Can Zoom or Teams detect AI note-takers?

Zoom, Teams and Google Meet are, from a detection standpoint, systems that manage participants. They know who connected, through what client, when they joined and left, who is sharing a screen, and whether their own recording function is running.

That is the extent of it. A meeting platform is an application on your computer. It does not enumerate your other applications, and it has no visibility into what else is reading your audio output.

So the answer splits cleanly depending on how a tool is built.

Tools like Otter.ai and Fireflies.ai capture meetings by joining them. Their assistant connects as a guest, receives the stream as a participant, and consequently appears in the attendee list under a name like “Notetaker”. Hosts can see it, question it, and remove it. Some organisations block unknown external participants entirely, which is why these tools sometimes fail to join at all.

This is not a flaw. It is how the product delivers what it promises: a reliable capture that works for the whole team regardless of whose laptop is in the meeting. Visibility is the cost of that.

A local assistant takes a different route. The call audio is already arriving at your computer and playing through your speakers. Software running on your machine can read that output and your microphone without touching the meeting connection at all.

From the platform's perspective, nothing has changed: one participant, you, with no extra connection and no second client. There is no attendee to list, because none was ever created.

Screen sharing. This is the detail most discussions miss entirely.

Not joining the meeting keeps a tool out of the participant list and does nothing about what happens when you present. The moment you share your screen, everything drawn on it — including an assistant's overlay — is in the stream the other side receives. On a demo or a technical review, where screen sharing is most of the call, that is the whole exposure.

Some tools handle this and some do not. Both macOS and Windows provide a way for a window to exclude itself from screen capture at the operating system level; Blackbox uses it, which is why its overlay stays on your display and is absent from the shared stream. If you are evaluating any tool in this category, this is the property to test — share your screen in a test call and look at the recording.

Here is the part worth being direct about. Everything above describes what a platform can observe. None of it describes what you are allowed to do, and the two get conflated constantly.

Recording laws vary. Some jurisdictions require consent from one party to a conversation; others require every party. On a call spanning several countries, the strictest rule in the room is usually the one that matters. Employment contracts and client agreements frequently add requirements of their own, and many enterprises have explicit policies about AI tooling on calls involving their data.

A tool being invisible to the platform changes none of that. If you need consent, you need it whether or not anything shows up in the attendee list. The sensible position is to know your own obligations first, and treat technical visibility as a separate, lesser question.

There is a real distinction between recording a conversation and reading your own notes during one. A tool that captures and stores the meeting is doing the first. A tool that surfaces an answer from documents you already own, on your screen, while you speak, is closer to the second — nearer to having the whitepaper open in another window than to running a tape recorder.

That distinction is worth understanding, and it is not a licence. Your policy, your contracts and your jurisdiction still decide. The architecture only decides what the platform can see.