OpenAI Models to Prioritize Iran Strike Targets
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI reached a Pentagon agreement roughly two weeks before publication allowing its AI in classified environments; Sam Altman said the military can't build autonomous weapons with it, but the deal only requires following the military's own "quite permissive" guidelines.
- A defense official told the outlet OpenAI's models could prioritize strike targets by analyzing text, image, and video inputs including logistics data like plane and supply locations — a generative AI use case "being tested in earnest for the first time in Iran."
- Anduril won a $20 billion US Army contract the week before publication; its existing OpenAI partnership (announced end of 2024) focuses on counter-drone analysis, and six US service members were killed in Kuwait on March 1 when an Iranian drone attack went unintercepted by US air defenses.
- Anthropic was designated a "supply chain risk" by the Pentagon after refusing to allow its AI for "any lawful use," prompting Trump to order the military to stop using it; Anthropic is fighting the designation in court.
- OpenAI joined the Pentagon's GenAI.mil back-office platform in February for drafting policy documents and contracts, following Google Gemini and xAI's Grok onto a system Defense Secretary Pete Hegseth has pushed across administrative roles.
Why it matters: OpenAI is now threaded into three Pentagon layers at once — classified targeting analysis, Anduril's $20 billion counter-drone stack, and the GenAI.mil back-office platform — meaning AI-generated recommendations are migrating from passive analysis to active combat decisions in a live war zone, with the article itself flagging that human "double-checking" of model outputs may be the only guardrail.
