An artificial intelligence (AI) startup has sued the U.S. government following a federal directive that forced Anthropic to block foreign nationals from accessing its most advanced models, according to a lawsuit filed Monday in a Washington, D.C., federal court.
The litigation-technology firm, Legion, claims the Trump administration’s strict export controls have caused “immediate, irreparable and existential” harm to its operations.
The legal battle began less than two weeks after Anthropic disabled access to its highly anticipated Fable 5 and Mythos 5 models. The restriction was triggered by a June 12 order from the Commerce Department aiming to keep the proprietary tech out of foreign hands.
Legion, which develops software tools for attorneys, is based in the U.S. but employs Canadian nationals working remotely from Canada. By cutting off access to Fable 5, Anthropic “instantaneously” stripped the startup of its core development tool, according to the filing.
“Each day the directive remains in force disrupts Legion’s product, operations, sidelines its engineers, and erodes the company’s ability to survive in a field defined by continuous access to the most capable models,” the company said.
The lawsuit names defendants including Commerce Secretary Howard Lutnick, who previously issued a formal warning to Anthropic CEO Dario Amodei. Lutnick’s letter stipulated that Anthropic must obtain explicit government permission before exporting its advanced models internationally or allowing foreign nationals to access them.
Neither the White House, the Commerce Department, nor Legion immediately responded to requests for comment. Anthropic declined to comment on the litigation but stated it is “grateful to the administration” for working to resolve the issue, adding that it remains committed to “making sure the U.S. leads in AI.”
The federal crackdown highlights growing anxieties over Anthropic’s new Mythos architecture, a general-purpose AI model that boasts unprecedented coding and reasoning capabilities.
According to Anthropic, Mythos Preview represents a massive leap in autonomous cyber capabilities, possessing the ability to find and exploit software vulnerabilities with minimal human intervention. During internal testing, the AI reportedly discovered thousands of zero-day vulnerabilities across every major operating system and web browser.
“The vulnerabilities it has spotted have in some cases survived decades of human review and millions of automated security tests,” Anthropic said in a statement.
Because the tool is sophisticated enough to outperform all but the most skilled human hackers, Anthropic initially restricted its release to a heavily vetted group of 200 partner organizations under a defensive initiative dubbed Project Glasswing. Tech giants and financial institutions — including Apple Inc., Microsoft Corp., Google, Amazon.com Inc., NVIDIA Corp., and JPMorgan Chase — were granted early access to use the model for defensive cybersecurity patching.
Despite Anthropic’s controlled rollout, federal regulators intervened after discovering the potential for users to jailbreak — or bypass the safety guardrails of — Fable 5, a version of Mythos altered to prevent it from executing cybersecurity tasks.
Security concerns were further amplified by reports that an unauthorized group of users in a private online forum briefly gained access to Mythos in April.
While Anthropic touts the model as a revolutionary tool for digital defense, some independent experts urge caution regarding the company’s claims. Gang Wang, an associate professor of computer science at the University of Illinois, noted that independent researchers have not yet been granted the hands-on access required to verify the model’s purported capabilities.
As the legal challenge moves forward, the case marks a critical flashpoint in the domestic oversight of artificial intelligence, signaling a more aggressive, national-security-focused phase in the global cyber arms race.
