According to Microsoft’s official blog, Microsoft expanded Project Perception with continuous agent risk assessment, adaptive permissions, and threat containment to treat AI models as insider risks.