Pentagon Threatens to Terminate $200 M Anthropic Deal Unless Claude Is Open for “All Lawful” Military Uses

Silicon Valley’s latest clash with the Pentagon echoes the 1945 Los Alamos scientists’ post-Trinity debates and the 1969 ARPANET export fights: technologists help build a powerful tool, then recoil at the government’s intended use. Since the Church Committee (1975) exposed domestic spying, U.S. intelligence has repeatedly tried to bend new tech—telephone switches, bulk internet taps, and now foundation models—toward total access, while some inventors attempt to limit it. The Pentagon’s threat signals a maturation of the AI-industrial complex: procurement dollars are shifting from bespoke defense primes to civilian labs whose ethics teams can slow deployment. Whether Anthropic holds the line or is replaced by less-constrained rivals will shape norm-setting for autonomous and surveillance tech in the next half-century; like the 1968 Nuclear Non-Proliferation Treaty, today’s guardrails (or their absence) could define acceptable state behavior with AI for generations.