In just a few months, one AI system quietly became the only model running on classified networks. What followed was not a celebration of innovation, but a standoff.
At the center of the dispute is a simple but explosive question, who decides how powerful AI can be used, the company that builds it, or the government that funds and deploys it?
As deadlines loom and legal tools like the enter the conversation, this episode reveals something bigger than a contract fight. It is an early test of how democracies will control military AI, and what happens when ethics collide with national security.
Analysts increasingly describe the conflict as a defining clash between AI safety principles and military demand for unrestricted technological capability. The debate is raising broader questions about whether governments or technology companies will ultimately control how powerful AI systems are deployed.
Reports indicated that talks between Anthropic and U.S. defense officials had restarted despite the public dispute. The discussions aim to find a compromise allowing the Pentagon to use Claude while preserving the company’s safety guardrails, potentially restoring Anthropic’s access to defense AI systems.
Major investors and technology partners backing Anthropic reportedly urged the company to repair relations with the U.S. Department of Defense. Some fear that losing access to defense contracts could weaken Anthropic’s long-term strategic position in the rapidly expanding military AI sector.
Following news of the Pentagon partnership, OpenAI reportedly experienced a wave of users uninstalling or deleting its applications, as critics argued the company had crossed an ethical line by supporting military AI use. Online campaigns and developer criticism highlighted growing public debate over the role of AI companies in warfare.
Anthropic CEO Dario Amodei reportedly sent a memo criticizing OpenAI’s decision to enter into a Pentagon AI deal, arguing that the company compromised on safety restrictions that Anthropic had refused to remove. The remarks highlighted a growing philosophical divide among leading AI companies over whether advanced AI systems should be used freely in military contexts.
reached an agreement to provide AI capabilities to the U.S. Department of Defense after tensions between the Pentagon and Anthropic escalated. CEO later acknowledged the deal was finalized quickly and said companies ultimately cannot fully control how the military uses their AI once deployed.
Following Anthropic's refusal to lift AI guardrails on Claude for military use, Defense Secretary Pete Hegseth designated the company a supply chain risk after the February 27 deadline passed without compliance. This action halted all federal use of Anthropic technology across classified and unclassified systems, effectively terminating contracts worth up to $200 million. The dispute centered on Anthropic's restrictions against mass domestic surveillance and fully autonomous weapons, which the Pentagon demanded be removed for 'all lawful purposes.' Rivals like OpenAI, Google, and xAI had to more flexible terms, easing their transition into defense roles. Hegseth announced the split publicly on X, calling Anthropic's stance at odds with American values. This marks a pivotal moment in defense-tech relations, setting precedents for AI governance in warfare.
In response to the deadline passing without compliance, the Pentagon directed all U.S. government agencies to immediately stop using Anthropic's technology on classified and unclassified systems, escalating the dispute over military AI restrictions.
Following Anthropic's rejection of the final Pentagon offer to lift AI guardrails, Defense Secretary Pete Hegseth designated the company a 'supply chain risk' per a Trump administration order, halting federal use of Claude AI and effectively terminating contracts without invoking the Defense Production Act.
The Pentagon has issued a deadline of 27th February to Anthropic, demanding that the company allow broader military use of its Claude AI model without internal restrictions or face possible consequences such as contract loss, supply-chain exclusion, or legal action under authorities like the .
Anthropic issued a public statement that it was engaged in good-faith talks with the Pentagon and reaffirmed its ethical stance while negotiations continued ahead of the looming deadline. Anthropic was signalling that it was still willing to talk and cooperate with the U.S. government, but would not abandon its ethical principles around how its AI technology can be used.
Reports confirmed the Pentagon was prepared to potentially invoke the Defense Production Act to compel Anthropic to permit unrestricted military use of Claude or face being designated a supply chain risk and other punitive actions.
U.S. Defense Secretary Pete Hegseth summoned Anthropic CEO Dario Amodei to the Pentagon and issued a deadline for the company to remove that limit Claude’s military use or face consequences like contract termination or legal compulsion.
The Pentagon was reported to be considering ending its relationship with Anthropic and possibly designating it a ‘’ due to continued resistance to lifting usage restrictions, highlighting escalating tensions.
U.S. military used Claude in its 2026 Venezuela operation that resulted in the capture of Nicolás Maduro, underscoring the AI’s battlefield role and contributing to rising friction over its usage policies.
By late January 2026, Reuters and other outlets reported that the Pentagon and Anthropic were at odds, with the DoD pushing for unrestricted military use of Claude while Anthropic resisted on ethical limits such as autonomous weapons and domestic surveillance.
In July 2025 the U.S. Department of Defense awarded Anthropic about $200 million as part of its broader AI modernization efforts, deepening the relationship even as negotiations over guardrails intensified.
In November 2024 Anthropic partnered with and to provide its Claude AI to U.S. intelligence and defence agencies, and by mid-2025 Claude was the only AI model used on classified Pentagon networks as tensions over usage guardrails began brewing.