President Donald Trump’s decision to restrict access to advanced AI models is reviving a decades-old First Amendment issue with a modern twist.
Some free speech advocates are considering artificial intelligence models as an expression tool It deserves constitutional protection, and the export controls imposed by the regime amount to unconstitutional prior restrictions, requiring companies to seek government permission before publishing.
Like government regulations, legal debates are evolving in real time.
Last week, AI company Legion sued the government in federal court He said the sweeping export controls imposed on Anthropic’s Frontier models exceeded his authority, but he did not specifically argue about freedom of speech.
But a white hat AI researcher, who was granted anonymity to avoid jeopardizing his legal strategy, says it may only be a matter of time before a lawsuit is filed.
“There’s an argument that when you access a model or interact with a model, it’s a type of audio, and I think that’s a jaw-breaking act,” the researcher told POLITICO. “Export controls are a violation of speech, because when you publish research or security vulnerabilities, you are expressing yourself by accessing that technology and solving the problem.” The White House referred West Wing Playbook to the Commerce Department, which did not immediately respond to a request for comment.
Part of the concern stems from the Trump administration’s order earlier this month to cut off access to Anthropic’s Fable 5 and Mythos 5 frontier AI models after authorities raised concerns about vulnerabilities in the systems.
Last Friday, The regime partially retreated. Although Secretary of Commerce Howard Lutnick authorized the limited restoration of Mythos 5 to a limited group of vetted U.S. critical infrastructure and cybersecurity organizations, broader restrictions remain in place and Fable 5 continues to face severe restrictions.
But that doesn’t resolve potential First Amendment issues.
The free speech theory is based on a decades-long battle over encryption software that also involved export controls.
1990s, Electronic Frontier Foundation I successfully attempted the challenge. The federal government’s export control of encryption code is a landmark case that helped establish the principle that computer code can constitute protected speech under the First Amendment. 9th U.S. Circuit Court of Appeals Bernstein v. United States In 1996, a court ruled that the source code was protected speech under the First Amendment and that government regulations that prevented its release were unconstitutional.
Tyler Torn, a technology and freedom of expression legislative analyst at the Individual Rights Expression Foundation, argues that the same argument increasingly applies to artificial intelligence.
“Chatbots are new tools, but what users do with them is as old as human history: ask questions, test arguments, compare ideas, and ask for help navigating the accumulated record of human language,” Tone wrote in the paper. recent posts. “These are activities that the First Amendment carefully protects, and that protection will become increasingly important 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 rights to receive information, placing them squarely within long-standing First Amendment principles. He argues that government control over access to AI models risks creating “choke points” between citizens and increasingly important means of accessing knowledge.
This constitutional debate reflects one of the battles that defined the technology wars of the 1990s.
Phil Zimmermann, the developer of the Pretty Good Privacy encryption software, spent three years under federal criminal investigation after the government claimed he violated U.S. export control laws by publishing his encryption program online.
Although prosecutors declined to charge him, the episode marked a landmark moment in the so-called “cryptocurrency wars” and helped establish the principle that computer code enjoys First Amendment protection.
Zimmerman sees clear parallels between past government efforts to control encryption and today’s efforts to limit access to frontier AI models.
If his case had gone to court, he said, “free speech … would probably have been the most successful strategy for winning in court.” He added that his team “intended to make that argument” as the centerpiece of their defense.
But while Zimmerman believes the First Amendment challenge to AI export controls is worthwhile, he argues that it’s not the most persuasive argument. Rather, he says, the administration’s own national security rationale goes in the opposite direction: Cybersecurity researchers need access to the same powerful models that adversaries will inevitably have access to.
“My argument is just as urgent as what Mr. Trump is trying to say,” Zimmerman told Politico. “If Trump says, ‘That’s an emergency power…because it’s really urgent,’ and I’m saying it’s really urgent, but it’s really urgent, I’m saying it shouldn’t be done that way.”
Essentially, he believes the best way to stop bad guys from using dangerous AI models is to give good guys access to those same models.
“We need to get these tools to harden our software,” he said. “The bad guys are definitely going to get their hands on these tools… and they’re going to attack us. We need these tools to identify vulnerabilities so we can remediate them.”
Courts may accept free speech arguments, Zimmerman said, but “three years is not enough” to wait for a lawsuit to wait until more sophisticated AI systems become widespread.
Like this content? Consider subscribing to POLITICO West Wing Playbook: Reinventing the government newsletter.
