OpenAI changes Pentagon deal after backlash

OpenAI is revising its recent deal with the US Department of Defense after criticism from users and internal debate over how its AI could be used in military operations.

The company has confirmed that it is adding new language to the agreement to block its systems from being used for domestic spying. In an updated statement, OpenAI said its tools “shall not be intentionally used for domestic surveillance of U.S. persons and nationals,” and that intelligence agencies such as the NSA would not be able to use the technology without further changes to the contract.

CEO Sam Altman admitted the rollout of the agreement had not gone smoothly. Posting on X, he said the company “shouldn’t have rushed to get this out on Friday,” and described the situation as having “looked opportunistic and sloppy.”

The deal was announced shortly after the Trump administration cut ties with rival AI firm Anthropic, following a dispute over how its Claude model could be used. Anthropic had insisted its technology should not power mass domestic surveillance or fully autonomous weapons. President Donald Trump directed federal agencies to “immediately cease” using Anthropic’s systems and labeled the company a national security risk.

The Pentagon has said it does not intend to use AI for mass surveillance of Americans or to build weapons that act without human involvement, but it wants flexibility for lawful military applications.

Altman told employees in a company-wide meeting that OpenAI would not control how the Defense Department uses its systems in real-world operations. “So maybe you think the Iran strike was good and the Venezuela invasion was bad,” he said. “You don’t get to weigh in on that.”

He added that operational calls ultimately sit with Defense Secretary Pete Hegseth. “I think this was an example of a complex, but right decision with extremely difficult brand consequences and very negative PR for us in the short term,” Altman said, according to reports of the meeting.

The updated contract will allow OpenAI to deploy its models across the Pentagon’s classified networks. Last year, the company secured a $200 million defense contract covering non-classified uses.

The controversy has had visible fallout. Data from Sensor Tower shows daily uninstalls of ChatGPT jumped 200% after the Pentagon partnership was made public. At the same time, Anthropic’s Claude climbed to the top of Apple’s App Store rankings.

OpenAI is also weighing a separate agreement with North Atlantic Treaty Organization (NATO). Initial comments suggested the company was targeting classified systems, but a spokesperson later clarified the opportunity relates to NATO’s “unclassified networks.”

AI is already widely used in defense for tasks such as logistics planning and analyzing large data sets. Companies like Palantir Technologies provide platforms that combine satellite imagery and intelligence reports with commercial AI models to support military decisions. Critics warn that large language models can still make errors or fabricate information, raising concerns about relying on them in high-stakes settings.

Professor Mariarosaria Taddeo of Oxford University has argued that with Anthropic sidelined, “the most safety-conscious actor” is now “out from the room.” “That is a real problem,” she said.

Meanwhile, Elon Musk’s xAI has also secured a deal to supply models for classified use. Referring to competition, Altman told staff: “But there will be at least one other actor, which I assume will be xAI, which effectively will say ‘We’ll do whatever you want.’”