Microsoft Teams can now block AI note-taker bots: what it means for your meeting notes

Microsoft has announced a Teams policy that lets IT administrators auto-reject AI note-taker bots across an entire organization. It’s called BlockDetectedBots, it’s reported to roll out from August 2026, and it targets the bots behind some of the most popular note-taking tools — Otter, Fireflies, Read AI, Fathom. If your AI note taker joins meetings as a visible participant, this matters, because the switch isn’t yours to flip. It belongs to the IT admin of whichever organization hosts the meeting.

Everything below is based on Microsoft’s announcement and reporting as of mid-2026; specifics can shift before rollout, so check current documentation before making decisions. Full disclosure up front: we build Eidetic, a note taker that doesn’t use a bot, so we have a stake here. The facts stand on their own.

What Microsoft announced

As described so far, the policy works like this:

• Tenant-level, not per-meeting. An admin sets it once for the whole organization. Individual hosts don’t need to spot and eject bots by hand anymore.
• Detection plus auto-reject. When a known note-taker bot tries to join a meeting, Teams identifies it and refuses entry automatically. Reporting names Otter, Fireflies, Read AI, and Fathom among the detected services.
• Admin-controlled. It’s an IT policy. Employees — and external guests — don’t get a vote.
• Rolling out from August 2026, per the announcement. Exact timing and detection coverage may change.

The key detail is scope: the policy applies to meetings hosted in that tenant. That includes bots brought along by people outside the organization who were merely invited.

Why the bot backlash reached this point

This didn’t come out of nowhere. Anyone who takes calls has lived the awkward moment: three humans on a call and a fourth tile that says someone’s notetaker “has joined the meeting”. Nobody agreed to it; now everyone talks differently.

The institutional reaction has been building for a while:

• Zoom, Meet, and Teams already show consent prompts when recording starts, and hosts can already admit, remove, or block bot participants by hand.
• Several US universities have banned Read AI campus-wide after unwanted bots kept appearing in classes and internal meetings.
• The American Bar Association has advised keeping AI note-taker bots out of privileged legal conversations.
• The legal pressure is real for vendors too: Otter is facing ongoing privacy litigation, and Fireflies is facing biometric-privacy litigation.

BlockDetectedBots turns all of that ad-hoc resistance into a single org-wide default. The direction of travel is clear: platforms and institutions increasingly treat an uninvited recording bot as a liability, not a convenience.

What breaks if your notes depend on a bot

If you rely on a bot-based note taker, your notes stop working in any organization that enables the policy — silently. The bot tries to join, gets rejected, and you find out after the meeting that there’s no transcript.

The guest scenario is the one that stings. Your own company might never enable the block, but the policy follows the host, not you. A sales call on a prospect’s Teams tenant, a client kickoff, a job interview, a vendor review — exactly the meetings where your notes matter most are the ones hosted on someone else’s infrastructure, under someone else’s policy. There is no setting on your side that fixes it.

And this is Teams first, not Teams only. Hosts on other platforms can already block bots manually; a one-switch tenant policy is the pattern every platform now has a template for.

What still works after the block

Two approaches don’t involve a bot at all, so the policy can’t touch them. Here’s how the options compare:

Prices checked July 2026.

Option 1: Teams’ own Copilot notes

Teams has a native AI note taker: Copilot can produce recaps and notes without any third-party bot joining, so the block doesn’t apply to it. The catches: it requires a paid Microsoft 365 Copilot license and admin enablement, processing happens in Microsoft’s cloud, and it’s controlled by the host organization. If you’re a guest in someone else’s tenant, whether it runs — and who gets the notes — is their call, not yours. Licensing details change often; check current Microsoft docs.

Option 2: capture on your own machine

The other pattern sidesteps the meeting platform entirely: software on your own computer records your microphone and system audio, so it hears both sides of any call — Teams, Zoom, Meet, Slack, a browser tab — without anything joining as a participant. There is no bot to detect and no bot to reject, and it works identically for in-person conversations. We’ve written up how this works in how no-bot AI note takers work.

That’s the approach we took with Eidetic(again: our product). It’s a native macOS menu-bar app for Apple silicon that captures mic and system audio, shows a live transcript beside your own notes during the call, and runs transcription, editable summaries, and meeting chat on-device — nothing is uploaded, and it works offline. Unlike every cloud meeting service it’s a one-time purchase: €49 after your first 6 free meetings, no subscription. Honest limits: it’s single-user, Apple-silicon Macs only, and the local speech model struggles with mixed-language sentences.

One thing “no bot” does not mean: secret recording. Consent laws vary by country and US state, and courtesy doesn’t vary at all — tell participants you’re recording notes, the same way you’d announce any recording.

For IT admins: blocking bots without breaking people

Worth remembering before you flip the switch: almost nobody runs a bot because they want to surveil colleagues. They run it because notes are tedious, memory is unreliable, and follow-ups slip. Block the bots without offering an alternative and the need doesn’t disappear — it goes underground, into phone recordings and unsanctioned tools you can’t see.

The compliance-friendlier pattern is to point people somewhere: Copilot notes if your org licenses it, or a local, no-upload tool for everyone else. A note taker that processes audio on the employee’s own machine introduces no third-party cloud processor, no new data-processing agreement, and no data-residency questions — meeting content never leaves the device. Eidetic also exports to Obsidian and fires webhooks, so notes can flow into existing systems without a cloud middleman. If you’re auditing bot-based vendors right now, our rundown of Otter.ai alternatives covers which tools work without a bot.

FAQ

Does Teams have an AI note taker?

Yes — Copilot in Teams can generate meeting notes and recaps natively, with no bot joining. It requires a paid Microsoft 365 Copilot license and admin enablement, so many users don’t have it even though their org uses Teams. If that’s you, a no-bot app running on your own machine is the alternative that doesn’t need anyone’s licensing decision.

How do I set up an AI note taker in Teams?

Two routes. Inside Teams: ask IT whether your org licenses Microsoft 365 Copilot and has it enabled for meetings, then turn on transcription when you meet. Outside Teams: run a capture app on your own computer — Eidetic on macOS records mic and system audio and transcribes on-device, which keeps working even in orgs that block bots. Either way, tell participants you’re taking recorded notes.

How do I remove an AI note taker from a Zoom meeting?

As host, open the Participants panel, find the bot (it appears as a named participant like “Otter.ai” or “Fireflies Notetaker”), and remove it as you would any participant. To keep bots out in the first place, use the waiting room and admit only people you recognize, or restrict joining to signed-in users. Exact menu names shift between Zoom versions, so check current Zoom docs — and note that removing the bot doesn’t delete whatever it captured before removal.

Is there an AI note taker for in-person meetings?

Yes. Bots can’t attend a conference room, but tools that record through your computer’s microphone can. Eidetic handles in-person conversations the same way it handles calls: capture on-device, transcribe locally, summarize into the same searchable library — and because processing is local, it works offline. Get verbal consent from the people in the room first; recording laws apply offline too.