Last updated: March 2026

AI Safety vs Military Use

What happens when the Pentagon tells an AI company to remove safety restrictions, and the company says no?

According to the public reporting available in late February and early March 2026, that is effectively what happened on February 27. Dario Amodei reportedly had a $200 million contract on the table. The dispute centered on two restrictions in Claude’s terms of service: no mass domestic surveillance and no fully autonomous weapons. He refused, and the situation escalated quickly.

The Ultimatum

The Pentagon gave Anthropic a deadline: Friday, February 27, 5:01 PM ET. Drop the restrictions or lose the contract. Defense Secretary Pete Hegseth had already summoned Amodei to make the demand in person. The Pentagon’s position was straightforward: AI companies working with the military must allow their models to be used “for all lawful purposes.” No carve-outs. No exceptions.

Hegseth backed the demand with two threats. First, designate Anthropic a “supply chain risk,” which would blacklist the company from all military contracts and force every defense contractor to cut ties with them. Second, invoke the Defense Production Act (a Korean War-era law) to compel Anthropic to comply.

Amodei pointed out the contradiction in a public statement: “These latter two threats are inherently contradictory: one labels us a security risk; the other labels Claude as essential to national security.”

He didn’t budge. “We cannot in good conscience accede to their request.”

The Political Escalation

An hour before the deadline, Trump posted on Truth Social, calling Anthropic “Leftwing nut jobs” who made a “DISASTROUS MISTAKE trying to STRONG-ARM the Department of War.” He ordered every federal agency to immediately stop using Anthropic’s technology, with a six-month phaseout period.

After the deadline passed, Hegseth followed through on the supply chain risk designation. His statement on X accused Anthropic of delivering “a master class in arrogance and betrayal” and trying to “seize veto power over the operational decisions of the United States military.”

“America’s warfighters will never be held hostage by the ideological whims of Big Tech. This decision is final,” he wrote.

The $200 million contract was dead. And every federal agency running Claude had six months to find alternatives.

Anthropic’s Response: See You in Court

Anthropic didn’t fold. They announced they would challenge the supply chain risk designation in court, calling it “legally unsound” and a “dangerous precedent for any American company that negotiates with the government.”

The company also pushed back on Hegseth’s claim that all military contractors would have to cut ties with Anthropic entirely. Under federal law, they argued, the designation only applies to Claude’s use in Department of War contracts, not how contractors use Claude for other customers.

Anthropic framed its position around two arguments. First, current AI models aren’t reliable enough for fully autonomous weapons. Deploying them that way would endanger American soldiers and civilians. Second, mass domestic surveillance of Americans violates fundamental rights.

“To the best of our knowledge, these exceptions have not affected a single government mission to date,” the company said.

Silicon Valley Picks a Side

Here’s where the story takes a turn nobody expected.

Within hours, 636 employees at OpenAI and Google signed an open letter supporting Anthropic. These are employees at Anthropic’s direct competitors, publicly backing a rival against the US government.

The letter was blunt: “They’re trying to divide each company with fear that the other will give in. That strategy only works if none of us know where the others stand.”

The signatories called on their own company leaders to maintain the same red lines Anthropic was defending: no mass surveillance, no fully autonomous weapons.

Then the executives started talking.

Sam Altman told CNBC he didn’t “personally think the Pentagon should be threatening DPA against these companies.” An OpenAI spokesperson confirmed the company shares Anthropic’s red lines. Hours later, Altman announced OpenAI had signed its own Pentagon deal, but with the same safeguards Anthropic had demanded. “The DoW agrees with these principles, reflects them in law and policy, and we put them into our agreement,” he wrote on X.

Google’s Chief Scientist Jeff Dean weighed in: “Mass surveillance violates the Fourth Amendment and has a chilling effect on freedom of expression.”

Ilya Sutskever, who had been largely silent since leaving OpenAI, resurfaced to publicly praise both Anthropic and OpenAI for not backing down.

Apple’s MLX creator Awni Hannun put it simply: “Saying it is easy. Doing it is hard, especially when you have to pay a huge price for it.”

The Twist: Claude Was Already in the War

In a development that undercuts the entire standoff, The Guardian reported on March 1 that US Central Command had used Anthropic’s Claude for intelligence assessments during air strikes on Iran, hours after Trump’s ban order. The military was apparently still running Claude in operational systems even as the political fight played out.

This detail suggests the Pentagon still found Claude valuable enough to keep using it in operational contexts, even while publicly blacklisting the company that made it.

What This Actually Means

The $200 million contract is not the main point. Anthropic has already scaled into a much larger business, and the broader issue here is the precedent around government pressure and model-policy red lines.

The real stakes are about precedent. Can the US government force a private company to remove safety restrictions from its AI products? Can a tech company maintain ethical red lines when the Pentagon is the customer?

Musk’s xAI has already moved to fill the gap, signing an agreement to let the military use Grok in classified systems with extremely broad terms. But as multiple analysts have noted, Grok’s capabilities don’t match Claude’s for the Pentagon’s core needs.

OpenAI’s approach is the most interesting play. Altman managed to sign a Pentagon deal while maintaining the exact same restrictions Anthropic was punished for. Whether that deal survives scrutiny from the same administration that just blacklisted Anthropic for identical terms remains to be seen. (For more on how these companies compare on the technical side, see our ChatGPT vs Claude vs Gemini breakdown.)

What Sticks

Forget the contract dispute. The lasting significance is simpler: for the first time, competing AI companies stood together on a safety issue. Not in a joint press release. Not in an industry pledge nobody enforces. Actual employees at rival companies, putting their names on a letter, telling their own employers to hold the line.

That may be the most durable precedent from this episode, and it likely matters more than the contract value itself.

Related guide: OpenAI’s $110B funding round and what it signals for the AI market.