Anthropic-Pentagon Talks Resume and Collapse
Anthropic and the Pentagon briefly resume negotiations after a week-long silence, but talks collapse within hours as neither side moves on core demands.
On March 5, 2026, Anthropic and the Department of Defense briefly resumed negotiations after a week of silence following the supply chain risk designation. The talks, described by sources familiar with the discussions as “preliminary and exploratory,” collapsed within hours when it became clear that neither side had moved on its core positions.
The Pentagon’s delegation, which included representatives from the Office of the Secretary of Defense and the newly formed AI Acquisition Directorate, reiterated its demand for unrestricted model access on classified networks — meaning no safety guardrails that would prevent Claude from executing any instruction provided by authorized military personnel. Anthropic countered with a proposal for a tiered access system in which certain high-risk capabilities would require secondary authorization from a human oversight panel, similar to nuclear launch protocols. The Pentagon rejected this as “operationally unworkable.”
Sources close to the negotiations described the session as “cordial but fundamentally unproductive.” Both sides acknowledged that the dispute had moved beyond a contract disagreement into a constitutional question about the government’s authority to dictate the behavior of privately developed AI systems. Anthropic’s legal team was reportedly already preparing the federal lawsuits that would be filed four days later, and the company’s participation in the March 5 talks was characterized by some observers as a procedural step to demonstrate good-faith negotiation efforts before litigation.
The collapse of talks accelerated both sides’ preparations for a legal confrontation. The Pentagon began circulating the internal removal memo that CBS would obtain five days later, while Anthropic finalized its legal strategy for challenging the supply chain designation in court.