Pentagon Warns Anthropic Could 'Subvert' Defense AI Systems - GovInfoSecurity
The Pentagon's Case Against Anthropic Comes Down to This: Safety Features Make the Company Dangerous
The Justice Department made its argument clear in a court filing Tuesday: Anthropic's own safety guardrails are proof the company can't be trusted with American warfighting systems.
In a response to Anthropic's lawsuit challenging its designation as a supply-chain risk, DoJ attorneys argued that the company's restrictions on autonomous weapons and mass surveillance - the very ethical lines Anthropic drew publicly - are exactly why the Pentagon no longer wants Claude anywhere near classified systems. If Anthropic can restrict what its AI does, the reasoning goes, it can also subvert that AI.
"AI systems are acutely vulnerable to manipulation, and Anthropic could attempt to disable its technology or preemptively alter the behavior of its model either before or during ongoing warfighting operations, if Anthropic—in its discretion—feels that its corporate 'red lines' are being crossed," DoJ attorneys wrote in the filing.
The argument inverts Anthropic's own logic. The company built those restrictions precisely because it believes current AI systems aren't reliable enough for fully autonomous weapons and shouldn't power mass surveillance of American citizens. In a March 4 statement, Dario Amodei argued those positions reflected technical judgment, not political stance. The Pentagon has now turned those judgments into evidence of untrustworthiness.
"Our only concerns have been our exceptions on fully autonomous weapons and mass domestic surveillance, which relate to high-level usage areas, and not operational decision-making," Amodei wrote.
Judge Rita Lin of the US District Court for the Northern District of California will hear Anthropic's request for an injunction blocking the designation next Tuesday, March 24. If she sides with Anthropic, the company could resume defense work while litigation continues. If not, the supply-chain risk label stands - and defense contractors would be required to remove Claude from Pentagon programs.
The practical stakes are significant. According to the DoJ filing, Claude is currently the only AI model cleared for use on the department's classified systems. Defense Secretary Pete Hegseth and other officials have argued this concentration creates unacceptable risk. The department is working to replace Anthropic's tools with AI from OpenAI, Google, and xAI - a transition the DoJ acknowledged won't be instant.
"The department cannot simply flip a switch at a time when Anthropic currently is the only AI model cleared for use," the filing states, "and high-intensity combat operations are underway."
The broader industry reaction has been decidedly one-sided. Microsoft, Google DeepMind, a federal employee labor union, former military leaders, and AI researchers have all filed briefs supporting Anthropic. Not a single organization has filed in support of the government's position.
The case raises questions that extend beyond one company's fortunes. If safety guardrails are disqualifying for defense work, what does the Pentagon actually want from AI vendors? And if the answer is unfettered access to AI capabilities, what does that say about the role of ethical constraints in national security?
Anthropic has until Friday to file its counter-response. The company's position has been consistent: it wants to work with the Defense Department but won't remove the narrow restrictions it believes current AI capabilities warrant. The DoJ's filing suggests that position itself is the problem.
Primary source: Anthropic CEO Dario Amodei's March 4 statement on the Department of War designation (https://www.anthropic.com/news/where-stand-department-war)