The Pentagon challenges AI ethics

TL;DR: Over the long weekend, the Pentagon reportedly threatened to cut off Anthropic after it refused to allow the use of Claude for mass surveillance of Americans or fully autonomous weaponry—it’s the only AI model currently available in the military’s classified systems. Defense Secretary Pete Hegseth is apparently close to branding Anthropic a “supply chain risk,” which would require contractors to ditch Claude. The fight is an early look at how the US government might strong-arm AI labs into abandoning their own ethical standards and whether those labs will resist.

What happened: It's one thing to beef with a rival AI company, and another to lock horns with a government department that controls the largest military budget on Earth. Anthropic is reportedly now in a fast-escalating standoff with the Pentagon over where national security ends and company-imposed ethical guardrails begin. Axios reported that Hegseth is close to cutting Anthropic off entirely, and even designating it a “supply chain risk,” a move usually reserved for foreign adversaries, not a domestic tech company.

In recent months, the Pentagon has pushed Anthropic, OpenAI, Google, and xAI—all of whose models it uses in some capacity—to loosen restrictions for AI models so they can be used for “all lawful purposes,” including weapons development, intelligence collection, and battlefield operations with essentially no company-imposed limits. But Anthropic has put its foot down on two red lines: no mass surveillance of Americans and no fully autonomous weaponry. After months of back-and-forths, the Pentagon is now threatening to sever all ties with the AI company. "It will be an enormous pain in the ass to disentangle, and we are going to make sure they pay a price for forcing our hand like this," a senior official told Axios.

Per another Axios report, tensions rose after the military used Claude in its January operation to capture former Venezuelan President Nicolás Maduro. According to a senior administration official, an Anthropic exec asked Palantir about Claude’s use; the Pentagon read that as implied disapproval. Anthropic denies this, saying it hasn’t commented on specific operations.

The Pentagon partnership: Anthropic signed a contract worth up to $200 million with the Department of Defense last July, making Claude the first frontier AI model used on the Pentagon's classified networks. Claude Gov, the version of the AI model tailored to national security work, has received high praise internally, with an official telling Axios that other models "are just behind." But OpenAI, Google, and xAI have all reportedly shown more openness to the Pentagon's demands, and the department is confident all three will agree to their ”all lawful purposes" language, a senior official told Axios.

Why it matters: A "supply chain risk" designation would require Pentagon contractors to certify that they don’t use Claude in their workflows. The penalty sends a strong message: Imposing your own ethical limits on military use could carry severe economic consequences.

This standoff could also be a defining test for the AI industry. Anthropic's CEO Dario Amodei has argued AI should support national defense "in all ways except those which would make us more like our autocratic adversaries." One Pentagon insider described Anthropic as the most "ideological" of the AI labs. Internally, the pressure is showing, too—last week, Anthropic's head of safeguards research resigned, warning that "the world is in peril."

What's next: While the Pentagon is threatening to hit Anthropic where it hurts, the AI company says it's having "productive conversations, in good faith" with the DoD. —WK