Abstract:
Anthropic, OpenAI, Google and SpaceXAI have recently been subject to antitrust lawsuits filed by the U.S. Federal Court.
The plaintiffs accuse the four artificial intelligence companies of reaching an illegal agreement to slow down the development of artificial intelligence technology through public statements and coordinated actions, thereby restricting industry competition and reducing the value that consumers receive from purchasing AI services such as ChatGPT, Claude, Grok and Gemini.
The lawsuit was submitted to the U.S. District Court for the Northern District of California on September 18. The case was filed by four paying AI service users, who subscribed to AI products provided by the four companies mentioned above, and hoped that the court would recognize the case as a class action lawsuit on behalf of related paying users across the United States.
The plaintiffs did not claim that these companies’ individual decisions to slow down their own AI research and development were illegal per se, but focused the lawsuit on whether competitors could coordinate their actions and jointly decide how quickly the entire industry should advance technology development. The plaintiffs believe that if a company alone chooses to develop AI more carefully, it may bear the business risk of losing customers, revenue, talent, and technological leadership; but if major competitors jointly slow down development, it can limit the speed of product improvement while avoiding the above competitive pressures.
The incident focused on in this lawsuit occurred on September 12. On the same day, Anthropic CEO Dario Amodei published an article on AI security, calling on the entire industry to coordinate actions to control the development speed of cutting-edge artificial intelligence in order to buy more time for security measures and related protection technologies.
Amodei pointed out in the article that the development speed of artificial intelligence has been very fast, and the industry needs to reconsider whether it should continue to advance at the current speed. He also warned that if AI agents continue to develop rapidly, a large number of autonomously operating AI systems may appear in the future, and they may even gain the ability to control a large number of Internet resources in a short period of time.
The plan proposed by Amodei does not call for a halt to artificial intelligence research and development, but hopes that the industry can jointly develop a more cautious way of advancing. He believes that if all companies are worried about competitors accelerating, then any company slowing down alone may be at a disadvantage in business competition, so this "prisoner's dilemma" needs to be solved through industry-level coordination.
On the same day that Amodei published his article, OpenAI CEO Sam Altman, SpaceXAI founder Elon Musk, and Google DeepMind co-founder and head Demis Hassabis responded publicly and expressed support for related directions.
The plaintiff believes that this series of public statements are not independent security commitments, but evidence that competitors coordinate the speed of AI development. They pointed out in the complaint that if several major AI companies jointly decide to update their products slower than the level in a normal market competition environment, it may constitute coordinated behavior to restrict output and thus trigger U.S. antitrust laws.
Plaintiff lawyer Nick Rowley said that they are not opposed to AI companies slowing down research and development on their own for safety reasons. What is really questioned is that competitors jointly restrict technology development through private agreements. In the plaintiff's view, this approach turns the responsibility for safety decisions that should have been independently borne by each company into a joint industry restriction of competition.
The complaint further states that the relevant coordination may not have begun on September 12. The plaintiffs date back to July this year, when senior employees from several major AI labs signed a statement acknowledging the industry’s “intense competitive pressures that are unwilling to unilaterally slow down development” and calling on governments to support global efforts to slow down the development of automated AI.
Based on this, the plaintiff believes that the series of public statements on September 12 were only the result of the gradual development of previous discussions, rather than an isolated incident. However, these accusations are currently claimed by the plaintiffs in the lawsuit, and the court has not yet made a ruling on whether the relevant facts constitute a violation of law.
Another important point of contention in this lawsuit is the boundary between AI security cooperation and antitrust law. Amodei was already aware of possible antitrust issues when he first proposed the industry coordination plan. He believes that if different AI laboratories need to collaborate across enterprises on security issues, the US government can participate in coordination, or at least provide legal conveniences, such as limited antitrust exemptions for specific types of security discussions.
Altman previously stated that OpenAI welcomes the establishment of a unified AI security framework developed by the federal government and believes that the industry does not have to wait for special antitrust exemptions or new laws to begin relevant security work.
The plaintiff does not agree with this. They argue that the importance of AI safety issues does not automatically provide antitrust immunity for commercial coordination among competitors. According to Section 1 of the U.S. Sherman Antitrust Act, agreements between competing companies that restrict trade or competition may be legally binding, and the plaintiffs believe that jointly determining the speed of AI product improvement may fall into this category.
The lawsuit particularly emphasizes that an important expectation of consumers purchasing AI subscription services is that model capabilities, functions and performance can continue to improve as competition continues. If the major AI companies collectively slow down development, consumers may still pay the same subscription fee but get a product that is updated more slowly.
Currently, the four plaintiffs hope that the court will prevent the relevant companies from continuing to implement the alleged coordinated behaviors, and request the court to confirm that the relevant behaviors violate U.S. federal antitrust laws. They also request that the case be expanded into a class action lawsuit covering paying users across the United States.
If the case ultimately qualifies as a class action lawsuit, the number of consumers potentially involved could be very large. ChatGPT, Claude, Grok, and Gemini all currently have large paying user bases, so the legal impact of the case may extend far beyond the four consumers who originally filed the lawsuit.
Anthropic, OpenAI, Google and SpaceXAI did not immediately respond to media requests for comment as of the time of the publication of the relevant reports. Therefore, it is not yet known whether the four companies will deny the existence of any form of coordination, or whether they will argue that the relevant discussions belong entirely to legitimate cooperation in the field of AI security.
The lawsuit also comes against the backdrop of ongoing debate in the United States over the direction of artificial intelligence regulation. On the one hand, AI companies and researchers are increasingly concerned about the security risks that cutting-edge models may bring. On the other hand, there are also different opinions within the US government on whether AI development should be restricted. As the capabilities of AI models continue to improve at an accelerating rate, how to draw the line between security cooperation and maintaining market competition is becoming an increasingly realistic legal issue.
If the court ultimately accepts the plaintiff’s antitrust theory regarding the “AI deceleration agreement,” the case may have an impact on future security cooperation between AI companies. On the contrary, if the court believes that enterprise coordination against security risks does not constitute anti-competitive behavior, then this may also provide greater legal space for the AI industry to establish a cross-enterprise security cooperation mechanism.
The case is still in its early stages, and the accusations made by the plaintiff have not yet been confirmed by the court. Whether the four companies have indeed reached a legally binding deceleration agreement, whether the relevant public statements constitute restrictions on competition in the sense of antitrust law, and the extent to which AI security cooperation should be subject to antitrust laws will all become core issues that need to be resolved in subsequent litigation.
Comments