OpenAI sets principles for national security AI work

OpenAI says governments are beginning to use advanced AI systems in increasingly sensitive areas, including national security, and that this shift requires clearer rules for how AI labs, public institutions, and civil society should work together. The company frames the issue as a balance: democratic governments should be able to use AI for legitimate defensive purposes, but those uses must not weaken public accountability or remove meaningful human judgment from critical decisions.

The company argues that frontier AI can help democratic societies protect people, defend critical infrastructure, improve public services, and respond to emerging threats. It highlights cyber defense and biological security as areas where AI could give defenders a real advantage, for example by helping detect cyberattacks or support public health and biodefense work. At the same time, OpenAI says powerful AI must be deployed in ways that reinforce the rule of law and democratic institutions, rather than concentrating power in the hands of a few agencies, companies, or automated systems.

OpenAI is publishing a set of National Security Principles to explain how it approaches government partnerships and sensitive uses of its technology. The principles are meant to provide more transparency as the company expands work with the U.S. government and allied partners. OpenAI says the framework was developed through a company-wide process involving teams across research, safety, policy, and government partnerships. It also brought in David Kris, a national security expert, to help facilitate the process and provide independent judgment.

A major focus is defensive cooperation with governments and allied institutions. OpenAI says it is expanding its work in cyber defense through its Daybreak program, including Trusted Access for Cyber partnerships with Australia, Canada, Japan, South Korea, France, Germany, Poland, the Netherlands, and European Union institutions such as ENISA. It also points to a growing relationship with the UK government around cyber, testing, and evaluation. These partnerships are presented as examples of how the company wants to support national security missions while keeping usage boundaries clear.

OpenAI says it is taking a similar approach in biosecurity. The company recently expanded trusted access to GPT-Rosalind, a model aimed at public health and biodefense support, for selected U.S. government and allied partners. In practical terms, this suggests OpenAI wants its models to help vetted public-sector users with defensive biological risk work, while limiting access to sensitive capabilities and keeping these partnerships under defined controls.

The principles also apply to OpenAI’s current and future national security and law enforcement partnerships, including existing work with the Department of War. OpenAI says earlier contractual restrictions for that agreement remain aligned with the new principles. Those restrictions include no use of OpenAI technology for mass domestic surveillance, no use to direct autonomous weapons systems, and no use for high-stakes automated decisions. The company presents these limits as examples of how it distinguishes between defensive uses of AI and applications that could undermine rights, accountability, or human control.

OpenAI also says that companies should not decide the most consequential questions about government AI use on their own. It argues that major decisions about AI in national security should be settled through democratic processes, with companies helping inform the debate rather than replacing lawmakers or public oversight. OpenAI says it supports legislation that would establish safeguards for some of the highest-risk military uses of AI, especially around domestic surveillance, autonomous weapons, and other applications that directly affect U.S. national security.

The broader message is that OpenAI wants to normalize certain government and national security partnerships, especially with democratic allies, while drawing public boundaries around uses it considers unacceptable or too risky without safeguards. The announcement is not only a policy statement; it is also an attempt to shape how governments, companies, and the public discuss the role of frontier AI in defense, law enforcement, cyber resilience, and biosecurity as these systems become more capable.