Pentagon weighs ‘supply chain risk’ label for Anthropic amid AI ethics standoff

Defense Secretary Pete Hegseth is “nearing” a decision to sever ties with Anthropic and designate the artificial intelligence firm as a “supply chain risk,” signaling a sharp escalation in tensions between the Department of Defense and one of the world’s leading AI developers.

A senior Pentagon official, speaking to Axios, indicated that such a designation would mandate that any entity seeking to conduct business with the US military must terminate its own relationship with the company.

“Untangling this relationship will be arduous, but we intend to ensure they pay a price for applying this kind of leverage against us,” the official stated, noting that such punitive measures are typically reserved for foreign adversaries.

Pentagon spokesperson Sean Parnell confirmed the friction to Axios, stating: “The Department of Defense’s relationship with Anthropic is under review. Our nation requires partners who are committed to ensuring our warfighters prevail in any conflict. Ultimately, this is about the safety of our troops and the American people.”

Anthropic’s Claude model currently stands as the sole AI framework cleared for use within the military’s classified systems and is widely regarded as a global leader in multiple applications. Despite the current standoff, Pentagon officials have “sincerely” praised Claude’s capabilities.

The model’s operational utility was recently highlighted by its reported use during the January operation involving the abduction of Nicolas Maduro.

The friction stems from months of contentious negotiations between Anthropic and the Pentagon regarding the specific terms under which the military may utilize Claude.

Anthropic CEO Dario Amodei is said to view these issues with gravity, maintaining what sources describe as a “pragmatic” stance. While the company is reportedly prepared to ease certain existing terms of use, it remains steadfast in ensuring its tools are not deployed for the mass surveillance of Americans or the development of autonomous weapons systems capable of lethal action without human intervention.

The Pentagon, however, contends these conditions are excessively restrictive, arguing that “gray areas” within the terms could render operational utility impossible.

In ongoing discussions with Anthropic and three other major AI laboratories—OpenAI, Google, and xAI—Pentagon officials are insisting on a mandate that would allow the military to utilize AI tools for “all lawful purposes.”

A source familiar with the dynamics noted that senior defense officials have harbored frustrations with Anthropic for some time and are now seizing the opportunity to bring the dispute into the public sphere.

Current mass surveillance laws do not explicitly account for the advancements of AI. The Pentagon already aggregates vast quantities of data, ranging from social media activity to concealed carry permits. However, significant privacy concerns persist regarding the potential for AI to scale these authorities to target civilians disproportionately.

An Anthropic spokesperson stated the company has engaged in “productive, good-faith discussions” with the Pentagon on how to navigate these “new and complex issues.”

The spokesperson reaffirmed the company’s commitment to providing frontier AI for national security purposes, highlighting that Claude was the first model of its kind deployed on classified networks.

Designating Anthropic as a supply chain risk would have far-reaching consequences, requiring the vast network of companies contracted by the Pentagon to certify they do not use Claude in their internal workflows. This poses a significant hurdle, as Anthropic recently disclosed that eight of the top 10 US companies currently utilize the model.

While the specific contract the Pentagon threatens to cancel is valued at up to $200 million—a fraction of Anthropic’s $14 billion annual revenue—the strategic impact is substantial. A senior administration official noted that rival models currently “lag behind” when it comes to specialized government applications, making an immediate transition difficult.

The Pentagon’s aggressive posture toward Anthropic serves as a definitive signal to OpenAI, Google, and xAI as they negotiate their own terms. While these firms have agreed to remove certain safeguards for use on unclassified military systems, they have yet to be deployed for more sensitive, classified operations.

While one senior official expressed confidence that the other three firms would eventually accept the “all lawful use” standard, a source familiar with those talks cautioned that many critical details remain unresolved.