OpenAI reveals more details about its agreement with the Pentagon

OpenAI has publicly detailed its new agreement with the Pentagon, a deal CEO Sam Altman admits was 'definitely rushed' and came with poor optics. The announcement followed the collapse of negotiations between the Department of Defense and Anthropic, after which President Trump directed federal agencies to phase out Anthropic's technology. OpenAI quickly secured its own contract for deploying models in classified environments, prompting immediate scrutiny about its safeguards versus Anthropic's stated red lines around autonomous weapons and mass surveillance. In response, OpenAI published a blog post defending its approach and outlining three explicit prohibitions for its technology.

OpenAI's framework prohibits mass domestic surveillance, autonomous weapon systems, and 'high-stakes automated decisions' such as social credit systems. The company argues its 'multi-layered approach'—retaining control over its safety stack, deploying via cloud API, keeping cleared personnel in the loop, and enforcing strong contractual protections—exceeds the safety measures of competitors who rely primarily on usage policies. However, critics like Techdirt's Mike Masnick contend the contract language, which states data collection will comply with Executive Order 12333, could still permit surveillance. OpenAI's head of national security partnerships, Katrina Mulligan, countered that deployment architecture matters more than contract language, emphasizing that cloud API deployment prevents direct integration into weapons hardware. Altman, facing backlash that saw Anthropic's Claude briefly overtake ChatGPT in Apple's App Store, stated the deal was an attempt to 'de-escalate' tensions between the DoD and the AI industry, acknowledging the high-stakes gamble for OpenAI's reputation.