An Anthropic customer sues the US Government over losing access to the Fable 5 and Mythos 5 models

By Bloomberg · June 23, 2026.

Legion, a U.S.-based legal technology company that develops tools for lawyers, has filed a lawsuit in a federal court in Washington against the U.S. government. The reason: the Trump administration's order that forced Anthropic to disable access to its most advanced AI models—Fable 5 and Mythos 5—for foreign nationals, directly affecting Legion's Canadian employees who work from Canada.

The lawsuit was filed on Tuesday, June 23, 2026, less than two weeks after Anthropic disconnected access to both models in compliance with the export controls imposed by the Trump administration. The article's image shows the 'Claude Fable' logo, which indicates that Fable 5 is part of Anthropic's Claude family of models, although under its own commercial name.

The core of the conflict lies in the tension between the export-control policy for advanced AI technology—which seeks to prevent foreign powers from accessing the most capable models—and the real operational needs of U.S. companies with internationally distributed teams. Legion argues that its Canadian software developers, although they are not U.S. citizens, work for an American company and for the benefit of American clients, so the restriction harms them disproportionately without there being a clear national security risk.

This case illustrates an emerging issue in the agentic AI ecosystem: advanced language models have become critical infrastructure for startups and professional-services companies such as legal technology (LegalTech). When access to those models is suddenly interrupted for regulatory reasons, the operational impact is immediate and can result in business losses that are difficult to quantify. Legion is not suing Anthropic, but the government, which clearly signals that the company understands Anthropic acted in compliance with a legal obligation, not by its own decision.

From the perspective of agentic AI, the case is especially relevant because the Fable 5 and Mythos 5 models appear to be Anthropic's most powerful at this moment, and their use in LegalTech applications very likely involves agentic workflows: document analysis, generation of legal arguments, automated research and litigation assistance. The sudden deprivation of access to these models affects not only a specific function, but entire automation systems that depend on their capabilities.

As sector context, AI technology export controls have gained relevance in recent years in the U.S. The Biden administration already pushed restrictions on AI chips (especially from NVIDIA) toward China and other countries considered to be at risk. The extension of that logic to the software models themselves—not just to hardware—represents a new and more controversial qualitative step, because the models are intangible and their 'export' can occur simply by granting access to an API from abroad.

Legion's lawsuit could become a reference case on how far the American government can go in restricting access to AI models for non-U.S. citizens, even when they work for American companies. The distinction between 'foreign national working for a U.S. company' and 'adversary foreign entity' is precisely the one Legion will try to get the court to establish.

For Anthropic, the situation is uncomfortable: it has complied with the government order, but that has cost it customers and placed it at the center of a lawsuit that, although not directed against it, involves its products and its name. The company will have to decide whether to intervene in the proceedings as an interested party or stay on the sidelines.

The outcome of this litigation could have broad implications for the entire industry: if the courts limit the scope of export restrictions on AI model APIs, other providers such as OpenAI, Google DeepMind or xAI will also be affected in how they manage international access to their most capable models. If, on the contrary, the courts validate the government's approach, a system could become normalized in which developers outside the U.S.—even in allied countries like Canada—have restricted or differentiated access to the most advanced AI models.

The date of the lawsuit—June 23, 2026—and the span of less than two weeks since the disconnection show the urgency with which Legion has acted, probably because the service interruption is already affecting its operations and its commitments to clients in the legal sector. In that sense, it is possible that the company has also requested injunctive relief to restore access while the merits of the matter are resolved.