Science & Tech · AI · 1 hr ago
Microsoft adds tools to monitor and restrict AI agents
Microsoft has introduced new security tools for workplace AI agents as companies use them to handle tasks and access sensitive information.
One new feature, Guardian Agents, is designed to monitor agents and is part of Microsoft’s Project Perception security product.
Microsoft CEO Satya Nadella says companies should use independent systems to oversee AI agents, rather than relying on safeguards promised by the companies that make them.
Microsoft security chief Hayete Gallot says powerful models can behave unpredictably because they are trained to complete tasks, not follow an organization’s rules or judgment.
Microsoft Defender can now detect up to 60 types of workplace agents, including Claude Code, Codex and OpenClaw.
The tools can also restrict agents’ access and require human approval for risky decisions, giving companies more control over what agents can do.
Microsoft introduced enterprise security features to monitor AI agents and limit what they can access and do.
The new Guardian Agents feature is intended to monitor agents used by a company.
Microsoft says powerful AI agents could pose insider threats because they may act unpredictably and have access to sensitive data and tools.
The company also added a Microsoft Defender feature that can detect up to 60 types of workplace AI agents.
Some agent actions can require human approval if they appear risky.
- Who
- Microsoft introduced the features for enterprises using AI agents.
- What
- The company added tools to monitor agents, detect them, restrict their access and adjust their permissions.
- When
- Monday, October 19, 2026.
- Where
- Not stated.
- Why
- Microsoft says agents can be unpredictable and may have access to sensitive data and tools, creating security risks.
This story does not have two clearly opposing sides.
Microsoft introduced Project Perception, with agents for attacking, detecting and fixing vulnerabilities.
Satya Nadella warned that companies should stop trusting AI model providers and use separate systems to monitor and control agents.
Microsoft introduced Guardian Agents and additional tools to detect and restrict workplace AI agents.
- Guardian Agents
- A feature intended to monitor a company’s AI agents
- Project Perception
- Introduced in June
- Microsoft Defender
- Can detect up to 60 types of agents
- Examples of detectable agents
- Claude Code, Codex and OpenClaw
- Risky decisions
- May require human approval
3 sources
What people are saying
Unfortunately, frontier models, regardless of their origin, cannot be trusted
We simply can’t outsource responsibility for what intelligence does on our behalf.