Anthropic and the Pentagon: Tensions Over Access to Claude AI

The CEO of Anthropic, Dario Amodei, has returned to the negotiating table with the U.S. Department of Defense. The goal is to salvage the company's relationship with the U.S. military following an incident that jeopardized their collaboration. Discussions between the two parties collapsed last Friday after weeks of public disputes. These were fueled by the startup's refusal to grant the Pentagon unlimited access to its artificial intelligence, while competitors like OpenAI rushed to fill the void.

Amodei is currently in talks with Emil Michael, the Under Secretary of Defense for Research and Engineering, aiming to finalize a new agreement. This agreement would allow the U.S. military to continue using Anthropic's Claude AI models. According to the Financial Times, citing anonymous sources, Michael attacked Amodei on social media, calling him a "liar" with a "God complex" and accusing him of "endangering the security of our nation."

For Anthropic, securing a new agreement could be a matter of survival. Secretary of Defense Pete Hegseth has stated that he is considering designating Anthropic as a supply chain risk. This category is typically reserved for companies with ties to foreign governments that pose risks to U.S. national security. Such a designation would have repercussions for the American tech ecosystem, forcing companies to abandon Claude and sever ties with Anthropic if they wish to continue working on defense contracts.

A recently leaked memo, sent by Amodei to Anthropic staff on Friday, likely inflamed already tense relations between the company and the Trump administration. In this memo, Amodei reportedly criticized OpenAI's agreement with the Pentagon as "security theater" and described the messaging from both parties as "lies." Amodei suggested that Anthropic's relationship with the federal government had deteriorated because, unlike OpenAI or its leaders, "we did not donate to Trump" and "we did not offer dictatorial-style praise to Trump."

Amodei's memo also mentions that the Department of Defense was close to accepting Anthropic's terms. However, this would have been contingent on removing a specific phrase regarding the analysis of mass-acquired data, which was the only line in the contract that precisely matched the scenario that concerned Anthropic the most. Amodei found this very suspicious.

Anthropic's unpleasant conflict with the Pentagon has centered on the Department of Defense's insistence on having unlimited access to the company's technology. Anthropic has refused to compromise on its two red lines for military use: no mass surveillance of Americans and no lethal autonomous weapons, AI systems capable of killing without human oversight. Hegseth insisted that the AI technology used by the department should be available for "any legal use," terms that Anthropic has rejected due to concerns that this could cross those red lines. xAI and OpenAI are said to have accepted these conditions.

Until last week, Claude was the only AI system with security clearance to handle classified information. Reports indicate that Claude is actively used in military operations, including during the U.S. raid on Venezuela and has been used for strikes in Iran, which likely prompted Trump to reconsider his ban on the federal government's use of Claude.