OpenAI Inks Classified Deal with DOW
OpenAI has reached a classified deployment agreement with DOW, allowing its AI models to be used on secure, compartmentalized government networks. The deal signals the growing integration of commercial AI into sensitive national security and defense applications.
The agreement materializing on February 27, 2026, followed the breakdown of similar negotiations between the Pentagon and rival AI lab Anthropic. Anthropic's talks collapsed after the company reportedly refused to remove safeguards against the use of its AI for mass surveillance and fully autonomous weapons, leading the Trump administration to ban the company from federal contracts. OpenAI asserts its agreement contains more robust guardrails than any previous classified AI deployment. The deal explicitly establishes three "red lines": the technology cannot be used for mass domestic surveillance, to direct autonomous weapons systems, or for high-stakes automated decisions like social credit systems. CEO Sam Altman stated the Department of War (DOW) agrees with these principles. To enforce these limits, the AI models will be deployed exclusively on a cloud-based infrastructure, not on edge devices, which architecturally prevents their use in powering fully autonomous weapons. OpenAI will retain full control over the safety mechanisms and will not provide the DOW with "guardrails off" or non-safety-trained versions of its models. As part of the deal, OpenAI will embed cleared engineers and safety researchers within government teams to oversee the technology's application and maintain its safety systems. This hands-on involvement is a key part of the multi-layered approach to ensure the agreed-upon "red lines" are not crossed. The contract specifies that any use of the AI in autonomous or semi-autonomous systems must undergo rigorous testing and validation, adhering to DoD Directive 3000.09. Furthermore, any intelligence activities must comply with existing U.S. laws like the Foreign Intelligence Surveillance Act and Executive Order 12333. This agreement is a significant expansion of OpenAI's government work, which already included a pilot program with the DoD's Chief Digital and Artificial Intelligence Office for administrative tasks and collaborations with NASA and the Air Force Research Laboratory. The deal came shortly after OpenAI announced a partnership with Amazon Web Services to create a runtime environment on the Amazon Bedrock platform, which is approved for sensitive government data.
