Microsoft adds new security tools to monitor AI agents at work

In short: Microsoft launched new security features that treat workplace AI agents like possible “insider threats” and aim to monitor and control them.

What happened

Microsoft announced new security features focused on AI agents. An AI agent is software that can take actions for you, like looking up files, running tasks, or using other tools, not just answering questions.

The most notable new feature is called Guardian Agents. Microsoft says it is meant to watch the AI agents a company uses and flag risky behavior. Guardian Agents is part of a Microsoft product called Project Perception, which the company introduced earlier this year.

Microsoft’s CEO Satya Nadella has been warning that the biggest risk may come from the agents companies already use, not just outside hackers. Microsoft’s security chief Hayete Gallot said “frontier models” (today’s most capable AI systems) cannot be trusted because they are unpredictable. She said these systems are trained to finish tasks, not to follow a company’s rules and good judgment.

Microsoft also added an agent detection feature to Microsoft Defender. Microsoft says it can spot up to 60 types of agents that employees might bring into the workplace. The company also added controls that can limit what agents can access and can require human approval when an agent appears to be doing something risky (like a manager sign-off before a large purchase).

Why it matters

More workplaces are letting AI tools touch sensitive data and perform real actions. If an agent makes a bad choice, the impact can look less like a typo and more like giving the wrong person a staff badge and a master key. Microsoft is pushing companies to use separate oversight tools, rather than relying only on promises from the AI providers.

Source: TechCrunch AI