OpenAI employees have publicly voiced concerns about the agreement on X. Some ChatGPT users reportedly canceled their subscriptions and switched to Anthropic’s Claude, which climbed to the number one spot in Apple’s App Store — a clear symbolic blow.

OpenAI now plans to amend its contract with the US Department of Defense (DoD) to include clearer safeguards. CEO Sam Altman published details on X from a post that had initially been shared internally.

The most significant addition: the AI system may not be intentionally used to surveil US citizens — including indirect use as an analytical tool after the purchase of commercial personal data.

The Department of Defense also confirmed that intelligence agencies such as the NSA are not permitted to use OpenAI’s services under the current agreement. Any such access would require a separate contract.

Altman further emphasized that the technology is not yet ready for many potential applications and that OpenAI does not fully understand all trade-offs required for safe deployment. The company intends to address these questions gradually, together with the DoD, and implement technical safeguards. Altman stated that OpenAI aims to operate through democratic processes and would refuse unconstitutional orders.

He acknowledged that Friday’s initial announcement was rushed and poorly communicated. Altman also reiterated that Anthropic should not be classified as a security-critical supplier (SCR) and should be offered the same contractual terms.

OpenAI Responds to Criticism Over Pentagon Deal

OpenAI researcher Noam Brown — known for last year’s breakthrough work on reasoning models — publicly supported the contract revisions on X. He noted that the original language left open questions, particularly regarding new AI-enabled surveillance capabilities. The world should not have to rely solely on trust in AI companies or intelligence agencies, Brown wrote.

Instead, he argued, such gaps must be addressed through democratic processes before deployment in intelligence contexts. Brown warned of a gradual normalization in which democratic oversight of critical political decisions could be bypassed.

He also announced plans to become more directly involved in AI policy discussions at OpenAI. Given the rapid pace of research progress, he said, it is essential that researchers have a voice in political and strategic decisions.

OpenAI’s sharp shift in tone may also be linked to multiple leaks over the weekend, which described how the company actively pursued the Pentagon deal while Anthropic was still negotiating.

According to The New York Times, Altman contacted Pentagon technology chief Emil Michael just one day after Michael issued an ultimatum to Anthropic. A framework agreement was reached within 24 hours. OpenAI accepted language allowing its AI to be used for “all lawful purposes,” despite Altman previously signaling solidarity with Anthropic and publicly rejecting the use of AI for domestic mass surveillance or autonomous weapons.