OpenAI CEO admits military AI decisions rest with government, not company

Sam Altman, CEO, OpenAI

• OpenAI can't control Pentagon's AI tech use post-deployment
• Altman: OpenAI sets safeguards, DoD handles operations
• Deal bans domestic surveillance, mandates human oversight for force

Did our AI summary help?

OpenAI CEO Sam Altman told employees this week that the company has no authority over how the US Department of Defense (DoD) uses its artificial intelligence technology once it is deployed — a distinction that comes amid an intensely controversial Pentagon partnership.

Speaking at an internal all-hands meeting, Altman emphasised that OpenAI’s role is to build safe, powerful AI and contribute technical guidance, but it doesn’t “get to make operational decisions” about military use of those tools. “So maybe you think the Iran strike was good and the Venezuela invasion was bad,” he said, “you don’t get to weigh in on that.”

The comments come four days after OpenAI announced a new agreement with the Pentagon to allow its models to be used in classified networks, a deal that followed a public breakdown in talks between the Defense Department and rival Anthropic.  President Donald Trump subsequently directed federal agencies to cease use of Anthropic’s tools and labelled the company a “supply-chain risk,” sharply escalating tensions in the AI-defence space.

Altman defended the arrangement while acknowledging the rollout “looked opportunistic and sloppy” and that it had drawn significant internal and public criticism. He said the Pentagon has expressed respect for OpenAI’s expertise and allows the company to build its own safety stack and apply its own safeguards.

Despite that, Altman pointed out that the DoD retains control over tactical and operational decisions, a position that has frustrated some staff after the company’s rapid announcement last Friday. Employees had expressed concerns about the ethical implications of deeper military integration, especially as Anthropic’s talks fell apart amid debates over surveillance and autonomous weapons.

OpenAI’s deal contains explicit limits — including prohibiting use for domestic surveillance and requiring human responsibility for force — but how the technology is put into action remains under the Pentagon’s authority, Altman said.