Microsoft: frontier models cannot be trusted.

Microsoft CEO Satya Nadella said earlier this month that we should assume that all AI models are “compromised,” and now Microsoft is going one step further. “Unfortunately, frontier models, regardless of their origin, cannot be trusted,” says Hayete Gallot, executive vice president of Microsoft Security. Microsoft is naturally positioning its new Project Perception as a solution to securing AI models. It’s an agentic security system designed to help cybersecurity professionals secure and control AI agents. [Link: Securing the agentic insider - The Official Microsoft Blog | https://blogs.microsoft.com/blog/2026/10/19/securing-the-agentic-insider/ | The Official Microsoft Blog]