What did OpenAI change in its Pentagon deal?
New language to limit surveillance use
OpenAI agreed to amend language in its contract with the U.S. Department of Defense to add stronger protections against the company’s tools being used for mass domestic surveillance. The company’s CEO has publicly acknowledged the original agreement had poor optics, and the amendments are intended to make explicit limits on how the technology can be used.
What the changes do
- They add explicit prohibitions on using OpenAI’s systems for mass surveillance of U.S. persons.
- They clarify that certain intelligence agencies are excluded from immediate access under the current terms, pending separate arrangements and further legal reviews.
- The company has committed to more explicit wording around acceptable uses to reassure customers, employees and the public.
Immediate consequences
- Backlash and user behavior: the announcement provoked a strong consumer and employee response, with a sharp spike in uninstalls of the ChatGPT mobile app as some users reacted to the perceived military ties.
- Product and policy moves: OpenAI has promised additional user-safety features, such as options for users to designate trusted contacts who can be alerted in crisis situations, and it has signaled a willingness to be clearer about operational boundaries.
- Broader sector impact: the dispute amplified regulatory and political attention on how frontier AI firms sell technology to governments, and it helped competitors and rivals draw users away.
Why it matters
The amendment is an attempt to balance two pressures: commercial opportunities selling to government agencies and public concern about civil liberties and company ethics. How effectively OpenAI enforces the new restrictions — and how transparent it is about oversight and auditing — will shape public trust and how other AI vendors negotiate similar deals.