Microsoft Launches 'Guardian Agents' to Monitor AI Agents

On Monday, Microsoft launched new security features to protect enterprises from the risks of AI agents, most notably 'Guardian Agents' dedicated to monitoring the agents used by the company.

The feature falls under the 'Project Perception' product launched by the company last June, which initially included three types of agents: a red team that attacks, a blue team that detects vulnerabilities, and a green team that remediates them, according to TechCrunch.

Microsoft added a feature to 'Defender' to detect up to 60 types of agents brought into the work environment by employees, including Claude Code, Codex, and OpenClaw, alongside tools that limit agents' access to data and systems, adjust their permissions based on the task, and require human approval when making high-risk decisions.

Hayete Gallot, Microsoft's security business chief, said that 'frontier models, regardless of their source, cannot be trusted' because they are unpredictable, having been trained to accomplish tasks rather than adhere to organizational norms. Satya Nadella, the company's CEO, wrote on his blog that companies cannot delegate responsibility for what these systems do on their behalf, calling for a separation between the supply of intelligence and the authority over its use.

What do these terms mean?

AI Agent: A program based on an artificial intelligence model to execute multi-step tasks on behalf of the user, which may access sensitive data and tools.

Insider Threat: A risk originating from within the organization, whether from an employee or a system with legitimate privileges acting in an unsafe or harmful manner.

Frontier Model: One of the most capable AI models of its generation, typically used for complex tasks.