type: timeline_event
By March 3, 2026, OpenAI had revised its Pentagon deal in response to mounting criticism over surveillance loopholes in the original contract language. CEO Sam Altman acknowledged publicly that the company "shouldn't have rushed" the announcement, calling the rollout "opportunistic and sloppy" even while defending the underlying decision to strike an agreement with the Department of Defense.
The revised contract added new explicit language stating that "the AI system shall not be intentionally used for domestic surveillance of U.S. persons and nationals." However, critics and legal analysts identified a significant remaining problem: the revised deal still contained a reference to Executive Order 12333, the Reagan-era framework governing foreign intelligence collection that has historically been used to authorize broad collection activities with limited oversight. Analysts argued that the E.O. 12333 carve-out meant that certain intelligence components of the Pentagon could potentially conduct surveillance of Americans' data under its authorities while remaining technically compliant with the contract.
OpenAI's head of national security partnerships Katrina Mulligan offered a technical safeguard argument, noting that "by limiting our deployment to cloud API, we can ensure that our models cannot be integrated directly into weapons systems, sensors, or other operational hardware." Critics countered that cloud API access to classified networks was itself the point of concern, since it would allow AI analysis of whatever data the military fed into the system.
MIT Technology Review characterized OpenAI's deal as "what Anthropic feared"—an agreement that on its face included the same restrictions Anthropic had sought but embedded those restrictions within language that legal experts said preserved meaningful flexibility for surveillance activities. The full text of the contract was not released publicly, making independent verification of the safeguards impossible. The episode exposed the limits of voluntary AI safety commitments in defense contracting when the companies seeking those contracts face severe competitive and regulatory pressure to reach agreement.