Legion LegalTech's complaint targets a Bureau of Industry and Security directive that forced Anthropic to cut off Fable 5 access for foreign nationals worldwide, and argues the legal classification underlying the directive was rescinded in May 2025.
Legion LegalTech Corp., a legal-tech startup identified in some reporting as San Jose-based, filed suit in federal court in Washington on June 23, 2026, to overturn the U.S. directive that forced Anthropic to cut off access to its Fable 5 and Mythos 5 AI models for any foreign national worldwide (complaint, lawsuit landing). The lawsuit's sharpest argument has nothing to do with AI policy. The export control the government is enforcing, the complaint argues, was already abolished last year.
The legal architecture of the case is built around what may be the government's most awkward vulnerability. The Bureau of Industry and Security used a June 12 "Is Informed" letter, supported by Section 4817(b)(1) of the Export Control Reform Act and the International Emergency Economic Powers Act (IEEPA), to force Anthropic to pull access. But according to the complaint, the only direct AI-model-weights export classification that ever existed, ECCN 4E091 (the Export Control Classification Number covering AI model weights as a controlled item), was rescinded in May 2025 with no replacement. That creates a situation the lawsuit frames bluntly: "Commerce cannot enforce a control that does not exist" (Gizmodo).
The argument cuts against the typical framing of the case as a trade-and-tech-nationalism dispute. This is, at heart, an administrative-law challenge. The government is moving fast enough on AI that it is leaning on a control with no underlying legal text, and threatening criminal and civil penalties for violating it.
Legion's complaint adds several other lines of attack. Hosted AI access, the company argues, does not transfer model weights, source code, training data, or technical know-how, so no qualifying "export" occurs at all (The Next Web). The directive, it says, was issued without the required rulemaking or interagency coordination, conflicts with a June 2 Trump executive order that disclaimed any licensing or preclearance regime for new AI models, and was arbitrary and capricious: a narrow jailbreak-fix-codebase concern drove a categorical worldwide ban while comparable capability remained available through OpenAI's GPT-5.5 (MediaNama).
The case is not abstract for Legion. Founded in 2024, the company drafts pleadings, discovery, motions, and exports Word and PDF files for lawyers. Its engineering team includes Canadian nationals working from Canada, the exact population the directive targets. As CEO Arthur Rothrock told Bloomberg, the worry is precedent: "Who's to say they can't do this any other time against another company, like OpenAI?" (Bloomberg).
A read of the complaint, attributed to a Reuters court-filing summary circulating on X, characterized the government's demand as requiring Anthropic to disable Fable 5 access for every foreign national on Earth within roughly 90 minutes, under threat of criminal and civil penalties (Reuters via Rohan Paul). That figure is not corroborated in the primary complaint text available and should be read as the Reuters read, not as independently verified.
Anthropic is not a defendant, but the company told Bloomberg it is "grateful to the administration" and "committed to working alongside the government" (Gizmodo). The White House and Commerce had not responded at filing.
The Fable 5 episode itself is brief. Anthropic released the model on June 9, 2026, and pulled it on June 12, after roughly three days of public availability. Fable 5 sits in the same capability class as Claude Mythos Preview, the model Anthropic had previously said was too dangerous to release without elaborate safeguards. The directive followed Amazon-internal research reportedly finding workarounds that bypassed those safeguards, and prior reporting that China-connected entities had accessed Mythos (AIWeekly).
Collateral damage has already been documented. The United Kingdom's AI Security Institute had been actively evaluating the affected models and lost access as a side effect of the BIS directive (Startup Fortune).
The broader picture is messier than one case. Anthropic and the U.S. government are already locked in separate litigation in Washington and California over the Pentagon's designation of Anthropic as a supply-chain risk, a designation that followed Anthropic's refusal to allow Claude to be used for autonomous weapons or domestic surveillance. The Legion case now asks a court to decide whether the government can reach the same kind of outcome by executive directive rather than by formal rule.
The most direct question the suit puts before the court is the smallest one: does a directive that disables a commercial AI product for an entire category of users survive when the underlying classification it depends on has already been deleted from the books?