OpenAI Banned Military Use — Microsoft's Pentagon Contract Didn't
SkimNews Take
OpenAI's usage policy was enforceable only at its own API, meaning Microsoft's Azure distribution could route the same models to military clients under separate terms—revealing that foundation-model "responsible use" guardrails are only as binding as the partner distribution chain allows.
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI's 2023 policy explicitly banned military access to its AI, but the Pentagon had already begun experimenting with Azure OpenAI — Microsoft's version of the same models — and spokespeople for both companies say Azure products were never subject to OpenAI's usage policies.
- Microsoft declined to say when Azure OpenAI became available to the Pentagon, noting only that the service wasn't approved for "top secret" government workloads until 2025.
- OpenAI removed its blanket military-use ban in January 2024, with some employees learning of the change from The Intercept before company leaders addressed it at an all-hands meeting.
- OpenAI announced an Anduril partnership in December 2024 for "national security missions" and told employees the scope was narrow and limited to unclassified workloads — in contrast to Anthropic's Palantir deal, which covered classified military work.
- Altman publicly supported Anthropic's red lines against mass surveillance and autonomous weapons, but outside legal experts told WIRED the recent Pentagon agreement appeared to leave room for those activities; OpenAI later amended the terms to address surveillance concerns.
- At a Tuesday all-hands, Altman reportedly told employees OpenAI doesn't get to decide what the Defense Department does with its AI software and said he's interested in selling models to NATO.
Why it matters: The 2023 ban was effectively hollow for the Pentagon, which accessed OpenAI's models through Microsoft's Azure while the public-facing policy said military use was prohibited. That precedent helps explain the current internal backlash — employees, civil liberties advocates like the Institute for Law and AI, and the public at large are left trusting OpenAI's word about what its amended Pentagon agreement actually permits, since the full terms remain undisclosed.
