OpenAI publishes principles for government and national-security partnerships
No mass domestic surveillance, no use of force without human judgment
Confirmed
The takeaway
On July 8, 2026 OpenAI published a framework for its government, national-security and law-enforcement work.
Status
- Claim
Confirmed
- Our reporting
- High confidence
- Importance
- 3 of 5
- Last verified
- 5 October 2026
Your AI and this story
- GPT-6.2 Astra38 days after its cutoff
- Claude Opus 5.58 days after its cutoff
- Gemini 4 Hydrogen99 days after its cutoff
- Grok 4.738 days after its cutoff
None of these four assistants can know about it. The closest, Claude Opus 5.5, stops 8 days before it.
Key facts
- Three things OpenAI ‘will not support’: mass domestic surveillance; high-stakes decisions, including over the use of force, without appropriate human judgment and accountability; uses that evade legal obligations, oversight or accountability (PDF)
- Excludes systems that ‘autonomously identify, select, and engage targets without appropriate levels of human authorization and human-defined operational parameters’, while saying human judgment ‘does not require a human decision on every discrete system action’
- Calls the US its home jurisdiction with ‘a distinctive role’ and says it will prioritise US allies, but that ‘alliance or partner status is not enough on its own’
- Applies like ‘a common law system’: documented applications of the framework are to build a body of ‘case law’
- Enforcement through contractual usage restrictions, model behavior specifications, audit logs, misuse classifiers and AI-enabled oversight tools (PDF)
- Developed with national-security lawyer David Kris and applies to existing Department of War work (search summary of OpenAI’s post; the post itself returned 403 to our fetcher)
- Allied trusted-access programs cited: Australia, Canada, Japan, South Korea, France, Germany, Poland, the Netherlands; UK cyber-testing partnership; GPT-Rosalind for US public-health and biodefense partners (Investing.com)
What happened
OpenAI posted a five-page PDF and a blog post setting out how it decides on national-security and law-enforcement deals. The framework rests on three commitments: government uses of AI that benefit people, that strengthen democratic governance and avoid concentrating power, and that keep appropriate human judgment and accountability in high-consequence settings. It names three uses OpenAI will not support. The date (Wednesday, July 8, 2026) comes from secondary coverage (Investing.com, Defense Tech Legal Network); the openai.com page could not be fetched directly.
Why it matters
This is OpenAI’s written red lines for military and police use, published while the US government was fighting Anthropic over similar restrictions. The wording leaves room for autonomous systems that act within human-set parameters. It is the reference point for later disputes over OpenAI’s Pentagon work.
Sources
4 sources from 4 sites. Numbers match the chips in the text.
4 sources: 2 primary, 2 press
Primary
- OpenAI: Our approach to government and national security partnershipsopenai.com, official
- OpenAI Principles for National Security Partnerships (PDF)cdn.openai.com, official
Press
- Investing.com: OpenAI publishes national security principles amid government AI tensionsng.investing.com, press
- Defense Tech Legal Network newsletter (July 13): OpenAI releases national security principlesdefensetechlegalnetwork.substack.com, press
Changes
- Filed