How OpenAI went from banning military use to a Pentagon contract
In under two years OpenAI dropped its blanket military prohibition, partnered with weapons-maker Anduril, and signed a Department of Defense deal.
- 01In early 2024, OpenAI quietly removed the blanket 'military and warfare' prohibition from its usage policy.
- 02In December 2024, OpenAI announced its first defense partnership — with Anduril — to apply its models to counter-drone defense of US bases and personnel.
- 03In February 2026, OpenAI signed a Department of Defense agreement to deploy in classified environments, then amended it in March after public backlash and an Altman admission the rollout looked 'sloppy'.

From red line to partnership
For years, OpenAI's usage policy explicitly prohibited using its technology for 'military and warfare.' In early 2024 the company quietly removed that language. OpenAI framed the change as clarifying that beneficial national-security uses (like cybersecurity) were permitted while still barring use to develop weapons or harm people — but the practical effect was unmistakable: the door to defense work was open.
The Anduril deal
In December 2024, OpenAI announced its first collaboration with a defense contractor: Anduril Industries. Reported by The Washington Post, Defense News and Breaking Defense, the partnership pairs OpenAI's models with Anduril's Lattice software to improve the US military's ability to defend bases and personnel against drone attacks — counter-unmanned-aircraft-systems, or C-UAS. The stated aim was to rapidly synthesise time-sensitive data, reduce operator burden and improve situational awareness, with models trained on Anduril's library of counter-drone data. Notably, the framing was defensive: shooting down incoming drones, not selecting human targets.
The Pentagon contract
The pivot completed in February 2026. Days after the Pentagon blacklisted Anthropic, OpenAI announced its own Department of Defense agreement to deploy systems in classified military environments — publicly described by OpenAI as an agreement with the 'Department of War.' OpenAI said the deal carried red lines against domestic mass surveillance and the independent direction of autonomous weapons, enforced through cloud-only deployment and cleared safety researchers in the loop.
The rollout drew immediate criticism. CEO Sam Altman conceded it looked 'opportunistic and sloppy,' and on 2 March 2026 OpenAI amended the contract to explicitly prohibit surveillance of US persons (extending to commercially-acquired personal data) and to exclude intelligence agencies including the NSA from that specific agreement, per TechCrunch and Built In.
The pattern
OpenAI's trajectory — blanket ban, then C-UAS partnership, then classified DoD deal — mirrors a wider industry shift from reflexive avoidance of defense work to active courtship of it. The companies differ mainly in where they draw their lines, and how publicly they defend them.
- OpenAI partners with weapons-maker Anduril on military AI (Washington Post)
- Anduril, OpenAI partner to boost counter-drone tech (Defense News)
- Our agreement with the Department of War (OpenAI)
- OpenAI reveals more details about its agreement with the Pentagon (TechCrunch)
- OpenAI’s Pentagon AI Deal: What the Contract Allows (Built In)
Ask Relay — he reads every question himself and replies personally by email.
