OpenAI publishes national security partnership principles

OpenAI published a document setting out principles for how it will engage with governments on national security and law enforcement work, as it expanded partnerships with the US government and allied states in areas including cyber defence and biosecurity. The company said the framework reflected “an evolution” of its approach, developed with input from national-security expert David Kris and from listening sessions across research, safety, policy and government-partnerships teams.

The document commits OpenAI to supporting government uses of AI that “benefit people,” “strengthen democratic governance and avoid concentration of power,” and “preserve appropriate human judgment and accountability” in high-consequence settings. It sets three explicit limits: no support for mass domestic surveillance, no high-stakes decisions — including on the use of force — made without meaningful human judgment and accountability, and no use that helps evade legal oversight. Targeting-related military uses are not ruled out categorically, but the document says systems performing “critical targeting functions” bearing directly on force decisions require heightened scrutiny.

The principles carry no binding legal force and leave OpenAI itself to judge compliance case by case, comparing the process to case law that “will become clearer through documented applications” over time. The publication coincided with reporting on an expanded OpenAI for Government initiative, including contracts with US defence and intelligence customers, and followed earlier moves by the company into national-security work after historically restricting military use of its models.