Microsoft debuts new security tools in Microsoft 365 E7 and its Project Perception system that treat AI agents as potential insider threats via risk assessments

Your AI agents will now face continuous risk assessments and potential restrictions, impacting their operational autonomy.

Microsoft has introduced new security tools within its Microsoft 365 E7 suite and Project Perception system designed to monitor and control AI agents, treating them as potential insider threats. These features continuously assess AI agent risk based on data access and behavioral deviations. When risk levels increase, the system can restrict an agent's capabilities, require human approval for critical actions, or deactivate it. Microsoft is also expanding its detection of AI agents on employee computers to 60 types, including popular tools like Claude Code and GitHub Copilot. The company's Project Perception system is being enhanced with "Guardian Agents" to investigate and address AI agent security incidents. These new capabilities are currently in private preview, with wider availability anticipated at Microsoft's Ignite conference in November. The E7 bundle, priced at $99 per user per month, also includes Microsoft 365 Copilot and other productivity and security tools.

This development signals a significant shift in how organizations will manage AI within enterprise environments, moving from a trust-based model to a rigorously controlled and monitored one. By integrating AI agent security into high-tier bundles like Microsoft 365 E7, Microsoft is making these advanced controls a standard offering for businesses seeking to deploy AI tools safely. This escalation in the AI security arms race, with competitors like Google and AWS also vying for dominance in this space, suggests that robust AI governance will become a key differentiator and a mandatory consideration for any enterprise AI strategy.

The introduction of continuous risk assessment and automated containment for AI agents directly affects businesses adopting AI, as it introduces new layers of operational overhead and potential friction. Companies will need to adapt their workflows to accommodate these new security protocols, potentially slowing down AI deployment but also mitigating the risks of autonomous AI actions. The market is being pushed towards solutions that prioritize security and control over unfettered AI autonomy, setting a precedent for future AI development and integration within corporate IT infrastructures.