OpenAI published a set of national security principles on Wednesday that establish how the company intends to engage with government partners and restrict certain military and law enforcement applications of its technology. The company said the guidelines were developed as part of a cross-company effort with national security expert David Kris and are meant to govern both present and future national security and law enforcement collaborations, including work the firm currently conducts with the Department of War.
Key elements of the principles
OpenAI spelled out contractual guardrails intended to limit potentially harmful uses of its models. Among the explicit restrictions are:
- No deployment of OpenAI technology for mass domestic surveillance.
- No use of the company’s models to directly control autonomous weapons systems.
- No application of the technology for high-stakes automated decision-making.
The company framed these limits as part of a contractual approach that will apply across relevant government engagements.
Expanding partnerships on cyber and biodefense
OpenAI has broadened its trusted-access arrangements over the past month under its Daybreak cyber defense program. The list of partners in these Trusted Access for Cyber agreements includes Australia, Canada, Japan, the Republic of Korea, France, Germany, Poland, the Netherlands and EU institutions such as ENISA. Separately, OpenAI noted a partnership with the UK government focused on cyber testing and evaluation.
In June, the company announced expanded trusted access to its GPT-Rosalind model for select U.S. government and allied partners supporting public health and biodefense missions.
Context and related industry tensions
The publication of OpenAI’s principles comes against a backdrop of recent friction between the U.S. government and Anthropic. That dispute began in early 2026 when Anthropic declined to remove safety constraints on its Claude models for military use. In February 2026, President Trump directed federal agencies to stop using Anthropic’s technology, and Defense Secretary Pete Hegseth described Anthropic as a supply chain risk to national security.
Relations between Anthropic and the U.S. government eased in the weeks following a Commerce Department decision to lift most export restrictions on Anthropic’s advanced models in late June and early July.
What OpenAI says it will cover
OpenAI’s stated framework applies to both current and subsequent national security and law enforcement partnerships. The company characterized the principles as binding contractual measures that delineate permissible and prohibited uses, while also detailing its expanding trusted-access engagements with allied governments in cyber defense and targeted biodefense work.