Anthropic is embroiled in a standoff with Secretary Hegseth over how the company's AI offerings may be used in military applications:

For months, [Anthropic CEO Dario] Amodei has insisted that using AI for domestic mass surveillance and AI-controlled weapons are ethical lines the company will not cross, calling such use "illegitimate" and "prone to abuse." According to a source familiar with the Hegseth meeting, Amodeo stressed those positions again on Tuesday.
 
Hegseth has said Anthropic needs to allow the U.S. to use its AI in all "lawful" purposes, which could include AI-directed warfare and surveillance.

How this all gets hashed out will have important implications for the relationship between the US government and frontier AI companies. I expect many media outlets and commentators to offer views on a plethora of issues related to this clash, but here I want to address a specific consideration that might or might not eventually be relevant. Does the First Amendment protect Anthropic from what Hegseth would like to do here? Below I lay out the high level argument that Anthropic could make for this, as well as identifying some factors about this specific situation that I think weigh either for or against that argument.

The argument in brief

Similar to the argument presented by social media companies in Moody v. NetChoice, Anthropic could argue that any safeguards it applies to its AI systems meant to prevent mass surveillance or AI-controlled weapons constitute expressive activities protected by the First Amendment. If their safeguards were covered by the First Amendment, Anthropic could then argue that for Hegseth to make good on his threats would constitute retaliation for protected speech. Requiring Anthropic to remove safeguards or otherwise allow uses it doesn't wish to allow might also constitute compelled speech under this analysis.

I don't necessarily expect this to even come up in this specific case. It could come out tomorrow that Hegseth and Anthropic have reached an agreement privately and the matter is resolved. But I think these issues at the intersection of AI and the First Amendment will come up eventually, and on the off chance they do here, I wanted to preregister some of my thoughts on the topic.

Factors in favor

• Unlike in Moody, there isn't a countervailing speech interest on the part of social media users. This situation would be Anthropic vs the US government (although in other situations I think a similar consideration would apply to AI companies and the interests of their users).
• Anthropic can assert a very clear viewpoint and how the actions of the government are directed at opposing this viewpoint. This would assist Anthropic in arguing that their actions are "expressive".
• Assuming the reporting on this situation is correct, the existence of the threat by Hegseth seems pretty clear. This would help Anthropic argue retaliation.
• The timing and optics might be favorable for Anthropic with Hegseth and his department being involved in First Amendment litigation involving Senator Mark Kelly. Courts might be concerned whether this is an emerging pattern with Hegseth and be willing to look at debatable First Amendment arguments more closely.

Factors against

• This is occurring in a national security context, where Hegseth and the government are going to get a lot of deference.
• The government is likely also going to get a lot of deference on the choice of what vendor to use. The more that the government can emphasize that part of the situation the more it benefits them I think.
• In her concurrence in Moody, Justice Barrett seemed to express skepticism about how the use of AI impacts the expressiveness for a companies conduct, raising question about whether this being an AI product will make the First Amendment argument more challenging.
• The government could argue that Anthropic's actions are best understood as just that, actions, rather than speech or expressive conduct. The fact that the government is acting as a procurer of services in this context I think helps this argument.