OpenAI releases national security principles

OpenAI published a new set of National Security Principles on July 8 that spell out how the company says its AI models can and cannot be used by governments, intelligence agencies and law enforcement partners. The document says OpenAI will support some national security uses, particularly in defensive areas such as cyber and biosecurity, while drawing lines around surveillance, force decisions and other high-consequence uses. The company released the policy in a post on its website and in a separate PDF laying out the full principles. ### What did OpenAI actually release on July 8? OpenAI said on July 8 that it was publishing its National Security Principles “to offer transparency” into how it approaches government partnerships and national security uses of its technology. The company said the principles were the product of a cross-company process and that former U.S. national security official David Kris helped facilitate the work and provide independent judgment. (openai.com) The PDF says the framework is meant to govern “national security and law enforcement uses of AI” and reflects what OpenAI called an evolution in its approach as AI capabilities, safeguards and the threat environment change. The document says the company wants government uses of AI to benefit people, strengthen democratic governance, avoid concentration of power, and preserve human judgment and accountability in high-consequence settings. (openai.com) ### Which uses does the policy ban? The most explicit prohibition in the document is on “mass domestic surveillance.” The principles also say OpenAI will not support “high-stakes decisions — including decisions over the use of force — without appropriate human judgment and accountability.” OpenAI’s July 8 post says those restrictions are consistent with limits it had already written into an earlier government agreement: no use for mass domestic surveillance, no use to direct autonomous weapons systems, and no use for high-stakes automated decisions. (cdn.openai.com) The company said it believes it should be clear and transparent about those limits. ### Does this mean OpenAI is open to intelligence and military work? (cdn.openai.com) OpenAI said the principles apply to current and future national security and law enforcement partnerships, including existing work with the Department of War. The company said it is expanding work with the U.S. government and allied partners in “critical defensive areas,” particularly cyber and biosecurity. In the same July 8 post, OpenAI said it had established Trusted Access for Cyber partnerships in the past month with Australia, Canada, Japan, the Republic of Korea, France, Germany, Poland, the Netherlands and European Union institutions including ENISA. (openai.com) It also said it has a partnership with the UK government around cyber, testing and evaluation, and last month expanded access to its GPT-Rosalind model for select U.S. government and allied partners working on public health and biodefense missions. ### How does this fit with OpenAI’s broader rules? OpenAI’s existing usage policies, effective October 29, 2025, already say national security or intelligence uses require the company’s review and approval. Those policies also say OpenAI can withhold access when it believes that is necessary to protect its services, users or others. The new principles go further by setting out a public framework for what kinds of government uses OpenAI says it will support and what kinds it says it will reject. (openai.com) The document says democratic governments need access to powerful AI both to govern it responsibly and to defend against threats, but adds that such access must be constrained by law, civil-liberties protections and democratic oversight. (openai.com) ### Where can readers see the full policy? OpenAI posted the policy on its website on July 8 under a page titled “Our approach to government and national security partnerships,” alongside a separate PDF titled “OpenAI Principles for National Security Partnerships.” The company said the principles apply now to existing partnerships and to future government, national security and law enforcement work. (openai.com) (cdn.openai.com)