On July 14, 2025, Anthropic announced that the US Department of Defense, through its Chief Digital and Artificial Intelligence Office, had awarded it a two-year prototype agreement with a $200 million ceiling. The agreement covers developing “frontier AI capabilities” for national security, including prototypes tuned on defense data, work to forecast and mitigate adversarial uses of AI, and the sharing of technical insights with the department.
The award was notable because Anthropic has positioned itself as the most safety-focused of the leading AI labs, and the announcement leaned into that identity rather than away from it. The company argued that “the most powerful technologies carry the greatest responsibility” and that AI systems used in government must be “reliable, interpretable, and steerable,” pointing to its safety testing and usage policies as what makes its models suited for national-security work.
The agreement was one of several near-simultaneous defense deals across the industry, as the same CDAO office made comparable awards to other frontier labs. It followed Anthropic’s earlier moves into the public sector, including the Claude Gov models for national-security customers and partnerships to run Claude through secure government cloud infrastructure. Together these marked a clear shift: the major AI labs moved from keeping the military at arm’s length to actively competing for defense contracts.
For a general reader, the deal captures how quickly the relationship between AI labs and the military changed. A company built around the message that advanced AI is dangerous and must be handled carefully chose to apply that same framing to defense work, arguing that responsible labs, rather than less cautious competitors, should be the ones building national-security AI.