Microsoft says AI agents are like insider threats as frontier models cant be trusted | TechCrunch

▤ Summary

Microsoft introduced new security features aimed at protecting enterprises from AI agents that behave like insider threats. The company launched Guardian Agents as part of Project Perception to monitor agent activity and added new detection and permission controls in Microsoft Defender. It also said the new tools can identify dozens of agent types, restrict access to data and systems, and require human approval for risky actions. Microsoft framed the products as a separate security layer for agent oversight, even as it positions itself to sell those controls to enterprise customers.

▦ Linked Companies

Microsoft

$1B+