OpenAI signs Pentagon AI deployment deal with three hard limits

OpenAI has reached a formal agreement with the U.S. Department of Defense for deploying its AI systems in classified environments, announcing the deal alongside a direct claim that its contract includes more safety guardrails than any previous classified AI deployment agreement -- explicitly naming Anthropic's prior arrangement as the benchmark. The deal establishes three firm redlines: no mass domestic surveillance, no autonomous weapons systems, and no automated high-stakes decisions such as social credit systems.

Until now, classified AI deployments with the Pentagon operated under narrower restrictions typically enforced through usage policies rather than contractual technical safeguards. This agreement is structurally different: OpenAI retains full control over its own safety stack and can run and update classifiers independently of the government. Cleared OpenAI engineers will be forward-deployed with government teams, with safety and alignment researchers kept in the loop. The contract is cloud-only -- no edge devices -- which the company says closes a pathway to autonomous weapons use. Critically, the contract language locks in current surveillance and autonomous weapons laws even if future legislation weakens those standards, and OpenAI can terminate the agreement outright if the government violates its terms. The company also asked the government to offer the same contract terms to all AI labs, specifically requesting that officials resolve a dispute with Anthropic over its "supply chain risk" designation -- a label OpenAI publicly opposes.