Anthropic, the AI company behind the Claude model, has firmly rejected a high-stakes demand from the Pentagon to eliminate key safeguards on its technology’s military use. The dispute highlights growing tensions between national security priorities and ethical concerns in AI deployment.
The conflict centers on a $200 million contract signed in July, under which Anthropic became the first frontier AI provider to integrate its models into classified U.S. government networks. Claude has supported mission-critical tasks, including intelligence analysis, cyber operations, and operational planning.
In recent negotiations, the Department of Defense, referred to in some contexts as the Department of War following a 2025 executive order by President Trump restoring the historic name, insisted on contract language allowing “any lawful use” of Claude without company-imposed restrictions.
Anthropic has maintained narrow red lines: prohibiting the tool’s application in mass domestic surveillance of U.S. citizens or in fully autonomous weapons systems that operate without human oversight.
Key Points of Contention
- Mass Domestic Surveillance: Anthropic warns that AI could aggregate scattered data into detailed profiles of individuals at a massive scale, a capability it views as incompatible with democratic principles. The company supports AI in lawful foreign intelligence but draws a clear line at domestic monitoring.
- Fully Autonomous Weapons: Current AI systems lack the reliability for independent lethal decisions, Anthropic argues. Without robust guardrails, such tools risk endangering warfighters and civilians by bypassing human judgment.
CEO Dario Amodei addressed the issue directly in a company statement, declaring that Anthropic “cannot in good conscience accede” to the Pentagon’s request. He emphasized that threats to remove the company from supply chains or invoke the Defense Production Act would not alter its stance. Amodei noted the contradictory nature of labeling Anthropic a security risk while simultaneously deeming Claude essential to national defense.
The Pentagon issued an ultimatum during a Tuesday meeting between Defense Secretary Pete Hegseth and Amodei, setting a Friday deadline for compliance. Failure to agree could lead to offboarding Anthropic from DoD systems, designation as a supply chain risk—a label typically applied to adversarial foreign entities or forced cooperation under emergency powers.
Undersecretary for Research and Engineering Emil Michael criticized Amodei sharply on social media, accusing him of having a “God complex” and risking national safety by attempting to control military decisions. Michael argued that proposed safeguards are unnecessary, as existing laws and Pentagon policies already prohibit the contested uses, and stressed the need to counter advancements by adversaries like China.
Anthropic countered that recent contract revisions offered “virtually no progress” on its core concerns, with new phrasing allowing safeguards to be bypassed at will. The company expressed willingness to continue talks and assist in a smooth transition if offboarded, while offering collaboration on R&D to enhance system reliability— an offer the Pentagon has not accepted.
This standoff occurs as other AI firms, including xAI, have agreed to similar “all lawful purposes” terms for classified work, with negotiations advancing for OpenAI and Google. Anthropic’s position underscores a broader debate on balancing AI’s military potential with ethical constraints, especially as frontier models grow more capable.
With the deadline approaching and public rhetoric intensifying, the outcome could shape future government-private sector partnerships in AI. Anthropic remains committed to responsible deployment, prioritizing safeguards that align with democratic values even at the cost of lucrative defense ties.
