AI: Anthropic gets a major reprieve in its battle against the Pentagon

New episode in the confrontation between the Trump administration and Anthropic. American justice has just brutally slowed down a Pentagon offensive against the company behind Claude. A decision that could reshape the balance of forces in a sector that has become highly strategic for the economy and American sovereignty.

Federal judge stops Pentagon, Anthropic's AI stays in the game

Yesterday, March 26, in San Francisco, federal judge Rita Lin, of the Northern District Court of California, put a stop to the Pentagon's ambitions. She issued a preliminary injunction that immediately suspends two key actions by the Trump administration.

Concretely, justice freezes both the classification of Anthropic as a threat to national security and the order imposed on federal agencies to cease all use of Claude, its artificial intelligence model.

The judge did not mince her words. She described these measures as “arbitrary, capricious and constituting an abuse of discretionary power”. Better yet, she ruled on the ethical merits of the case:

Nothing in current law supports the Orwellian notion that an American company could be labeled a potential adversary and saboteur of the United States for expressing disagreement with the government.

A scathing statement, which alone sums up the seriousness of the matter. It clearly suggests that the State would have crossed a red line by sanctioning a company for its public positions.

To understand this escalation, we have to go back to the summer of 2025. At the time, Anthropic and the Pentagon were negotiating a strategic partnership. The objective is ambitious: to integrate Claude as the first advanced AI model authorized to operate on classified networks.

However, the discussions bogged down, then broke up in February 2026. The Pentagon hardened its position and demanded unrestricted access to the technology, including for sensitive uses. Among them: the development of lethal autonomous weapons and mass surveillance devices on American territory.

Anthropic categorically refuses. For the company, these uses cross non-negotiable ethical boundaries. From then on, the breakup becomes inevitable, and opens the way to a legal confrontation that is now public.

From a legal complaint to a provisional legal victory

The reaction from the Trump administration is not long in coming. At the end of February, Donald Trump ordered all federal agencies to stop using Anthropic products.

In the process, the Pentagon included the start-up on its blacklist of risky suppliers, a list which until then only included foreign companies like Huawei or Kaspersky.

Faced with these measures, Anthropic responded on March 9, 2026 by summoning the American government before a federal court in Washington DC. The company denounced a flagrant violation of the First Amendment: sanctioning a company for having exercised its freedom of expression amounts, according to it, to “destroying” a major player in American AI.

This appeal receives unexpected support of 37 engineers and researchers from Google DeepMind and OpenAI, who submitted a voluntary submission in favor of Anthropic.

During a 90-minute hearing on March 24, Judge Lin questioned the government's lawyers on one specific point: Is Anthropic being punished for publicly criticizing the Pentagon?

The implicit answer in his decision, rendered two days later, is unambiguous. She speaks of “classic illegal retaliation, contrary to the First Amendment”.

This decision goes well beyond the simple Anthropic case. It establishes a principle: no administration can use its economic power to muzzle a private company on the grounds that it has refused to collaborate on projects contrary to its ethical values.

With 32% share of the enterprise AI market recorded in 2025, according to Menlo Ventures, Anthropic is not a secondary player, its weakening would have direct repercussions on the technological competitiveness of the United States against China. The legal battle is not over, but Anthropic has, for now, marked a decisive point.