Is the Government Punishing Anthropic for Speech?

On Monday, Anthropic sued the federal government in federal district court in California, alleging that the various punitive actions taken by Defense Secretary Pete Hegseth and President Donald Trump are unlawful. The complaint states five causes of action. In addition, in a separate "petition for review" filed in the DC Circuit, Anthropic challenged its designation as a supply-chain risk. The company needed to file two separate actions because the DC Circuit has exclusive original jurisdiction to review supply-chain risk designations but, as an appeals court, generally lacks jurisdiction for other federal question cases. In today's essay, I'm going to discuss just one of the claims in the federal district court complaint. I express no opinion about the other claims in that complaint or about the supply-chain risk complaint.

I want to raise a question about Anthropic's contention that the government is retaliating against it for free speech. Recall that Anthropic doesn't want its products used for mass surveillance or in autonomous weapons but that the Defense Department sought to change those terms to allow any lawful use. The hammer of the supply-chain risk designation is pretty obviously designed to punish Anthropic, but is it designed to punish Anthropic for speech? To answer that question, I'll consider a hypothetical one.

Suppose that a company I'll call Humane Traps (HT) sells traps to the National Parks Service (NPS) for humanely capturing wolves, coyotes, and other animals so that they may be relocated in order to increase their population in ecosystems in which they previously were plentiful but have been diminished or eliminated due to hunting and habitat encroachment. The terms of service of the contract specify that relocation of the animals is the only use to which the traps may be put. Now suppose that the NPS notifies HT that it wishes to modify the terms of service to allow any lawful use. HT has reason to believe that the NPS wishes to use its traps to capture and then kill wolves and coyotes to serve the interests of ranchers who graze their cattle on land adjoining some national parks. This use would violate the principles of HT, which is dedicated to conservation. HT resists the change, and then the federal government imposes various penalties. There's no doubt that the government will have retaliated, but it will not have retaliated for HT's speech. It will have retaliated for HT's conduct--or more precisely, for HT's refusal to participate in and thus be complicit in what it regards as wrongful conduct.

Put differently, refusal to participate in some action out of some principled stance is invariably motivated by views about that action, and it is often accompanied by speech condemning the action, but the refusal itself is not speech. That's why, when religiously-scrupled companies like Hobby Lobby or religious orders like the Little Sisters of the Poor sought exemptions from obligations related to the provision of health insurance that included coverage for forms of contraception that they regarded as tantamount to abortion, they relied chiefly on the Religious Freedom Restoration Act, not freedom of speech.

Subject to two important caveats that I'll raise momentarily, I don't see Anthropic's case as fundamentally different. Anthropic, like HT in my hypothetical, objects to some uses of its products. It does not want to be complicit in mass surveillance or the use of autonomous weapons for which its systems are not yet ready. That's a principled, even noble, stance. It is accompanied by expressions of principle. But the refusal itself is not speech.

Now come the caveats.

First, if a person or firm objects to being required to create bespoke expression, then that does raise a free speech claim pursuant to 303 Creative LLC v. Elenis. Although the Anthropic complaint doesn't make such a claim, arguably it could do so because large language models like Claude are not physical machines in the way that animal traps are; they are in a sense built out of expression--including Claude's constitution. I don't know enough about how Anthropic's custom versions of Claude or other products work to say whether such a claim would or should be successful. Thus, I won't venture an opinion on the matter. In the next week or so, I hope to post a guest essay on this subject by someone more knowledgeable than I about how large language models and other forms of AI work. Stay tuned!

Second, I want to be clear that even if (as I've argued above), Anthropic's refusal to agree to terms that would allow its products to be used for mass surveillance or autonomous weapons is not speech, its retaliation claim could still succeed based on a showing that the government is not simply punishing Anthropic for its refusal to engage in objectionable conduct if there is evidence that the government's subjective motive in punishing Anthropic is to retaliate for the latter's speech in other respects. And indeed, there is such evidence. The complaint quotes President Trump's description of Anthropic as a "WOKE COMPANY" and Secretary Hegseth's denunciation of the company's "sanctimonious rhetoric" and "corporate virtue-signaling." Those are essentially admissions that at least part of what irks the administration about Anthropic is what the company expresses. Here, as in other contexts, the Trump administration is remarkably transparent in its illicit motives.

Finally, in saying that retaliation for complicity-avoiding conduct is not, ipso facto, retaliation for speech, I am addressing only the legal question whether it gives rise to a First Amendment claim. Needless to say, but I'll say anyway, as a matter of policy and basic decency, government should not retaliate against a firm for taking a principled stance. At most, it should simply find alternative suppliers with fewer (or more charitably, different) scruples. (Whether it has found such a supplier in OpenAI is not entirely clear.)

-- Michael C. Dorf