AI Agents as Insider Threats: Inside Microsoft's Bet on a Unified Security Stack

Microsoft is working on Project Perception, which involves using Guardian Agents to monitor and contain anomalous AI agents. However, experts are concerned that this approach may not be effective, as using agents to police other agents could potentially increase the risk rather than mitigate it. This is due to the unpredictable nature of the AI agents being monitored. The use of Guardian Agents is intended to be part of a unified security stack. Experts are warning that this method may multiply the risk instead of containing it. The project's goal is to address the potential insider threats posed by AI agents.