OpenAI frames its national security work with public principles.

OpenAI outlines its national security principles as it expands cyber and biosecurity partnerships, including GPT-Rosalind access, with U.S. and allied governments.

OpenAI is publishing a series of principles governing its partnerships with governments and the use of its technology in national security. The move, presented as a transparency effort, involved national security expert David Kris for an independent review, as well as internal consultations.

These principles come as the company expands its work with the U.S. government and allied partners on two defensive fronts: cyber and biosecurity. As part of its Daybreak cyber defense program, OpenAI states it has forged "Trusted Access for Cyber" partnerships with Australia, Canada, Japan, South Korea, France, Germany, Poland, the Netherlands, and European Union institutions like ENISA, in addition to cooperation with the British government on testing and evaluation. On the biosecurity front, access to its GPT-Rosalind model was expanded last month to U.S. and allied public sector entities for public health and biodefense missions.

The text also covers existing contracts, including the one with the Department of War. OpenAI reiterates the restrictions set during that agreement: no domestic mass surveillance, no control of autonomous weapon systems, no high-stakes automated decisions. The company also states it supports legislative initiatives aimed at regulating the most sensitive military uses, believing that these arbitrations fall within the democratic process and that its role is to inform these decisions, not to make them in place of institutions.