AI agents are starting to look less like ordinary software and more like insiders, with access to company data, apps, and systems. Microsoft is building a new security model around that shift, and we’re breaking down how it works as we sit down and sync up with Rocket IT’s weekly technology update.

In this episode, you’ll hear more about:

• How AI agents are starting to work across company data, apps, and systems
• How Microsoft is identifying and separating AI agents from the employees using them
• How new permissions can limit what an agent can access based on the task
• How businesses can require human approval when an agent takes higher-impact actions
• How Microsoft plans to monitor changing agent behavior and step in when something looks risky
• Why AI agents need clear boundaries as they take on more work inside a business

Video Transcript

A lot of the AI tools businesses have used so far have been pretty straightforward. You ask a question, generate some content, summarize a document, and a person decides what happens next.

AI agents can take that a step further. Depending on how they’re set up, they can work with company files, use applications, call other tools, and take actions on someone’s behalf. Once AI starts doing that kind of work, it needs a different level of oversight.

That’s the idea behind a new group of security controls Microsoft is introducing for AI agents.

One of the first challenges is simply knowing which agents are being used. Microsoft is expanding its security tools to identify 60 types of local AI agents, including products such as Claude Code, OpenAI Codex, and GitHub Copilot.

From there, the goal is to separate the employee from the agent working for them. Instead of an agent automatically using everything the employee can access, it can have its own identity and its own permissions. That gives the business a clearer record of what the employee did, what the agent did, and what the agent was actually allowed to reach.

Microsoft is also introducing something called Dynamic Permissions. The basic idea is to give an agent only the access it needs for the job it’s working on. Take an agent that helps process expense reports. It may need to read the report, check company policy, and work inside the system used to process the reimbursement. It probably doesn’t need access to every file, mailbox, or application the employee can use.

Those permissions can also change as the task changes. Access can be temporary, and certain higher-impact actions can still stop and wait for a person to approve them before the agent moves forward.

Microsoft is extending that same idea to what happens while the agent is working. An agent may start with the right permissions and then run into something unexpected, like bad instructions hidden inside a document or a task that starts moving in a different direction. New risk controls are designed to respond as that behavior changes instead of relying only on the permissions the agent received at the beginning. That could mean limiting what the agent can change, reducing what it can access, requiring human approval, or stopping it altogether if the risk increases.

There’s also a feature called Guardian Agents. These are designed to help security teams keep track of other AI agents, investigate why their risk may have changed, and apply the rules the organization has already put in place. That part is aimed at a future where a business may have a lot of agents working across different systems at the same time. Even if an organization never reaches that scale, the idea behind it is pretty simple: if an AI agent is working inside the business, you should know it’s there, know what it can access, and have a way to step in when necessary.

As these tools become more capable of working across company systems, the visibility and controls around them need to keep pace. None of this means AI agents are automatically a problem. The same access that creates risk is also what allows them to be useful. The goal is to give them enough room to do the job without giving them more authority than the job actually requires. If you’re exploring AI agents and aren’t sure where those boundaries should be, reach out to Rocket IT using the link in this video’s description. And to stay up to date on trending technology news, hit that subscribe button and the bell to catch us on next week’s episode of Sync Up with Rocket IT.