OpenAI reveals more details about its agreement with the Pentagon

OpenAI is publicly defending its new contract to deploy AI models in classified Pentagon environments, a deal CEO Sam Altman admitted was “definitely rushed” and had poor “optics.” The agreement was announced just after negotiations between rival AI lab Anthropic and the Pentagon collapsed, which led to the Trump administration designating Anthropic a supply-chain risk. The move places OpenAI in a sensitive government role and raises immediate questions about the enforcement of its ethical red lines around military and surveillance applications.

In a blog post, OpenAI detailed a “multi-layered approach” to safety, contrasting it with competitors who it claims rely primarily on usage policies. The company explicitly prohibits its models from being used for mass domestic surveillance, autonomous weapon systems, and high-stakes automated decisions. According to OpenAI’s head of national security partnerships, Katrina Mulligan, the key safeguard is its deployment architecture; by limiting access to a cloud API, the models cannot be integrated directly into weapons or operational hardware. However, critics like Mike Masnick of Techdirt argue the deal’s compliance with Executive Order 12333 creates a loophole that “absolutely does allow for domestic surveillance.”

The strategic gamble, according to Altman, was to “de-escalate things” between the AI industry and the Department of Defense after the fallout with Anthropic. The decision has already had market repercussions, with public backlash reportedly helping Anthropic’s Claude briefly overtake ChatGPT in Apple’s App Store. The episode highlights the intense pressure on leading AI labs to navigate lucrative but ethically fraught government partnerships, setting a contentious precedent for how the AI sector engages with national security.

OpenAI's rushed Pentagon deal signals a strategic decision to become the U.S. government's default AI provider, calculating that the long-term advantages of incumbency will outweigh the immediate reputational cost and ethical scrutiny.