Microsoft adds risk scores and flexible permissions for AI agents

Microsoft is adding a way to score how risky an AI agent's behavior is, and to change what that agent is allowed to do based on the score. The feature is aimed at large companies. It points to where safety controls for AI tools are heading.

What changed

An AI agent is a program that can take actions for you, such as opening files or sending messages, not just answer questions. Microsoft's addition has two parts. The first is risk scoring, which rates how risky each agent's activity is. The second is flexible permissions. Instead of staying fixed, an agent's access can change as its risk rating changes.

Who it affects

The material describes this as an enterprise feature, so it is built for big organizations that run many AI agents. Most small businesses and individuals will not use it directly. Still, the idea matters to anyone whose data might be handled by an AI tool, including client, patient or customer information.

What is not known yet

The material does not say when the feature arrives, what it costs, or which Microsoft products include it. It also does not explain how the scores are calculated or what an agent loses access to when its risk rises. Those details will decide how useful it is.

What to do

If AI tools can reach sensitive data in your business, ask your vendors how they limit and monitor what those tools can do. Useful questions include:

• What data can the AI tool see and change?
• Is its access reduced when something looks unusual?
• Can we review a record of what it did?