The question
What we told them
Short answer: you probably would not know. Not today.
On 29 Oct 2026 Microsoft said Defender can now discover local AI agents on Windows and macOS machines. It covers more than 50 supported agents and their connected tool servers. Runtime protection against malicious instructions is in preview on Windows. Both come from Microsoft's own announcement.
Read that the other way. A vendor does not build a finder for something that is rare. Coding helpers and small agents are already on work machines, installed by people who wanted to get something done faster.
The risk is not that staff use AI. We want them to. The risk is that an agent runs with the access of whoever installed it: their files, their logins, their email. Nobody reviewed it. Nobody decided what it may touch.
Our opinion: a detection tool is the second step. The first is a list. Ask every person which AI tools they use on work devices, and make it safe to answer honestly. If people fear being told off, the list will be short and wrong.
Then decide three things. Which tools are allowed. What each may access. Who says yes to a new one. If you run Microsoft Defender, check what your plan includes before you count on the new discovery.
Build one approved list of AI tools this week. Ask staff what they run, write down what is allowed and who decides, then check devices against it. You cannot manage agents you have not counted.