Anthropic's negotiations for a $200 million contract with the Pentagon ultimately failed due to irreconcilable differences over the use of its AI. The company insisted on safeguarding against domestic surveillance of Americans and deployment in autonomous weapons without human oversight. Pentagon officials, including chief technology officer Emil Michael, wanted unfettered use of the technology, viewing it as any other software tool and arguing that company usage policies should not override U.S. law. A last-minute push saw the Pentagon offering to accept most of Anthropic's terms, provided the company removed a specific clause prohibiting the "analysis of bulk acquired data," a demand Anthropic viewed suspiciously as directly enabling domestic surveillance.
The tensions escalated significantly in late 2024 and early 2025, despite Anthropic having previously secured a $200 million Pentagon contract and tailored Claude models for military use, including handling classified materials. The dispute culminated in Defense Secretary Pete Hegseth setting a 5:01 p.m. Friday deadline for Anthropic to compromise. When the company held firm on its protective clauses, Hegseth declared Anthropic a security risk and cut it off from government work. This designation, typically reserved for foreign national security threats, was unprecedented for an American company, leading Anthropic to announce its intent to sue the Pentagon.
The collapse of the deal also involved personal clashes and stylistic differences between the negotiators, particularly between Emil Michael and Anthropic CEO Dario Amodei. Michael, frustrated by Anthropic's stance and Amodei's unavailability for a crucial call, had simultaneously negotiated an alternative agreement with OpenAI. Following Anthropic's rejection of the Pentagon's terms, OpenAI CEO Sam Altman quickly announced a deal with the Pentagon, with Defense Secretary Hegseth reposting the announcement. This swift pivot to OpenAI, a rival, highlights the Pentagon's determination to acquire AI capabilities under its own terms, emphasizing a philosophical divide on AI ethics in warfare.
