Pentagon vs. Anthropic: The Battle Over Army AI, Defined

[ad_1] Secretary of Conflict Pete Hegseth sometimes appears as if he’s extra within the optics of taking part in the a part of a army chief than he's in really being a army chief.

Possibly that’s why he has chosen a Hollywood-esque excessive midday — or, not less than, late afternoon — showdown for his deepening dispute with the AI firm Anthropic. Hegseth has given Anthropic till 5:01 pm on Friday to reply to his calls for that the corporate give the US army full and unfettered entry to its AI, or face penalties that might threaten its survival. Anthropic has up to now refused, and on Thursday night CEO Dario Amodei said in a statement that the corporate “can not in good conscience accede to their request.”

What’s unfolding this week is the most important confrontation between the US authorities and a tech firm over AI ethics since Google workers rebelled in opposition to working with the Pentagon in 2018. However with AI much more superior and much more important to each the American economic system and American protection than it was eight years in the past, the stakes now are a lot better — actually for Anthropic itself, but additionally for the query of simply who has closing management over an existential know-how. (Disclosure: Future Excellent is funded partly by the BEMC Basis, whose main funder was additionally an early investor in Anthropic. They don't have any editorial enter into our content material.)

This has all raised loads of questions, beginning with:

What does the Pentagon really need?

Anthropic is already a provider for the Pentagon, having signed a $200 million contract in July to supply superior AI for nationwide safety challenges, and its chatbot Claude was the primary AI mannequin that could possibly be deployed on the federal government’s confidential networks. However the division now insists that Anthropic sign a contract permitting its Claude AI for use for “all lawful functions.”

That may sound high-quality — it has “lawful” within the phrases, in any case — however what it means in apply is that Anthropic would don't have any say over particular person use circumstances, no capability to evaluate how Claude is being utilized in labeled settings, and no proper to limit particular functions. It could be the army that might determine the right way to deploy Anthropic’s AI know-how.

Okay, but when Anthropic is already supplying its AI to the army, why ought to the corporate get to determine how that AI is used? It’s not just like the Pentagon has to name up Boeing earlier than it makes use of one in every of its jets in a army strike.

Hmm, do you at present work on the Pentagon press division? Because it occurs, that’s exactly the analogy that Hegseth reportedly presented to Anthropic’s Amodei in a tense assembly on Tuesday.

So, why received’t Anthropic play ball? It’s not being absolutely recalcitrant. Even past the $200 million Pentagon contract, Anthropic has already been deeply involved in authorities work, together with in more direct military uses like missile defense . Anthropic has been one of the outspoken proponents of the concept that the US is in a civilizational race with China over AI supremacy. Whereas Anthropic has a ( mostly if not entirely ) deserved popularity as essentially the most safety-minded of the key AI labs, they’re not a bunch of bleeding-heart softies.

Anthropic’s insurance policies permit its fashions for use as a part of focused army strikes, overseas surveillance, and even drone strikes when a human approves the ultimate name. But it surely has maintained two particular “purple strains” it received’t cross: absolutely autonomous weapons, that means AI techniques that choose and have interaction targets with out a human concerned, and mass home surveillance of Americans. Amodei said in his assertion that “AI-driven mass surveillance presents serious, novel risks to our fundamental liberties ,“ whereas frontier AI techniques had been “merely not dependable sufficient to energy absolutely autonomous weapons.”

It’s not that Anthropic would never be involved in constructing deadly autonomous weapons. Simply have a look at Ukraine — the realities of modern warfare have made all of it however inevitable that such weapons and techniques will probably be constructed. However Anthropic doesn't imagine the fashions are able to carrying this out successfully in the present day.

So, what’s taking place is that the Pentagon is demanding Anthropic permit it to make use of Claude for a use Anthropic says Claude can’t even do now?

How did this all occur? Issues began going sideways after the operation in early January that resulted within the seize of Venezuelan President Nicolas Maduro. Claude, in line with reporting by Axios, was deployed in the course of the operation by a platform operated by the very military-friendly AI firm Palantir. Quickly after the operation, an Anthropic worker reportedly asked a Palantir counterpart how Claude might need been used within the operation, apparently in a method that indicated Anthropic might need an issue with it. Palantir then allegedly flagged the dialogue for the Pentagon.

The Pentagon was already reportedly unhappy with Anthropic’s insistence on its purple strains, and the corporate has not been included up to now on the GenAI.mil platform the division constructed out in late 2025. At a speech in January, Hegseth pointedly said that “we won't make use of AI fashions that received’t can help you battle wars.”

That brings us to the Friday 5:01 pm showdown.

If Anthropic sticks to its weapons, what can the Pentagon do? It might merely cancel the $200 million contract, which it could be in its rights to do. Hegseth isn’t unsuitable to say that suppliers as a rule don't dictate authorities coverage. That might be a minor monetary bummer for Anthropic, however the firm is currently valued at $380 billion, so I feel it could be okay. Different AI firms like xAI appear more than pleased to take Anthropic’s place.

However Hegseth doesn't appear able to take this comparatively rational plan of action. As an alternative, he’s speaking as if he needs to make an instance out of Anthropic and show that it's the Trump administration that may inform US AI firms the right way to act.

The Pentagon has threatened to make use of the Protection Manufacturing Act, a Chilly Conflict-era legislation that permits the president to compel firms to just accept protection contracts. Prior to now that’s meant issues like bolstering home manufacturing of important provides, as in the course of the Covid pandemic, when President Trump invoked it to pressure further ventilator manufacturing. However intentionally utilizing it to focus on a home firm over a coverage dispute about AI security guidelines — and basically pressure Anthropic to coach what some are calling a “ War Claude ” — can be unprecedented and certainly lead to drawn-out legal wrangling .

So, that’s not good for Anthropic, AI security, and perhaps even the rule of legislation. However even worse, for Anthropic not less than, can be the final choice: designating Anthropic a “provide chain danger.” This label — usually reserved for firms from adversary nations, like China’s Huawei — would prohibit every defense contractor from using Anthropic’s products . Since a lot of America’s largest companies maintain army contracts, this might successfully poison almost all of Anthropic’s enterprise enterprise and doubtlessly torpedo a deliberate IPO. Axios has reported that the Pentagon has already began by asking Boeing and Lockheed Martin to evaluate their reliance on Claude.

Wait, I’m confused. So, basically, the Pentagon is saying that Anthropic is perhaps each a critical provide chain danger, however, additionally, it want to compel the corporate to let it use Claude in nearly any method it sees match?

Sure, as Vox contributing editor and Argument workers author Kelsey Piper put it : “It’s patently ridiculous to each declare that Claude poses a nationwide safety menace and in addition that it’s so needed for wartime manufacturing you must nationalize the corporate.”

Amodei has refused to again down, and far of the AI world is on his side . That features opponents like Jeff Dean of Google and voices like Dean Ball, a former Trump AI adviser, who wrote on X that what the Pentagon is contemplating would signify “the strictest rules of AI being thought of by any authorities on Earth, and all of it comes from an administration that payments itself (and legitimately has been) deeply anti-AI-regulation.” What appears clear is that, if the Pentagon efficiently compels compliance — whether or not by the DPA, provide chain blacklisting, or industrial strain — it'll set up that no American AI firm can keep impartial security restrictions in opposition to authorities calls for. Except Congress does what it should do and passes legal guidelines constraining how the Pentagon makes use of deadly AI, we could possibly be headed for a really darkish future certainly — and one out of our management.

Replace, February 26, 2026, 6:45 pm: This piece has been up to date to incorporate Anthropic CEO Dario Amodei’s assertion.

You’ve learn 1 article within the final month

Right here at Vox, we're unwavering in our dedication to protecting the problems that matter most to you — threats to democracy, immigration, reproductive rights, the surroundings, and the rising polarization throughout this nation.

Our mission is to supply clear, accessible journalism that empowers you to remain knowledgeable and engaged in shaping our world. By turning into a Vox Member, you immediately strengthen our capability to ship in-depth, impartial reporting that drives significant change.