SpaceXAI filed a lawsuit against a California law that could, even if it doesn’t win, upend AI disclosure requirements nationwide
AI transparency laws across the US are under threat from a lawsuit being pursued by SpaceXAI.
Elon Musk’s company is challenging California’s AB 2013, which requires AI companies to provide high-level summaries of training data, arguing that it breaches First Amendment rights.
While the specific requirement SpaceXAI — formerly known as xAI — is challenging concerns training data, legal experts are worried that an appeals court ruling could have far wider implications for transparency laws, potentially nullifying both those already in force and in the future. “Right now the public knows almost nothing about what these models are trained on, and xAI wants to keep it that way,” said Ben Rashkovich, Senior Counsel at Legal Advocates for Safe Science and Technology (LASST), which today filed an amicus brief opposing the suit alongside a host of other organizations and individuals.
“Transparency is the floor for AI governance, not the ceiling — and because the Ninth Circuit’s reasoning here will shape every AI regulation that comes after, this case is about whether the First Amendment can be used to knock that floor out.”
When xAI first filed the suit on December 29, it asked for a preliminary injunction in the case, but was denied. The injunction would have temporarily d regulation. The case has now moved to a US Court of Appeals, which can decide to grant the injunction or make a ruling on the law.
If the judges side with SpaceXAI and make a broad ruling on the First Amendment argument, it could have wide-ranging implications for the transparency measures in laws such as California’s SB 53, Illinois’ SB 315 or New York’s RAISE Act, as well as simpler transparency laws. The legal argument put forward by the company is threefold. On the First Amendment, it argues that by mandating disclosures, AB 2013 is unlawfully compelling it to speak. It also says California is doing so in a discriminatory way on the basis of the company’s viewpoints, which would mean the court must apply a strong preference for preserving xAI’s free speech rights. Should the court rule that a strict interpretation of the First Amendment applies, and that AB 2013 breaches it, nearly any disclosure law — AI-related or otherwise — could be at risk. “I think you could plausibly say xAI doesn’t want government disclosure requirements for AI, period … It’s a very doctrinal brief because it’s a very doctrinal argument,” said Rashkovich.
xAI also made two other arguments: that disclosure of these alleged trade secrets amounts to destruction of property, and that the law is “unconstitutionally vague.” Rulings in favor of either argument could invalidate AB 2013, but likely pose less of a threat to other AI transparency laws.
The case, and its repercussions, will be decided on which level of First Amendment test the appeal judges decide to apply. SpaceXAI wants “strict scrutiny,” which would apply the First Amendment as a right with few exceptions. The lower court applied an intermediate level, which requires that the state is deemed to have a valid interest in the information being disclosed. California says its interest is based on the need to provide consumers with accurate information to guide purchasing decisions. Even this intermediate level would give companies a stronger basis on which to sue, and potentially hold up legislation.
California and LASST** **are thus arguing the case should be considered under “rational basis,” which would mean a less stringent interpretation of xAI’s First Amendment rights and give the government more flexibility to compel speech. This would protect other disclosure frameworks that have already been implemented, and retain them as a feasible path for future policymaking.
“The cost of these legal challenges is significant,” Rashkovich says. “If AI companies keep moving for preliminary injunctions against common-sense AI regulations, suddenly these laws are pausing or possibly unpausing unpredictably.” That means that information could be withheld from the public while model capabilities rapidly accelerate, potentially developing dangerous capabilities.
Among other arguments, the amicus brief says the state clearly has an interest because of pressure to regulate AI. That pressure is itself evidenced by its long list of almost 30 co-signatories, which span First Amendment scholars, industry groups, tech safety policy groups and child safety organizations. These include Americans for Responsible Innovation, the Electronic Privacy Information Center and the Surveillance Technology Oversight Project.
Rashkovich believes the courts are likely to tilt in California’s favor, and Americans for Responsible Innovation spokesperson Chris MacKenzie likened the lawsuit to existential risks: something that has a relatively low probability of becoming a problem, but which could have tremendous consequences should things go the wrong way.
“If we close this avenue off for ourselves, we’re closing off quite a bit — not just laws in three states, but a whole avenue of public transparency that Congress and state lawmakers are both interested in,” he said. “What’s the likelihood that [SpaceXAI] wins? Probably pretty low. But we want to mobilize, with enough resources to reduce that threat.”