Pentagon may end ties with Anthropic over AI use: report

The Pentagon is considering ending its relationship with AI firm Anthropic over disagreements on military-use restrictions, Axios reported.

The department seeks to allow the military to use AI models for all lawful purposes, including sensitive weapons and intelligence operations, but Anthropic has refused to fully agree, citing limits on mass surveillance of Americans and autonomous weapons.

Negotiations have been strained, with concerns about the practicality of managing specific use cases and potential operational conflicts.

Anthropic, which signed a contract worth up to US$200 million with the Pentagon last year, has emphasized its commitment to supporting US national security but has faced internal and external disagreements over the scope of AI deployment.

The Pentagon is also in talks with other AI providers, with some showing more flexibility in meeting the department’s standards for classified and unclassified use.

The future of the partnership remains uncertain.

🔗 Source: Axios

🧠 Food for thought

Implications, context, and why it matters.

Anthropic ties its stance to in-house AI risk research

Anthropic links its refusal to its CEO’s views plus internal research on AI risks 1.
CEO Dario Amodei calls domestic mass surveillance and fully autonomous weapons “bright red lines.” He says the U.S. should not copy its “autocratic adversaries” 1.
The position sits alongside growing military use of Claude, Anthropic’s AI model, inside U.S. defense settings.
Claude became the first frontier model (a top-tier, general-purpose AI system) used on the Pentagon’s classified networks. The Wall Street Journal wrote that it ran via Palantir’s tools (software used by governments and large organizations to integrate and analyze data) during the operation to capture Venezuelan President Nicolás Maduro. Anthropic says it cannot comment on whether Claude was used in any specific operation 2.

The dispute could shape how the AI industry works with defense

The Pentagon wants “all lawful purposes” access from Anthropic and its rivals OpenAI, Google, plus xAI 3.
One competitor has reportedly accepted the Pentagon’s terms, while two others are described as “more flexibility than Anthropic” 4.
The shift can push AI labs to weaken ethical guardrails in pursuit of defense revenue, which may reward compliance over safety work.
The result will clarify whether AI companies can hold firm on government use limits, or whether national security needs will overrule company rules and shape future AI use in warfare 5.