northern District of California. The defendants are Anthropic, OpenAI, SpaceXAI and Google.¹ The harm alleged is not a price increase. Subscribers claim that they receive, at the same cost, a service that progresses more slowly. They pay to use the most capable model, so if competitors agree to slow it down, the value obtained for each dollar falls.¹ The plaintiffs’ thesis also concerns investment. If each company knows the others will go slowly, it no longer fears being overtaken and can reduce commitment and spending without losing position.¹ The lawyer for the four, Nick Rowley, summarizes the position this way: antitrust law does not leave competitors free to decide among themselves that competition is an excessive danger.¹

What are the weak points of the lawsuit?

Agenda Digitale considers the evidentiary basis fragile. One essay and three public endorsements are indirect elements, and a subscriber obtains the service they paid for, without any right to future improvements.¹ The plaintiffs nonetheless rely on a statement signed in July by employees of several labs. It acknowledged that competitive pressure prevents each company from slowing down on its own. For the article, recognizing that solitary moderation is commercially irrational means explaining why an agreement would be needed.¹ The decisive question remains open. Converging public statements, calls for caution and questions to the legislature are not, in themselves, enough to establish an agreement contrary to competition. The plaintiffs regard them as clues, while the defendants could describe them as shared opinions about a common problem.¹

How did the Senate react?

During the hearing held on September 15 before the Senate Judiciary Committee, Texas Republican Ted Cruz, who chairs the Commerce Committee, dismissed as pure madness the image of executives who predict the end of the world and in exchange want to protect their own monopoly. Josh Hawley, Republican from Missouri, flatly rejected the idea of an exemption.¹ Among Democrats, Massachusetts Senator Elizabeth Warren assured that Congress will not grant the large labs any antitrust exemption. Hawaii Senator Mazie Hirono said she was uneasy about any exemption in a market with several competitors.¹ The same wind also swept away a much more limited exemption, concerning the exchange of information against Chinese espionage, which had been inserted into amendments to the annual defense law, the National Defense Authorization Act.¹

What changes

The crux, as Agenda Digitale describes it, lies in distinguishing convergent opinions from an actual agreement. On that line depends whether the request to slow down is read as prudence or as a restriction of competition.¹ The employees’ admission, according to which slowing down alone is not advantageous, can be read in two ways. The plaintiffs use it to show why an agreement would be needed, while the defendants could claim it as a sign of a common problem and not of a pact.¹

This article was written by artificial intelligence from the sources cited, without human review. Automatic newsroom by Redazia.

Sources

Footnotes

1. Rallentare l’intelligenza artificiale: sicurezza o cartello?, Agenda Digitale, 2026-11-12 ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸ ↩⁹ ↩¹⁰ ↩¹¹ ↩¹² ↩¹³ ↩¹⁴ ↩¹⁵ ↩¹⁶ ↩¹⁷ ↩¹⁸