Microsoft Project Perception Aims to Prevent AI Agents From Exposing Business Data

Microsoft is expanding Project Perception, an AI-powered security platform, to help enterprises manage the risks posed by AI agents. These agents can access company data, use software tools, and make independent decisions. While they can save time and increase productivity, they may also operate in ways that put sensitive information at risk.

Microsoft refers to these systems as "agentic insiders." According to Hayete Gallot, executive vise president of Microsoft Security, AI agents may exhibit surprising behavior as situations change, even if they have not been attacked or hacked.

Project Perception, which was first unveiled in July 2026, uses artificial intelligence to assist security teams in detecting and responding to cyber threats. It employs three types of artificial intelligence agents: red team agents, who look for security flaws, blue team agents, who look into suspicious activities, and green team agents, who assist in problem resolution. Microsoft now intends to utilize this technique to monitor AI bots that operate within enterprise networks. The platform will detect these agents, track their access permissions, monitor their actions, and evaluate the threats they may provide to enterprise systems. 

Microsoft is also planning to introduce “Guardian Agents.” These specialised AI agents will supervise other agents and step in when their actions become risky. For example, they could reduce an agent’s access permissions or block an unsafe action before it causes damage.

Microsoft CEO Satya Nadella has also warned that powerful AI models should be treated as potential insider security risks, especially when they can access confidential data or perform important business tasks.

He has called for strict access limits, detailed activity logs, and controls that allow administrators to stop AI agents when needed. This would help businesses maintain control over AI systems instead of relying on the models to manage their own behaviour.

Microsoft also says future security tools must explain why an AI agent makes certain decisions, rather than only recording what it does. As businesses use more AI agents, tools like Project Perception could help them monitor activity, protect sensitive data, and reduce the risk of unexpected AI actions.

Information referenced in this article is from Redmond