OpenAI Inks New Deal for Classified Military AI
OpenAI has reached a new agreement with the Pentagon's Office of Warfighting (DOW) to deploy its AI models on classified networks. This marks a significant deepening of the relationship between the AI leader and the U.S. national security apparatus, with the deal specifying that “AI guardrails” will be in place.
This collaboration follows OpenAI's removal of an explicit ban on "military and warfare" applications from its usage policy in January 2024. The company replaced the specific prohibitions with a broader injunction not to use its services to "harm yourself or others," a change it said was for clarity. In June 2025, the Pentagon had already awarded OpenAI a $200 million contract to develop "prototype frontier AI capabilities" for both warfighting and enterprise domains. That award is part of a broader "OpenAI for Government" initiative that also involves work with agencies like DARPA. The deal was announced just after the Trump administration ordered all federal agencies to phase out technology from rival AI developer Anthropic, designating the company a "supply chain risk." The move against Anthropic, an American company, was described as an unprecedented action historically reserved for U.S. adversaries. Anthropic's blacklisting followed a contract dispute in which the company refused to remove two key safety guardrails from its agreement with the Pentagon. The company would not consent to its AI models being used for mass domestic surveillance or to power fully autonomous weapons systems without human control. OpenAI has stated its own agreement contains "red lines" that prohibit its technology from being used for mass domestic surveillance, to direct autonomous weapons, or for high-stakes automated decisions. Initially, the contract's language only required compliance with existing laws, drawing criticism that this could still permit surveillance programs. Following the public backlash, OpenAI amended the agreement to explicitly forbid the intentional surveillance of U.S. persons and nationals. The updated terms also affirmed that intelligence agencies like the NSA cannot use OpenAI's services without a new, separate agreement. The deal brings OpenAI more in line with defense-tech regulars like Palantir and Anduril Industries, both of which are major contractors for Project Maven, the Pentagon's signature AI program for analyzing drone and satellite imagery. Google was an early partner on Project Maven but withdrew in 2018 after employee protests over the military use of its AI.
