Anthropic Wins Legal Round Against U.S. Government in AI Battle

A federal appeals court has rebuffed the administration's attempts to restrict Anthropic's latest AI models, marking a major turning point for the future of AI governance.

A federal appeals court handed the government a notable defeat on Thursday, ruling that the Commerce Department overstepped its authority when it moved to impose export and domestic deployment restrictions on Anthropic's Claude model family. The three-judge panel, sitting in the D.C. Circuit, found in a 2-1 decision that the agency had leaned too heavily on a broad reading of the International Emergency Economic Powers Act — a legal theory the court said required explicit congressional authorization before it could be applied to domestic AI software.The case had been building since late 2025, when the Commerce Department's Bureau of Industry and Security placed several frontier AI systems, including Anthropic's Claude series, into a new dual-use technology category subject to export licensing requirements.The government's argument was not unreasonable on its face: AI systems capable of advanced reasoning and complex code generation could, in theory, accelerate weapons research or give foreign adversaries a strategic edge. On that basis, the administration believed it already held the standing authority to act.Anthropologic filed suit in early 2026, arguing the rules were procedurally defective, economically damaging, and built on a legal framework with no clear statutory footing.The company's legal team cited the Supreme Court's 2022 West Virginia v. EPA decision extensively, arguing that regulatory decisions of such sweeping economic consequence demand a clear statement from Congress not an agency's creative reading of a decades-old trade statute.The appeals court sided with Anthropic on the core question. Writing for the majority, Circuit Judge Elena Forsythe held that the government had failed to identify a congressionally granted power to regulate the domestic distribution of AI software under the framework it invoked. The ruling does not strip the executive branch of all authority over AI-related exports, but it draws a firm line between controlling cross-border transfers and dictating how an American company may distribute its own products inside the United States. That distinction, the majority wrote, is not a minor procedural point — it goes to the heart of how much latitude agencies can claim in areas Congress has not directly addressed.The dissent, authored by Judge Michael Carver, pushed back sharply. He argued the majority had applied the major questions doctrine with more aggression than the case warranted, and that the national security record before the court more than justified the administration's caution. In his reading, the question was not whether Congress had spoken with enough clarity but whether the executive branch had acted rationally in an area where it has traditionally held wide discretion. That division sets up a credible path to en banc review — or, eventually, a petition to the Supreme Court.For Anthropic, Thursday's ruling is a clear short-term win but leaves longer-term questions wide open.The company can move forward with its current model deployments without the licensing burden the BIS rules would have imposed. The government, however, retains the right to appeal, and a more carefully drafted regulatory structure could survive future legal challenge. The rest of the AI industry is paying close attention. OpenAI, Google DeepMind, and Meta have all navigated versions of the same regulatory pressure in recent months, and this decision is the sharpest signal yet that courts are not willing to extend agencies unlimited latitude in this space without clearer direction from lawmakers.What happens next hinges largely on Congress. A handful of senators from both parties have already called for legislation granting the executive branch cleaner statutory authority over frontier AI development. Others maintain that the judicial check is performing exactly as intended, and that any new law must be drawn narrowly enough to protect American companies from regulatory overreach. The two camps remain far apart, and the legislative calendar offers no obvious opening. For now, Anthropic's models stay on the market, the government's legal options stay on the table, and the fundamental question at the center of this fight — who ultimately decides how the most powerful AI systems in the country get regulated — stays entirely unresolved.