Altman Admits Rushed Pentagon Deal, Tightens Surveillance Safeguards

OpenAI CEO Sam Altman said the company “shouldn’t have rushed” its newly announced contract with the United States Department of War, acknowledging that the timing and rollout created avoidable controversy.

In a post on X, Altman shared what he described as a repost of an internal memo outlining changes to the agreement. The updated language clarifies that OpenAI’s AI systems “shall not be intentionally used for domestic surveillance of US persons and nationals”. It further specified that the Department understands the prohibition on deliberate tracking, monitoring or surveillance of Americans, including through commercially acquired personal or identifiable information.

Altman added that the Pentagon had affirmed OpenAI’s tools would not be deployed by intelligence agencies such as the National Security Agency under the agreement. He said the company would work with defence officials to implement additional technical safeguards, noting that “there are many things the technology just isn’t ready for”, and that important safety trade-offs remain poorly understood.

The revisions follow backlash over the timing of the deal, which was announced hours after US President Donald Trump directed federal agencies to stop using tools from rival AI firm Anthropic, and shortly before the US’s strikes on Iran. The sequence fuelled criticism online, with some users reportedly shifting from ChatGPT to Anthropic’s Claude.

Anthropic’s own negotiations with the Department of War had recently broken down. The company, founded by former OpenAI researchers, had sought explicit guarantees against domestic surveillance and the development of autonomous weapons without meaningful human control. US Defence Secretary Pete Hegseth subsequently designated Anthropic a supply-chain threat.

Altman said he had privately urged officials not to treat Anthropic as a risk and expressed hope it would be offered similar terms. The episode has exposed widening tensions over how quickly frontier AI systems should be integrated into national security operations—and under what constraints.