The federal government’s AI testing body has been told to keep its findings under wraps. National Cyber Director Sean Cairncross directed the Center for AI Standards and Innovation, known as CAISI, to cease issuing public reports on its assessments of advanced AI models, citing national-security concerns.
The directive aligns with a June 2, 2026 executive order focused on promoting advanced AI innovation while tightening security protocols around frontier models.
What CAISI does, and who it works with
CAISI is the rebranded successor to what the Biden administration called the US AI Safety Institute. The name changed, but the core mission stayed roughly the same: test advanced AI models for risks in areas like cybersecurity and biosecurity before they reach widespread deployment.
The agency maintains voluntary testing agreements with the biggest names in AI development. OpenAI, Anthropic, Google DeepMind, Microsoft, and xAI all participate in the program, submitting their frontier models for national-security evaluations.
