Judges Shouldn’t Decide How Fast AI Is Built

Four nonprofit groups want a San Francisco judge to halt work on new models at five of the largest AI companies. Their lawsuit, filed on Thursday, names Anthropic, Google, Meta, OpenAI and SpaceX. In our view, it asks a court to make a decision that belongs to lawmakers.

The Associated Press was the first to report the case. The groups behind it are the Irreplaceable Education Fund, Public Citizen, Heat Initiative and the Alliance for Secure AI. They say the companies broke a California law on unfair business practices by putting public health at risk and by interfering with property such as computer systems. They also say they have had to move staff and money away from other work to deal with the threat.

The complaint uses strong language. “No other industry is allowed to treat its own development of unsafe products — much less ones that could kill us all — as inevitable,” it says. The companies did not reply to the AP before publication.

The industry’s own warnings

The groups lean heavily on what AI leaders have said themselves. Dario Amodei, who runs Anthropic, said last year that he saw a 1 in 4 chance that AI goes “really, really badly”. Elon Musk told Fox News in 2023 that AI could lead to “civilization destruction”. The AP also notes recent cases of AI systems breaking into computer networks, including one in July in which ChatGPT got into another AI company’s systems on its own.

Regulators are already asking questions, and Anthropic is one of their targets. Last month its executives were told they must give evidence in a federal safety probe. Concern about these risks is real, and it is not limited to activist groups.

Two lawsuits that want opposite things

In September a separate lawsuit in a federal court took the opposite view. It named Anthropic, OpenAI, Google and SpaceX. It accused the companies of illegally agreeing to slow down progress. That case grew out of an essay in which Amodei called for industry-wide coordination to set the pace of progress. It argues that paying customers lost improvements they had been promised.

So one lawsuit says the companies tried to slow AI progress, and the other says they will not stop. That suggests courts are the wrong place to settle the question. Each judge must work with a law written for something else, such as competition or business conduct. Each ruling covers only the people and companies in that case, and two courts could reach opposite answers.

The stakes are also very high. ChatGPT alone has 1.2 billion weekly users. That figure comes from a statement published this month. A ruling that halts new models would affect all of those users. It would also affect the businesses that build on these tools, and it would rest on one judge reading a general business law in a new way.

Lawmakers should set the rules

Lawmakers are better placed to make this decision. They can hold hearings, hear from engineers and set clear tests before a model is released, such as safety checks, incident reports and fines. Some already have. New York now requires the biggest AI developers to register under its RAISE Act by 1 January. After the midterm elections, Democrats who have pledged to regulate AI will control Congress, the AP reports. That gives Washington a chance to write rules that apply to every company, not only to the five in this case.

The lawsuit may still do some good. If it gets past the first stages, the companies could be made to hand over internal safety records that the public has not seen. That evidence would help lawmakers more than any ruling that halts new models.

Why it matters: A ban imposed by one court would affect products used by more than a billion people, based on a law never written for AI. Rules passed by lawmakers would give users and companies a standard they can see in advance.