Lindsey Choo and Ella Markianos, writing at Platformer:

Defense secretary Pete Hegseth reportedly gave Anthropic a deadline of Friday evening to give the US military full access to its AI model or face essentially unprecedented penalties, escalating a weeks-long battle over AI safeguards.

If Anthropic doesn’t give in, the Pentagon plans declare the company a “supply chain risk,” making it harder for Anthropic to sign deals with military contractors, Axios reported. The Pentagon may also invoke the Defense Production Act to force the company to change its model to fit the military’s needs, Hegseth told Anthropic CEO Dario Amodei in a meeting Tuesday, sources told the outlet.

As Choo and Markianos write in their (excellent) “Following” section of Platformer, the Defense Department wants an un-neutered version of Claude to use without restriction, without any guardrails. Anthropic is the only major artificial intelligence lab that does not provide the military with access to unrestrained models for “lawful” military activity — OpenAI and xAI, Elon Musk’s company, both provide the government with versions that will not refuse requests. If a normal user asks ChatGPT to help plan a military attack, the model will obviously refuse, as it is against the company’s terms of use, but OpenAI makes a version exclusively for the military. The problem is that no other model comes close to Claude for military use, according to the Defense Department, and Hegseth wants unrestrained access to it.

This clashes with Anthropic’s modus operandi: building safe models more prone to refusals and “alignment,” a term the industry uses to mean that the model works in humans’ best interests. These ideals clash with the Trump administration’s, which wishes to deploy the military against American citizens and shoot them in broad daylight, amid many other heinous crimes against humanity. If Claude were asked to assist with any of these operations, it would refuse, aligned with the “Claude Constitution.” Anthropic is obsessive about feeding Claude morals to the point where the company treats the model as if it were a human child, teaching it good and bad in the hopes that it might carry those teachings on in conversations with users. Anthropic models are notoriously good and safe.

The Trump administration isn’t about to back down to a left-leaning AI company, however, because authoritarianism doesn’t abide by rules or exceptions. So, in keeping with the regime’s approach toward regulation, it wants to force Anthropic to give it an immoral version of Claude through the Defense Production Act, a law passed during the Korean War. The law’s third provision states that it “authorizes the President to control the civilian economy so that scarce and critical materials necessary to the national defense effort are available for defense needs.” If the president, under the advice of his defense secretary, were to invoke this law, the administration would entirely assume control over Claude’s development.

The law is very clear that it authorizes the government to take control of the civilian economy as long as it deems such control necessary. It does not draw a distinction between civilian and military products, i.e., this law would allow the Defense Department to control all of Claude’s development, including the civilian models. Of course Anthropic would sue the administration if this were to happen, but that’s irrelevant. The regime wants control over a civilian company because it did not bow to its requests, nay, demands. And it is using the military — which is not engaged in any armed conflict — as a pretext to control some of the most powerful technology in American history, a technology used to build software at the world’s largest corporations. This is authoritarianism.

I sincerely applaud Anthropic for adhering to its moral compass, even if it may bring untoward consequences. This is a class act, heroic example of the united opposition Silicon Valley must bring to Washington. It’s too bad OpenAI, whose president was the single largest donor to the Trump campaign, lacks these morals.