Anthropic Pentagon Dispute Over Claude Military Use Rules

Anthropic CEO Dario Amodei exchanged multiple emails with Pentagon Deputy Secretary Emil Michael over several months about limits on Claude use in military contexts. Court documents reviewed by the Wall Street Journal show the core disagreement centered on fully autonomous weapons and certain surveillance applications.

Anthropic pushed for explicit bans on those uses. The Pentagon wanted access for any lawful national security purpose. Michael stated that if differences proved too wide the department would not force a solution.

Following the impasse the Pentagon added Anthropic to its supply chain risk list. Partners were advised against using Claude models on Department of Defense projects. A judge later paused parts of the restriction while the government appealed. Michael noted that roughly two-thirds of prior operations using Anthropic tools had already moved to alternative providers.

The episode highlights a widening gap between AI developers focused on narrow safety constraints and government agencies seeking maximum flexibility for defense needs.

Emails Show Specific Requests From Anthropic

Records indicate Anthropic asked the Pentagon to exclude Claude from scenarios involving fully autonomous lethal weapons. The company also sought restrictions on certain monitoring applications that could process large volumes of personal data without oversight.

Pentagon replies emphasized that all uses would remain within existing legal frameworks. Officials argued that broad restrictions would limit legitimate intelligence and logistics work.

Michael wrote that further escalation would not be productive if both sides remained fixed on their positions. The correspondence continued for months without resolution.

Pentagon Adds Anthropic to Risk Registry

After the discussions stalled the Pentagon listed Anthropic among suppliers that posed potential supply chain concerns. The designation triggered recommendations to defense contractors to avoid deploying Claude in active programs.

Contractors reported that procurement offices began flagging any proposed use of the model. Several planned integrations were paused or canceled as a result.

A federal judge issued a temporary hold on parts of the risk designation. The government has appealed the decision and continues to treat the matter as open.

Two-Thirds of Operations Shifted Away

Michael stated that two-thirds of the activities previously involving Anthropic models had already transitioned to other AI providers. The moves occurred before the formal risk listing took effect.

Defense teams cited concerns over future access and policy uncertainty as reasons for the change. Alternative models from both domestic and allied sources were adopted in their place.

The shift reduced immediate reliance on Claude inside Pentagon-affiliated projects.

Industry Watches for Further Signals

Observers are tracking three near-term indicators. First, the outcome of the government appeal will determine whether the risk listing stays in force. Second, any new public statements from Anthropic on revised military guidelines will show whether the company adjusted its stance. Third, contract award notices will reveal whether defense partners continue to avoid Claude or begin restoring limited use.

Each of these developments will clarify the practical boundaries between commercial AI safety policies and government procurement expectations.

Contractors and developers now face clearer trade-offs when choosing models for sensitive work. The episode illustrates how differing priorities on autonomous systems and data oversight can quickly affect access and adoption decisions across the defense sector.