OpenAI published formal principles governing how governments can use its AI systems for national security work, establishing clear restrictions on surveillance, autonomous weapons, and high-stakes automated decisions.

The principles come as the company expands partnerships with the U.S. government and allies through its Daybreak cyber defense program. In the past month, OpenAI established Trusted Access partnerships with Australia, Canada, Japan, South Korea, France, Germany, Poland, the Netherlands, and EU institutions including ENISA.

The company also announced expanded access to its GPT-Rosalind model for select U.S. government and allied partners supporting public health and biodefense missions. OpenAI has an existing partnership with the UK government around cyber defense, testing, and evaluation.

Key restrictions for government use

The principles prohibit three specific uses of OpenAI technology by government partners:

  • Mass domestic surveillance
  • Directing autonomous weapons systems
  • High-stakes automated decisions without human oversight

These restrictions apply to current and future partnerships, including OpenAI's existing work with the Department of War announced earlier this year.

OpenAI engaged national security expert David Kris to help develop the principles through a cross-company process involving listening sessions with employees from research, safety, policy, and government partnership teams.

The company said democratic societies should be able to use AI for cyber defense, biological security, protecting critical infrastructure, and delivering public services. But it emphasized that AI deployment must reinforce democratic accountability, meaningful human judgment, and the rule of law.

"Many of the most consequential questions about AI in government, especially in national security, should be answered through the democratic process," OpenAI stated. "The role of companies like ours is to help inform those decisions, not make them alone."

OpenAI supports legislative efforts to establish safeguards around high-risk military AI uses, including domestic surveillance and autonomous weapons systems. The principles document is available as a PDF on the company's website.