Microsoft Unveils Dynamic Security Controls for AI Agents
Microsoft introduced new defenses for AI agents, treating them as potential insider risks. The features enforce least-privilege access and dynamically restrict access, block actions, or require manual approval based on real-time risk.
On October 19, Microsoft introduced security features for AI agents to address unintended actions from flawed logic or malicious input. The update includes Dynamic Permissions, Dynamic Agent Risk, and Dynamic Conditional Access to adjust privileges or halt operations. Defender for Endpoint also expanded support to 60 local agent types, while Project Perception added investigation tools, keeping critical approvals under human control.
Key takeaways
- Dynamic Permissions and Conditional Access adjust privileges based on task needs, restricting access or requiring manual approval as risk changes.
- Defender for Endpoint expanded support to 60 local agent types, while Project Perception added response tools under human oversight.
Source：iThome Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human. Translation checked by AI.