ISTANBUL
OpenAI said Tuesday that it is revising a recently announced agreement to supply artificial intelligence tools to the US Department of Defense after criticism that the deal was rushed and raised concerns over potential domestic surveillance, also causing a flurry of subscription cancellations from ChatGPT users.
The San Francisco-based company’s CEO, Sam Altman, acknowledged in a message to employees that the agreement, struck shortly after the Pentagon dropped its previous AI contractor Anthropic, appeared “opportunistic and sloppy.”
“We shouldn’t have rushed to get this out,” Altman wrote in a message later reposted on US social media company X, adding that the issues surrounding military AI use are “super complex” and require clearer communication.
Under the revised terms, OpenAI will explicitly bar its technology from being used for domestic mass surveillance or by defense intelligence agencies such as the National Security Agency (NSA), according to Altman.
The company also reiterated that one of its “red lines” is the prohibition of using OpenAI models to direct autonomous weapons systems.
The agreement was announced days after Anthropic was removed as the Pentagon’s AI provider. Anthropic had previously stated that deploying AI systems for mass domestic surveillance would be incompatible with democratic values.
US President Donald Trump criticized Anthropic following its stance, calling the company “left-wing nutjobs" and ordering federal agencies to halt use of its technology. Subsequently, the departments of State, Treasury, and Health and Human Services moved to discontinue Anthropic’s AI products after the DoW designated the company a supply chain risk.
Trump has instructed all US government agencies to phase out Anthropic systems following Defense Secretary Pete Hegseth’s decision.
Despite OpenAI’s assurances that its contract includes stricter guardrails than prior classified AI agreements, the deal sparked online backlash, including calls on social media platforms such as X and Reddit to “delete ChatGPT.”
Nearly 900 employees from OpenAI and Google signed an open letter urging their companies to refuse requests from the Department of Defense to use AI models for domestic mass surveillance or autonomous lethal operations without human oversight. The letter was signed by 796 Google employees and 98 OpenAI staff.
The employees warned that the US government was attempting to pressure companies individually, writing that leaders should “stand together” in rejecting demands that could undermine democratic norms.
The controversy also revived references to the 2013 Snowden disclosures, which revealed extensive surveillance activities by the NSA.
Observers, including OpenAI’s former head of policy research, Miles Brundage, questioned how OpenAI secured an agreement addressing ethical concerns that Anthropic had deemed unacceptable. In posts on X, Brundage suggested some within OpenAI may have conceded to government demands, though he acknowledged the organization’s internal complexity and differing viewpoints.
OpenAI, which reports more than 900 million users of ChatGPT, said it aims to ensure its AI technologies are deployed in line with democratic principles while balancing national security considerations.