Zero Trust for AI Systems: Why Authorization Can’t Live Inside the Model

The Coalition for Secure AI's Preparing Defenders for AI workstream (WS2) released a paper in September 2026 titled "Zero Trust for AI Systems," a guide for applying zero trust security design to AI systems as a new category of insider threat. The paper argues that authorization cannot live inside the model itself, and the release follows several frontier incidents over the summer, according to CoSAI.