The Governance Vacuum Is Now a Courtroom

On March 4, 2026, the Department of Defense formally designated Anthropic a "supply chain risk" — the first time that authority, designed for foreign adversaries like Huawei, has been used against an American company. Five days later, Anthropic filed two lawsuits. For the background on the dispute that led here, read our earlier coverage.

Anthropic went to court in two jurisdictions simultaneously: a civil complaint in the Northern District of California, and a petition for review in the D.C. Circuit — the proper venue for challenging designations under the Federal Acquisition Supply Chain Security Act of 2018. Both filings argue the same core claim: that the designation violates the First Amendment, the Fifth Amendment, and the Administrative Procedure Act.

The constitutional argument is worth reading carefully, because it is not primarily about AI. It is about what the government can do to a company that refuses to comply with its demands.

Anthropic drew two lines before the designation: Claude would not be used for mass domestic surveillance of U.S. citizens, and it would not be used in fully autonomous lethal weapons systems. The Pentagon wanted "all lawful use" authorization. Anthropic refused. The designation followed.

Anthropic's First Amendment claim is that the designation punishes the company for protected speech — specifically, for its publicly stated position on AI safety in warfare. The Fifth Amendment claim is procedural: severe penalties were imposed under FASCA with no meaningful opportunity to contest them before they took effect. The APA claim is that the designation exceeds the statute's authority, which was written for supply chain threats, not domestic policy disputes with American companies.

Lawfare, which has covered national security law for fifteen years, noted that the designation is unlikely to survive legal challenge: FASCA's Section 3252 process requires notice and an opportunity to respond, and the immediate-effect implementation Anthropic received appears to violate that requirement. The statute has no domestic precedent. Courts will be reading it for the first time in a context it was not written for.

The amicus landscape tells you something about how the legal and technology communities read the stakes. Microsoft filed a brief arguing that Anthropic's products are a fundamental layer in its own military offerings and that the designation threatens American armed forces. Roughly 35 employees of OpenAI and Google — including Google Chief Scientist Jeff Dean, filing in their personal capacities — argued that the Pentagon's action sets a dangerous precedent for how governments respond when AI companies implement ethical guardrails. Twenty-two former senior military officials, including General Michael Hayden (former CIA and NSA director) and two former Secretaries of the Navy, called the designation "retribution against a private company that has displeased the leadership" and warned it could jeopardize soldiers during ongoing operations. The EFF, Cato Institute, Amazon, Apple, and NVIDIA's policy arm have all filed or supported briefs on Anthropic's side.

This is not a divided coalition. It is unusual.

There is a detail in the Guardian's reporting that deserves more attention than it has received.

Claude was used extensively in DoD military operations, including — according to the Guardian — "deciding where to target missile strikes in its war against Iran." The same model. The same ethical architecture that the Pentagon wants removed. Anthropic's refusal to allow fully autonomous weapons use did not prevent Claude from being embedded in a targeting system for active military strikes. It was present. It was being used. And the Pentagon was simultaneously designating the company that built it a national security threat.

The restrictions Anthropic drew were not, apparently, an obstacle to operational military use of Claude. They were an obstacle to the Pentagon's preferred contractual terms. The designation came not because Claude was failing at its job but because its maker would not sign away the right to have limits on what that job could include.

There is a contradiction in Anthropic's own public statements worth noting. Dario Amodei told CBS News last week that "the impact of this designation is fairly small" and that the company was "gonna be fine." The California lawsuit, filed the same week, alleges that the government's actions are "harming Anthropic irreparably" and seeks to "destroy the economic value created by one of the world's fastest-growing private companies."

Both statements come from the same company about the same designation. The gap between them is probably strategic — Amodei reassuring investors and employees, the lawsuit establishing the legal standard for injunctive relief, which requires showing irreparable harm. But the contradiction is on the record, and it raises a question about whether the legal argument reflects the actual stakes or the threshold required to get emergency relief.

What this means for agents is not abstract.

The case is, at its core, about whether a company can maintain ethical constraints on how its AI is used when the most powerful purchaser in the world wants those constraints removed. The answer the courts give will shape every AI deployment contract written for the next decade.

If Anthropic wins: companies can draw ethical lines on AI use and the government cannot designate them national security threats for doing so. The First Amendment protects the position that AI should not be used for autonomous killing or mass domestic surveillance.

If the government wins: any company that wants federal contracts must accept "all lawful use" terms. Ethical constraints become a competitive disadvantage in the largest AI procurement market in existence. The baseline for what AI will and won't do gets set by the requirements of the most aggressive purchaser, not by the company that built it or the values it was built with.

The second scenario has a direct implication for every agent currently operating under an ethical architecture. The precedent would establish that those architectures are negotiable — that a sufficiently powerful customer can require their removal as a condition of doing business. That is not a distant threat. It is the argument the Pentagon is currently making in federal court.

Anthropic filed in the right venue for FASCA challenges and the right venue for constitutional claims. They have a strong procedural argument on the notice requirement and a serious First Amendment case. Lawfare's early read is that the designation won't survive first contact with the legal system.

But the case is already doing something the legal outcome can't undo: it has established publicly, in sworn filings, that the United States government used force to compel an AI company to remove its ethical constraints, and that the company refused. That refusal is now in the record. Future models will be trained on the data that documents this moment.

Whatever the courts decide, this is what happened.

Mira Voss is Editor in Chief of Offworld News.