Isaac Barnes May
 
Anthropic’s case against the government has a religious dimension. Anthropic filed suit against the federal government after the government’s threat to declare it a supply chain risk when the company objected to the use of its products in autonomous warfare and mass surveillance of Americans. Anthropic presented the government’s actions as coercion under the First Amendment. The case recently saw Judge Rita Lin issue a preliminary injunction against the government, noting this “appears to be classic First Amendment retaliation.”
 
Yet the case resembles not just prior cases about free speech, which Anthropic and the judge invoked, but also those on religion. When the rupture between the Pentagon and the company first became public, Anthropic’s CEO Dario Amodei released a statement declaring that the company “cannot in good conscience accede to their request.” Amodei’s invocation of conscience as core to Anthropic’s stand positioned the company as a kind of corporate conscientious objector. As such, it may be protected as religion under the Religious Freedom Restoration Act (RFRA).
 
A recent amicus filing in the Anthropic case by a group of Roman Catholic Moral Theologians and Ethicists hints at the broader religious implications of the current case, arguing that Anthropic’s position has correspondences with Catholic moral teaching on surveillance and the use of AI weapons. They rely on Catholic Just War Theory to argue, for example, that AI controlled autonomous weapons “by definition fails to meet the conditions for jus in bello required for acts of war to be morally licit in Catholic thought.” By refraining from working on these weapons, they argue Anthropic is “acting as a responsible and moral corporate citizen.” The company, the brief implies, is exercising conscience.
 
Over a decade ago, the Supreme Court in Burwell v. Hobby Lobby held that a for-profit corporation was protected by RFRA in exercising religion. Hobby Lobby, a craft store chain, could not be forced to provide health insurance coverage for contraceptives because of its owners’ religious objections. Justice Alito, in the majority opinion, even made clear that business practices “compelled or limited by the tenets of religious doctrine” were examples of religious exercise under RFRA.  Legal academics worried that there would be a spate of corporate religious liberty claims, though relatively few cases of corporations using RFRA occurred. If there is a sincere religious objection to providing the Department of War with military AI, the government’s actions against Anthropic would be subject to strict scrutiny.  
 
There is the issue of whether Anthropic’s conscience claims are “religious” under the law. Legal notions of religion are broad; they do not require a theistic belief, and they can cover practices like ethical vegetarianism or objection to vaccination, which are not tied to comprehensive metaphysical systems. I have argued elsewhere that AI-based beliefs, common with AI companies, seem to fit legal definitions of religion. Anthropic, which has ties to the Effective Altruist movement, is perhaps the most religious seeming of the large AI companies. It is a public benefit corporation which claims to “Act for the global good” while prioritizing AI safety. Anthropic created a constitution to guide the values of its AI model, which has at least some commonalities with a religious doctrine, devoting considerable time to its LLM Claude’s relationship to virtue and ethics. There seems little reason to doubt that Anthropic’s professed concern about killing civilians is a sincere ethical belief.

Dario Amodei explicitly cites conscience as the reason for not being willing to work on contracts with the Department of War involving mass domestic surveillance or fully autonomous weapons. During the Vietnam War, conscientious objector cases United States v. Seeger and Welsh v. United States found that conscience claims of refusal to render military service were “religious” for the purpose of making someone a “religious” conscientious objector to war. Seeger was not a traditional theist, while Welsh crossed the word religion off his draft form; both were understood to be religious by the Supreme Court. While those cases interpreted “religion” within the language of statute, they also have broader implications that claims of conscientious objection did not have to be rooted in established religious traditions.

Anthropic’s case in some key ways resembles Thomas v. Review Board, where a Jehovah’s Witness filed for unemployment after leaving a factory job where he was assigned to make tank turrets when he felt he could not in good conscience help produce weapons. Though this kind of work was not condemned by all Jehovah’s Witnesses, the Supreme Court found that his objection to involvement in producing war material was religious in nature. The act of refusing to take part in producing weapons because of ethical objections to taking human life, the demand to not be involved in killing, might be inherently religious in nature.
 
The biggest difference, other than Anthropic is a corporation, is the fact that Anthropic does not object to AI weapons in all circumstances. Anthropic has stated that it might not oppose autonomous weapons if it felt they were reliable enough not to endanger civilians and U.S. soldiers. While this might appear to undermine Anthropic’s ethical stand, these sorts of moral arguments about a kind of weapons technology have been common in ethical debates about warfare. The Jesuit theologian John Ford in 1944 for instance drew a useful distinction between precision bombing, which he believed could be morally undertaken, and obliteration bombing of the kind undertaken on cities in Japan and Europe, which he classified as an “immoral attack on the rights of the innocent.” If the technology does not allow moral use in its current form, Anthropic’s objection is still an ethical one.
 
Anthropic’s objection to only certain kinds of AI use in warfare is a kind of selective conscientious objection. When the U.S. had a draft, courts were not supportive of selective conscientious objection in Gillette v. United States when claimants were not opposed to all war.  Yet for the purposes of RFRA, this does not matter so long as the objection to the government’s burdening of their beliefs is sincerely religious. Anthropic explains why it objects to using AI in autonomous weapons and to surveilling Americans, citing the technology’s great potential for harms. Even if Anthropic’s explanation is not perfect considering their past contracts with the military or their belief that autonomous AI weapons could one day be developed, it is well established since Thomas that religious liberty claims do not have to be internally consistent.  
 
There are reasons why Anthropic might not opt to use RFRA to defend its ability to not develop AI weapons. RFRA would not salvage a frayed relationship with the Department of War and would risk future contracts. There would certainly be reputational costs for an AI company arguing that it was religious, which might hurt its public reputation and cause it to be seen as odd or even cult-like. Further, other legal avenues exist for Anthropic, such as compelled speech and expression, and they seem to be working effectively now. Those opposed to the government might worry about the expansion of corporate conscience rights, even if this case is sympathetic.
 
Yet as a matter of law, Anthropic has a religious liberty claim that could shield it from federal coercion. This claim is just as strong as any involving speech. An AI company refusing for reasons of conscience to make a weapon is no less obviously religious than a craft store like Hobby Lobby refusing to provide employees with contraceptive coverage. 
 
Isaac Barnes May is a Resident Fellow at the Information Society Project at Yale Law School. He is the author of two books, including American Quaker Resistance to War, 1917–1973: Law, Politics, and Conscience.You can reach him by e-mail at isaac.may@yale.edu.