Key Takeaways
1. OpenAI updated its agreement with the U.S. Department of Defense to clarify limitations on domestic surveillance.
2. The update specifies that OpenAI systems cannot be used for monitoring U.S. individuals or using personal identifiable information.
3. The agreement aims to address concerns raised after OpenAI’s deal for its models in classified defense settings.
4. Critics worry that broad language in the agreement could allow for misinterpretation regarding surveillance practices.
5. OpenAI will work with the Department of War and other AI labs to discuss defense policies and operations.
OpenAI has recently updated its agreement with the U.S. Department of Defense, which it refers to as the “Department of War.” This update aims to make the limitations regarding domestic surveillance clearer.
New Terms for Clarity
According to the update released on March 2, 2026, OpenAI specified that its systems are not to be utilized for monitoring U.S. individuals, which includes obtaining or using commercially sourced personal or identifiable information. The company also noted that the Department of Defense confirmed that its service will not be employed by defense intelligence agencies such as the NSA, and any such usage would need a distinct agreement.
Response to Concerns
This update came after some backlash when OpenAI announced a deal to use its models in classified defense settings. Business Insider reported that Sam Altman, the CEO, shared an internal memo detailing that the company was collaborating with the Pentagon to “make some additions” to the agreement in response to worries about potential mass surveillance.
Before the March 2 update, OpenAI had highlighted three key “red lines” in its public description of the agreement, which included a prohibition on mass domestic surveillance. The contract language allowed for “all lawful purposes,” referencing existing U.S. legal frameworks for intelligence activities.
Criticism of Broad Language
Critics have raised concerns that relying on broad “lawful use” language and established surveillance authorities might leave openings for interpretation. Even with OpenAI’s claims that the contract and technical safeguards prevent mass domestic surveillance, doubts remain.
OpenAI’s public announcement outlines a cloud-only deployment approach and mentions that cleared personnel from the company will be involved. However, important operational information, such as the contract’s monetary value, the specific systems or models included, and the overall scope of the rollout, are not detailed in the company’s published summary.
OpenAI also stated that the Department of War intends to initiate a working group with leading AI laboratories, cloud service providers, and defense policy and operations leaders, and they expect to be involved in those discussions.
Source:
Link