President Donald Trump’s decision to restrict access to advanced AI models is resurrecting decades-old First Amendment questions – with a modern twist.
Some free speech champions see artificial intelligence models as an expressive tool worthy of constitutional protection, and that the kinds of export controls the administration imposed amount to an unconstitutional prior restraint, forcing companies to seek government permission before they publish.
The legal argument, like the administration’s regulations, is evolving in real time.
Legion, an AI company, last week sued the administration in federal court saying it exceeded its authority with the sweeping export controls it imposed on Anthropic’s frontier models but it did not specifically make a free speech argument.
But it may just be a matter of time before one is filed, according to a white-hat AI researcher granted anonymity to avoid jeopardizing legal strategy.
“There’s an argument to be made that when you’re accessing or interacting with the models it’s a form of speech, and I see it as jawboning,” the researcher told POLITICO. “Export controls are an infringement on speech because when you publish research or security vulnerabilities you’re expressing yourself by accessing the technology and fixing the problems.”The White House referred West Wing Playbook to Commerce, which did not immediately respond to a request for comment.
The concerns stem, in part, from the Trump administration’s order this month to suspend access to Anthropic’s frontier AI models, Fable 5 and Mythos 5 after officials raised concerns about a vulnerability in the system.
Last Friday, the administration partially retreated. Commerce Secretary Howard Lutnick authorized a limited restoration of Mythos 5 for a narrow group of vetted U.S. critical infrastructure and cybersecurity organizations, while broader restrictions remain in place and Fable 5 continues to face tighter limits.
But that doesn’t solve the potential First Amendment issue.
The free-speech theory draws on decades-old battles over encryption software that also involved export controls.
In the 1990s, the Electronic Frontier Foundation successfully challenged federal export controls on encryption code in a groundbreaking case that helped establish the principle that computer code can constitute protected speech under the First Amendment. The 9th U.S. Circuit Court of Appeals in Bernstein v. United States ruled in 1996 that source code was speech protected by the First Amendment and that government regulations preventing its publication were unconstitutional.
Tyler Tone, legislative analyst for technology and free expression at the Foundation for Individual Rights and Expression, argues the same reasoning increasingly applies to artificial intelligence.
“A chatbot is a new tool, but what the user does with it is as old as humanity: asking questions, testing arguments, comparing ideas, and seeking help navigating the accumulated record of human language,” Tone wrote in a recent post. “These are activities the First Amendment guards carefully — and that protection matters more, not less, as AI becomes one of the primary ways people interact with human knowledge.”
Tone argues that AI systems embody layers of human expression—from developers’ choices about training data and model behavior to users’ own right to receive information—placing them squarely within longstanding First Amendment doctrine. Government control over access to AI models risks creating a “chokepoint” between the public and an increasingly important means of accessing knowledge, he argues.
The constitutional argument echoes one of the defining technology battles of the 1990s.
Phil Zimmermann, the creator of the Pretty Good Privacy encryption software, spent three years under federal criminal investigation after the government alleged that publishing his encryption program online violated U.S. export control laws.
Prosecutors declined to indict him, but the episode became a landmark moment in the so-called “crypto wars,” helping establish the principle that computer code could receive First Amendment protection.
Zimmermann sees clear parallels between the government’s past efforts to control encryption and today’s attempts to restrict access to frontier AI models.
Had his case gone to trial, he said, “free speech would have been… probably our most successful strategy for winning in court.” He added that his legal team “intended to make that argument” as a central part of its defense.
But while Zimmermann believes a First Amendment challenge to AI export controls has merit, he argues it is not the most compelling case to make. Instead, he says the administration’s own national security rationale cuts in the opposite direction: cybersecurity researchers need access to the same powerful models that adversaries will inevitably obtain.
“My argument is exactly as urgent as what Trump is trying to say,” Zimmermann told POLITICO. “If Trump is saying, ‘Well, it’s emergency powers … because it’s really urgent,’ well, I say it’s really urgent too—but I’m saying that it’s really urgent that we not do it that way.”
Essentially, he believes the best way to stop a bad guy with a dangerous AI model is making sure the good guys have access to the same model.
“We have to get our hands on these tools to harden our software,” he said. “The bad guys are certainly going to get their hands on these tools … and they are going to attack us. We need these tools to identify the vulnerabilities so that we can fix them.”
While courts may embrace a free speech argument, Zimmermann said, “we don’t have three years” to wait for litigation while more capable AI systems proliferate.
Like this content? Consider signing up for POLITICO’s West Wing Playbook: Remaking Government newsletter.
