The Pentagon vs Anthropic: Why a tech giant is defying the US military on use of AI

anthropical, single of si vale’s leading unreal intelligence firms, is locked in a standoff with the Pentagon over how far powerful AI systems could be used for war and surveillance.

The dispute centers on Anthropic’s Claude chatbot, which has been running on US military classified networks and was reportedly used in planning the operation to capture Venezuelan President Nicolas Maduro. The Department of War has given the company until 5:01pm Eastern Time (22:01 GMT) on Friday to accept demands to lift key safeguards or risk being blacklisted as a “supply chain risk.”

Claude is deeply embedded in US defense workflows. The company says its models are already used across national security agencies for intelligence analysis, simulations, operational planning, cyber operations and other “mission‑critical” tasks. Anthropic became the first AI firm to deploy systems on the Pentagon’s classified networks, signing a contract worth up to $200 million with the Department of War last summer.

Other major AI providers have so far only reached deals to run their models on the military’s unclassified systems, putting Claude in a privileged position inside the US defense establishment.

Within Anthropic’s acceptable‑use policy for the Department of War are explicit bans on using Claude for mass domestic surveillance and for fully autonomous weapons. Those contractual safeguards reflect the company’s internal rules.

The Pentagon has demanded those limits be scrapped. Officials say they want to be able to use the system for “all lawful purposes” and according to US media, have pressed the firm to provide a “clean” version of the model stripped of moral and ethical constraints.

“You can’t lead tactical operations by exception,” an unnamed Pentagon official was quoted as saying by CNN, insisting that “legality is the Pentagon’s responsibility as the end user.” The military argues it cannot afford to be in a crisis and have to ask a private contractor for permission to remove guardrails.

US Secretary of War Pete Hegseth, who met Anthropic CEO Dario Amodei this week, has publicly complained that the Pentagon does not need neural networks “that can’t fight” and has threatened to designate Anthropic a “supply chain risk” – a label usually reserved for firms seen as extensions of foreign adversaries.

Pentagon spokesman Sean Parnell claimed the military “has no interest in using AI to conduct mass surveillance of Americans (which is illegal) nor do we want to use AI to develop autonomous weapons that operate without human involvement,” but stressed: “We will not let ANY company dictate the terms regarding how we make operational decisions.”

Anthropic says it is willing to keep working with US national security agencies, but will not drop two core restrictions on how its systems are used.

“Threats do not change our position: we cannot in good conscience accede to their request,” Amodei said in a statement on Thursday, adding that the Pentagon’s demands “have never been included in our contracts… and we believe they should not be included now.”

The company set two clear red lines for its AI, declaring it will not support mass domestic surveillance or fully autonomous weapons. It argues that large‑scale monitoring of Americans is “incompatible with democratic values” and that today’s models are “not reliable enough” to make lethal decisions without human control.

Amodei insists these carve‑outs have not prevented the US military from using Claude for other “mission‑critical” tasks and says the firm still wants to support US national security – but not at the cost of enabling mass surveillance at home or fully autonomous killing.

Amodei says the Department of War has warned that, if Anthropic keeps its safeguards, it could be removed from military systems and declared the aforementioned “supply chain risk” – a designation never before applied to an American firm.

Losing a contract worth up to $200 million would not be existential for Anthropic, valued at nearly $400 billion, but such a label could hit much harder. Any company doing business with the Pentagon would have to prove that its own systems do not rely on Anthropic’s technology, potentially complicating or chilling large enterprise deals with firms that also supply the US military.

For the Department of War, cutting ties would also be costly. Officials would have to replace internal tools built around Claude. One Pentagon source told US media that Elon Musk’s Grok AI system is “on board with being used in a classified setting,” but acknowledged that Grok is not regarded as being as advanced as Anthropic’s model.

The developer’s stance has triggered an unusual wave of public support inside Silicon Valley. Late on Thursday, hundreds of current employees at Google and OpenAI – two of Anthropic’s main rivals, both of which also supply AI models to the US military – signed an open letter backing the company’s refusal to comply with the Pentagon’s demands.

The petition, titled ‘We Will Not Be Divided’, had been publicly signed by 421 Google and 76 OpenAI staffers as of Friday. Citing a US media report, the letter accuses the Department of War of targeting Anthropic for “sticking to their red lines to not allow their models to be used for domestic mass surveillance and autonomously killing people without human oversight.”

“The Pentagon is negotiating with Google and OpenAI to try to get them to agree to what Anthropic has refused,” the signatories wrote, alleging that officials are trying “to divide each company with fear that the other will give in.” The letter calls on the two firms’ leadership to “put aside their differences and stand together to continue to refuse” the Defense Department’s demands.

The clash between Anthropic and the Pentagon has drawn interest from technology and defense analysts, who warn it may set precedents for how powerful AI is governed in any future conflicts. Adam Connor, vice president for technology policy at the Center for American Progress, told US media the dispute is likely to be read across the industry as a signal that defense officials do not want contractual limits on how military users can deploy advanced models.

If the Pentagon follows through on its threats, observers say it would mark a historic escalation – effectively turning one of America’s most advanced commercial AI products into a pariah inside its own defense ecosystem. Gregory Allen, a senior adviser at the Center for Strategic and International Studies, argued that treating Anthropic this way would be akin to burning one of the US tech sector’s “crown jewels” at a time when Washington is comparing the AI race with China to the space race with the Soviet Union. He suggested there are better ways to resolve the dispute than the “absolutist” stance the Trump administration has taken.

---

---