OpenAI adds new guardrails to Pentagon deal after US surveillance backlash

OpenAI says it has added new language to its agreement with the U.S. Department of Defense (which the company calls the “Department of War”) to make its restrictions on domestic surveillance more explicit. 

In an update dated March 2, 2026, OpenAI said the added terms clarify that its system is not to be used for domestic surveillance of U.S. persons, including by purchasing or using commercially acquired personal or identifiable information.

OpenAI also said the department affirmed the service will not be used by defense intelligence agencies “like the NSA,” and that any such use would require a separate agreement.

Why the revision is happening now

The update follows backlash after OpenAI announced it had reached a deal to deploy its models in classified defense environments. Business Insider reported that CEO Sam Altman posted an internal memo explaining the company was working with the Pentagon to “make some additions” to the agreement after concerns it could enable mass surveillance.

What the original framing looked like

Before the March 2 update, OpenAI’s published description of the deal emphasized three “red lines,” including no mass domestic surveillance, and also included contract language that allowed “all lawful purposes” while referencing existing U.S. legal authorities for intelligence activities.

Critics have argued that leaning on broad “lawful use” language and longstanding surveillance authorities can leave room for interpretation, even if OpenAI says the contract and technical controls prevent its systems from being used for mass domestic surveillance.

What’s still not disclosed

OpenAI’s public post describes a cloud-only deployment model and says cleared OpenAI personnel will be involved, but key operational details such as the contract’s value, the specific systems/models covered, and rollout scope are not listed in the company’s published summary.

What happens next

OpenAI says the Department of War plans to convene a working group with frontier AI labs, cloud providers, and defense policy/operations leaders, and that OpenAI expects to participate.