Anthropic Disputes Pentagon’s National Security Risk Claims in AI Lawsuit

A latest court filing reveals a surprising twist in the escalating dispute between the Department of Defense and Anthropic, the AI firm founded by Dario Amodei. Just days after President Trump and Defense Secretary Pete Hegseth publicly severed ties with Anthropic over concerns about its AI safety protocols, a Pentagon official reportedly signaled to the company that the two sides were “very close” to a resolution on key sticking points. This revelation, contained in sworn declarations submitted to a California federal court late Friday, casts a new light on the abrupt decision to designate Anthropic a national security risk and halt a potential $200 million contract.

The Shifting Narrative

The core of the conflict centers on Anthropic’s reluctance to grant the military unrestricted access to its AI technology. The Pentagon expressed fears that Anthropic might disable or alter its systems mid-operation, or potentially use its AI for mass surveillance or autonomous weapons systems. However, Anthropic argues that the government’s concerns are based on technical misunderstandings and claims never raised during months of negotiations. Sarah Heck, Anthropic’s Head of Policy and a former National Security Council official, stated in her declaration that the company never sought any operational control over military actions. She specifically refuted the claim that Anthropic demanded a role in approving military operations, calling it a “central falsehood” in the government’s filings.

Adding to the complexity, an email from the Pentagon’s Under Secretary Emil Michael to Anthropic CEO Dario Amodei on March 4th – the day after the supply-chain risk designation – indicated that the two sides were “very close” on the issues of autonomous weapons and mass surveillance. This directly contradicts subsequent public statements from Michael, who claimed there were no ongoing negotiations. Amodei himself had publicly stated that conversations with the Pentagon were “productive” just prior to Michael’s shift in tone, as reported by TechCrunch.

Technical Realities and Deployment Constraints

The dispute likewise hinges on the technical feasibility of the Pentagon’s concerns. Thiyagu Ramasamy, Anthropic’s Head of Public Sector and a veteran of Amazon Web Services’ government AI deployments, submitted a declaration detailing the security measures in place when Claude, Anthropic’s AI model, is deployed within government systems. Ramasamy asserts that once Claude is integrated into a secure, “air-gapped” environment operated by a third-party contractor, Anthropic loses all direct access. There is no “kill switch” or backdoor allowing the company to remotely disable or modify the system. Any changes would require explicit Pentagon approval and action.

Ramasamy further explained that Anthropic cannot even access the data inputted by government users, addressing concerns about potential data extraction. He also highlighted that Anthropic employees undergo rigorous U.S. Government security clearance vetting, a process typically reserved for personnel handling classified information. In fact, Ramasamy claims Anthropic is the only AI company where cleared personnel are directly involved in building AI models for classified environments. This level of vetting and access control is a key component of secure AI deployment, as detailed in numerous government cybersecurity guidelines, such as those published by the National Institute of Standards and Technology (NIST).

First Amendment Implications and the Supply Chain Risk Designation

Anthropic’s lawsuit argues that the supply-chain risk designation – the first ever applied to an American company – is a retaliatory measure against the company’s publicly stated views on AI safety, violating its First Amendment rights. The government, however, maintains that Anthropic’s refusal to allow all lawful military uses of its technology is a business decision, not protected speech and that the designation is a legitimate national security measure. This legal battle raises fundamental questions about the balance between national security concerns and the freedom of speech, particularly in the rapidly evolving field of artificial intelligence.

The designation itself carries significant consequences. According to reporting by the Associated Press, it not only terminates Anthropic’s $200 million contract with the Pentagon but also prohibits other defense contractors from doing business with the company. This effectively isolates Anthropic from a major sector of the U.S. Economy and could significantly hinder its growth.

Impact on the Defense AI Landscape

The fallout from this dispute extends beyond Anthropic. The Pentagon’s move sends a chilling effect throughout the AI industry, potentially discouraging other companies from establishing clear ethical boundaries around the military application of their technologies. The incident underscores the challenges of integrating AI into defense systems while upholding principles of responsible AI development. As noted in a recent CBS News report, the Department of Defense has issued a memo requiring all senior leadership to remove Anthropic’s products from their systems within 180 days, impacting critical areas like nuclear weapons systems, ballistic missile defense, and cyber warfare capabilities.

The Role of Third-Party Contractors

The reliance on third-party contractors for deploying and maintaining AI systems within the Department of Defense adds another layer of complexity. Ramasamy’s declaration highlights the critical role these contractors play in securing and operating AI models, effectively acting as a buffer between the AI developer and direct military control. This raises questions about the oversight and accountability of these contractors, and whether current security protocols are sufficient to mitigate potential risks.

What Comes Next

A hearing before Judge Rita Lin in San Francisco is scheduled for March 24th, where arguments will be presented regarding Anthropic’s lawsuit. The outcome of this case will likely set a precedent for how the government regulates the military use of AI and the extent to which companies can assert their ethical principles in the face of national security concerns. The court will need to weigh the government’s legitimate security interests against Anthropic’s First Amendment rights and the potential implications for innovation in the AI sector. Further legal proceedings and potential appeals are anticipated, suggesting a protracted legal battle ahead.