Inside OpenAI’s Pentagon Deal and the Three Red Lines on Military AI
The company framed the contract as part of a broader effort to align frontier AI with democratic oversight rather than unchecked military deployment.
Topics
News
- Enterprise AI Scales Fast, but Structural Change Lags, Study Finds
- Inside OpenAI’s Pentagon Deal and the Three Red Lines on Military AI
- OpenAI Partners with Amazon and Microsoft
- Nvidia, Global Telecom Groups Back AI-Driven 6G Push
- Meta’s AI Strategy Signals a New Reality: Compute Is the Competitive Moat
- Anthropic Pushes Back on Pentagon Pressure to Loosen AI Use Restrictions
OpenAI said over the weekend that it had reached an agreement with the Pentagon to deploy advanced artificial intelligence systems in classified environments, outlining three “red lines” that will govern how its technology may be used.
The announcement follows US President Donald Trump’s move a day earlier directing the government to stop working with Anthropic and to label the AI lab a supply chain risk. Anthropic has said it will challenge any such designation in court.
In a blog post, OpenAI said, “Yesterday we reached an agreement with the Pentagon for deploying advanced AI systems in classified environments, which we requested they also make available to all AI companies.” It added, “We think our agreement has more guardrails than any previous agreement for classified AI deployments, including Anthropic’s.”
OpenAI also said it does not believe Anthropic should be labeled a supply chain risk and has conveyed that view to the government.
At the center of the agreement are three red lines.
In a blog post titled “Our agreement with the Department of War,” the company said its technology cannot be used for mass domestic surveillance, cannot be used to direct autonomous weapons systems, and cannot be used for high-stakes automated decisions such as “social credit” systems.
The company emphasized that the deployment will be cloud-only, meaning its models will not run on edge devices.
That structure, OpenAI said, reduces the risk that its systems could be embedded in fully autonomous lethal weapons. “We retain full discretion over our safety stack, we deploy via cloud, cleared OpenAI personnel are in the loop, and we have strong contractual protections,” the company wrote.
OpenAI also said it will not provide models without safety guardrails and will keep cleared engineers as well as safety and alignment researchers involved in the deployment. The company said its architecture will allow it to independently verify that its red lines are not crossed, including by running and updating classifiers.
The contract language references existing US laws governing surveillance and military AI use, including the Fourth Amendment, the National Security Act of 1947 and the Foreign Intelligence Surveillance Act of 1978. It states that the AI system will not independently direct autonomous weapons where human control is required by law or policy and will not be used for unconstrained monitoring of US persons’ private information.
The Pentagon has signed AI agreements worth up to $200 million each over the past year with major labs including OpenAI, Anthropic and Google, as it looks to integrate AI more deeply into defense and intelligence operations.
OpenAI said it had previously declined to pursue a classified deployment contract because it did not believe its safeguards were ready. It added that it remains unwilling to remove key technical protections to improve performance for national security work.
Framing the deal as part of a broader political and technological balancing act, the company wrote, “We believe strongly in democracy. Given the importance of this technology, we believe the only good path forward requires deep collaboration between AI efforts and the democratic process.”
OpenAI said it could terminate the agreement if the government violates the contract terms, though it added, “We don’t expect that to happen.”
Meanwhile, Anthropic appears to be gaining consumer momentum despite its dispute with Washington. As first reported by CNBC, its chatbot Claude climbed to the top of Apple’s U.S. App Store free rankings over the weekend, overtaking OpenAI’s ChatGPT.
Data from SensorTower shows Claude was outside the top 100 at the end of January but has steadily risen through February, jumping from sixth midweek to first on Saturday. An Anthropic spokesperson said daily signups have hit record highs each day this week, free users are up more than 60% since January and paid subscriptions have more than doubled this year.
