Filed June 23, 2026 in the U.S. District Court for the District of Columbia, No. 1:26-cv-02225, by Spencer Fane LLP for Legion LegalTech, Corp. Defendants are the United States, the Department of Commerce, Secretary Howard Lutnick, the Bureau of Industry and Security, Under Secretary Jeffrey Kessler, the Executive Office of the President, and ten Doe defendants. See Legion v. United States (Anthropic export-directive challenge).
The text summarized here was retrieved from the plaintiff's own domain; the as-filed version should be confirmed against the D.D.C. docket.
The directive challenged
As alleged: "On June 12, 2026, the federal government ordered Anthropic, PBC to immediately disable two of its most capable artificial-intelligence models — Fable 5 and Mythos 5 — for every foreign national on Earth." The order "reached Anthropic by letter from Commerce's BIS, gave the company ninety minutes to comply under threat of 'prompt criminal and civil penalties,' and purports to rest on the export-control authority that Commerce and its BIS administer." Within hours, "hundreds of millions of users — including citizens of the United States' closest allies and Anthropic's own employees — lost access."
Legion describes itself as "a U.S.-based AI-native litigation-technology company that builds drafting and case-management tools for attorneys on top of frontier AI models," a commercial Anthropic customer with a licence to Fable 5, whose "software development team includes Canadian nationals working remotely from Canada, a Five Eyes intelligence partner and a Country Group A:5 nation under Defendants' own Export Administration Regulations."
First claim: ultra vires under ECRA
The argument is that no operative control exists to enforce. "The only export-control classification that ever directly covered advanced AI model weights — ECCN 4E091 — was rescinded in May 2025 with no replacement. No currently operative provision of the Commerce Control List classifies access to a hosted AI model, or its inferential text output, as a controlled item. Commerce cannot enforce a control that does not exist."
The second limb attacks the definition of export. The EAR's deemed-export framework "reaches only the release of 'technology' or 'source code,' expressly 'but not object code,'" and Legion argues that "providing a person with access to a hosted AI model that returns text in response to prompts releases no technology, source code, model weights, or technical data to the user" — placing hosted inference outside the export concept altogether.
The complaint then addresses the two authorities the directive invoked. Section 4817(b)(1) interim controls "requires notice-and-comment rulemaking and multilateral coordination that never occurred." The 15 C.F.R. § 744.22(b) "is informed" letter is "a targeted, case-specific tool for identified military-intelligence end-use risks, not a mechanism for imposing blanket worldwide bans untethered from any identified end user or end use."
The claim is pleaded as a nonstatutory ultra vires action rather than under the APA, because "ECRA precludes Administrative Procedure Act review of BIS control-list functions but does not preclude all forms of judicial review" — a form of equitable relief the complaint says "the D.C. Circuit has repeatedly held remains available under ECRA."
Second claim: IEEPA and the Berman Amendment
If the directive rests on IEEPA instead, the complaint argues the Berman Amendment forecloses it. 50 U.S.C. § 1702(b)(3) "categorically excludes from the President's IEEPA authority any power to regulate or prohibit — whether directly or indirectly — the exportation of information or informational materials in any format or medium of transmission." The complaint's framing of the provision's structure is that "the statute does not limit this authority — it withholds it entirely."
The application: "The expressive outputs that Legion receives from the Fable 5 model — drafted text, written legal analysis, summaries, and similar composed material — are informational materials within the meaning of this provision, and the directive's purpose and effect are to stop the flow of those materials to their recipients."
Two independent IEEPA defects are pleaded. No national emergency was declared as to "the code-analysis capability of a commercially hosted AI model," and the alleged threat lacks the "source in whole or substantial part outside the United States" that § 1701(a) requires, since "the model was developed by a U.S. company, hosted in the United States, and offered through a U.S. commercial platform." The complaint also invokes the major-questions framing, arguing the directive "asserts authority of vast economic and political significance, a de facto licensing regime under which any frontier AI model must 'go through the administration,' that IEEPA does not clearly and explicitly confer." Because IEEPA authority runs to the President and the President is not an APA agency, "nonstatutory ultra vires review is the only mechanism" available.
Third claim: arbitrary and capricious
The mismatch alleged is between "the asserted threat, a narrow jailbreak allowing a model to review software code, and the remedy selected — a categorical worldwide suspension of access by any foreign national regardless of nationality, location, employer, or purpose."
The underinclusiveness allegation is the sharpest: "The identical code-analysis capability remains publicly available through competing products the directive left untouched, including OpenAI's GPT-5.5." Less-restrictive alternatives — "restricting access only for nationals of adversary nations" — are alleged to have gone unconsidered.
On process, the complaint alleges, "on information and belief," that "the directive was precipitated not by any neutral agency assessment but by telephone calls from Anthropic's commercial rivals over a single evening, followed by a ninety-minute ultimatum and no disclosed independent analysis."
The retaliation context
The complaint places the directive in a sequence: after Anthropic "refused to grant the Department of Defense unrestricted access to its AI tools," the President "publicly ordered all federal agencies to stop using Anthropic's technology, the Secretary of Defense designated Anthropic a 'supply chain risk,' and the President threatened Anthropic with 'major civil and criminal consequences.'"
It relies on a prior judicial finding rather than only its own characterization, citing Anthropic PBC v. U.S. Dep't of War, No. 26-CV-01996-RFL, 2026 WL 836842, at *17 (N.D. Cal. Mar. 26, 2026), where a court "already found that an earlier government action against Anthropic under a national-security label was likely pretextual and that the government's real motive was unlawful retaliation." See Anthropic v. United States (Pentagon ban challenge).
The executive-order argument is structural: on June 2, 2026, ten days before the directive, the President signed "Promoting Advanced Artificial Intelligence Innovation and Security," 91 FR 34565, which "expressly disclaimed authority to impose 'a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models.'" The complaint's inference: "Where the Executive publicly disclaims a power and then exercises it, the inference that no statute confers the power is compelling."
Harm and stakes
Harm is pleaded as "immediate, irreparable, and existential," on the reasoning that "the pace of frontier AI advancement is blistering, and competitive ground lost during a suspension cannot be regained after the fact," and that because the claims lie against the federal government, "economic losses are unrecoverable as damages, and equitable relief is the only adequate remedy."
The systemic argument: the directive "asserts a power to switch off a lawful and already-published commercial information service relied upon by hundreds of millions of users worldwide, on a national-security label unsupported by any disclosed evidence and contradicted by the government's own conduct." Left standing, "it would establish that the Executive may, by unreviewed command, disable any frontier AI model at will."
Relief sought
Declaratory judgment under 28 U.S.C. § 2201 that the directive is unlawful and void; an order vacating it in its entirety; preliminary and permanent injunctions against enforcement, either generally or as applied to Legion and its personnel; a § 705 stay of the effective date if the APA applies; costs; and fees under the Equal Access to Justice Act.
Relationships
- supports: Legion v. United States (Anthropic export-directive challenge) — the operative filing
- related: Anthropic v. United States (Pentagon ban challenge) — the prior action whose pretext finding the complaint relies on
- related: Export Controls (AI) — pleads that no operative ECCN covers hosted-model access after the May 2025 rescission of 4E091
- related: Anthropic, Bureau of Industry and Security (BIS), EO — Promoting Advanced AI Innovation and Security (Trump, signed June 2, 2026)