Expressive Governance Is a First Amendment Threat Hiding in Plain Sight

When one of the United States’ leading artificial-intelligence companies told the federal government that its models could not be used safely for autonomous lethal weapons or the mass surveillance of US citizens, the government did not simply find another vendor. It branded the company a “supply-chain risk to national security”—a label built for foreign adversaries—and moved to cut off its ability to do business across the entire federal defense ecosystem. The President then ordered every federal agency to stop using its technology.

That is not a procurement dispute. It is punishment for protected speech, dressed in the ordinary clothing of administrative discretion. In a forthcoming book and in an amicus brief I filed in Anthropic v. US Department of War, now before a federal court in San Francisco, I call this phenomenon “expressive governance”: the executive’s use of routine administrative tools—funding, contracting, licensing, personnel decisions, regulatory designations—to penalize disfavored viewpoints while preserving the appearance of business as usual.

The danger of expressive governance is precisely that it looks lawful. Courts have long, and sensibly, deferred to the executive’s judgment about who receives a contract or a clearance. But that deference was built for genuine managerial choices, not for retaliation. When the government turns these low-visibility levers against a speaker, the constitutional violation can hide in the gap between what officials say and what they actually do.

And this company is not alone. The same template has been used against law firms stripped of access for representing disfavored clients, against universities threatened with the loss of billions in funding over their programs and their faculty’s speech, against media organizations menaced with regulatory reprisal, and against civil servants fired for signing dissent letters. Courts have already found several of these actions unconstitutional. In each, the mechanism was identical: facially neutral authority deployed to inflict real economic and reputational harm on those who would not fall into line.

The cumulative effect is the deeper injury. Every company, firm, and institution that depends on a federal contract, grant, or approval now absorbs a simple lesson: take a principled public stand the administration dislikes, and you may be next. As Luke Barnes observed of this case in a piece for the NYU Stern Center for Business and Human Rights, the message to an entire industry becomes that responsibility is a liability. That chill—the speech never spoken, the position never taken—is the structural harm the First Amendment exists to prevent.

The harder question is what courts should do about it. Striking down one order at a time accomplishes little when the tactic can simply reappear in a new administrative form. The amicus brief I filed in the Anthropic casebrief proposes a framework that gives judges a repeatable method rather than a one-off result. It rests on three tools.

First, the government should have to point to a law that actually authorizes what it is doing—a clear-statement rule. Before the executive may wield a discretionary power in a way that burdens speech, courts should require that Congress clearly authorized that use. A designation designed for foreign saboteurs cannot be stretched to punish a US company’s policy views absent a specific directive from Congress. None exists here.

Second, the burden of explanation should shift to the government. Once a plaintiff shows both a viewpoint-linked pattern—suspicious timing, targeting, revealing public statements—and a dependency the government can exploit, such as a contract or a license, the government should have to prove that it would have taken the same action for legitimate reasons. If the real explanation is the speech, it should not be able to hide that behind a routine-sounding label.

Third, the presumption applies with greatest force when the government targets what I call an "expressive intermediary." I don't use that term in the Section 230 sense of a platform hosting others' content; I mean an entity whose own expressive choices shape how a broad public finds, analyzes, and communicates information, so that punishing it distorts public debate well beyond the individual target. Large language models are a paradigmatic example: an AI provider's publicly stated safety principles are its own speech, and its models increasingly mediate how millions of people and institutions process information. When the government moves against such an entity, courts should presume the action distorts public debate; generalized invocations of "national security" should not suffice, and the government must instead offer concrete, viewpoint-neutral, individualized reasons.

This framework does not invent a new tier of scrutiny, and it does not handcuff the executive. The government remains free to disagree with a company, reject its terms, and take its business elsewhere. What it may not do is weaponize the machinery of national security to destroy a business for holding a view on a matter of public importance. The difference between those two things is exactly what a court, armed with the right method, is equipped to detect.

The stakes reach well past one company or one administration. If expressive governance succeeds, the price of conscience becomes ruin, and the space for honest public debate quietly contracts. Courts are the institution positioned to say no—and they should.