Pentagon Gives Anthropic Ultimatum on AI Model Access
The Pentagon has reportedly issued a 'final offer' to Anthropic, demanding unrestricted access to its Claude AI models by this Friday. The ultimatum threatens contract cancellation, a supply chain risk designation, and potential invocation of the Defense Production Act. This development marks a major escalation in the standoff over the AI company's resistance to allowing its technology to be used for lethal or surveillance applications.
Anthropic's refusal centers on two specific "red lines": preventing Claude from being used for domestic mass surveillance or in fully autonomous weapons systems that operate without human oversight. CEO Dario Amodei stated the company "cannot in good conscience accede" to the Pentagon's demand for unrestricted use, arguing current AI is not reliable enough for autonomous weapons. This ethical stance has created a high-stakes public clash over AI governance in national security. The standoff reportedly intensified after Claude was used via Palantir's classified platform during a January military operation to capture Venezuelan President Nicolás Maduro. Following the raid, Anthropic officials expressed disapproval of its technology being used for such purposes, which soured the relationship with the Pentagon and led to the current impasse over contractual assurances. Defense Secretary Pete Hegseth has demanded that Anthropic agree to "any lawful use" of its models, a broad mandate the company finds too ambiguous. In response to Anthropic's resistance, the Pentagon has threatened to designate the company a "supply chain risk," a label historically applied to foreign adversaries, which could bar it from sensitive systems. The Pentagon is also threatening to invoke the Defense Production Act (DPA), a Korean War-era law that gives the president broad powers to compel private companies to prioritize government contracts for national defense. Legal experts note that using the DPA to force a software company to alter its safety protocols and build products against its conscience would be an unprecedented expansion of the act's authority, which was designed to direct the production of physical goods and materials. This confrontation highlights a significant shift in the Pentagon's AI procurement strategy. A January 2026 memo outlined an "AI-first" mandate, requiring that the latest AI models be available to the military within 30 days of public release and that contracts include standard "any lawful use" language. This strategy aims to accelerate AI adoption by removing barriers and ensuring access to cutting-edge technology from companies like Google, OpenAI, and xAI, all of whom have DoD contracts. The dispute is unfolding as the DoD's Chief Digital and AI Office (CDAO) pushes to scale AI capabilities across the force, building on foundational projects like Project Maven. Started in 2017, Project Maven uses AI to analyze surveillance data to identify potential targets, a program that saw Google withdraw in 2018 after employee protests, with contractors like Palantir stepping in. Anthropic was reportedly the first frontier AI company to deploy its models on classified government networks and is currently the only one running on the Pentagon's classified systems via Palantir's AI Platform. This makes its potential removal a significant disruption to ongoing intelligence and operational workflows, even as the company has pledged a smooth transition to another provider if necessary.
