Block AI Notetakers: How to Keep Uninvited AI Notetakers Out of Your Meetings

Most AI notetakers work by sending a bot into the meeting as an extra participant. Because the bot has to get through the door, the host and the administrator have a say. Each of the three large platforms added controls for this in 2026.

Microsoft explains why. Its documentation says external meeting assistants may record or transcribe meetings without participant awareness and may store meeting data in third-party systems outside the organization's compliance boundaries.

Some organizations decided not to wait for the platforms. UC Today reports that the University of Washington blocked Read AI from its Zoom and Teams integrations on 21 January 2025, Chapman University followed on 13 August 2025, and UC Riverside restricted all non-native AI bots on 7 October 2025.

Settings and defaults change, and an administrator may have changed yours. Treat the table as a list of things to look for in your own admin console. The sources were read on 12 November 2026.

Two rows describe the platforms' own recording and AI features. They are included because people often assume those notices also cover a third-party bot, and they do not.

Cornell's IT guidance describes the path. When someone creates a notetaker account, some sign-up prompts grant the tool access to the person's calendar and permission to join meetings automatically. The person may not realize what they agreed to.

From then on the bot joins on that person's behalf, including meetings they do not attend. This is why removing a bot from one meeting changes little. It will be back for the next one until its owner changes the setting or revokes the access.

Cornell also notes a trap in cleaning up. Deleting the app from Microsoft's My Apps list does not revoke its access. The owner has to open the app's management page and revoke consent, and some tools appear there under a different name.

A July 2026 survey reported by Stacker suggests how common this has become. Of 500 employed US adults, 33.4 percent said an AI notetaker had been in their meetings. Among that smaller group, 34.7 percent said they were always asked permission and 25.1 percent said they never were. The survey was sponsored by a plaintiff-side law firm, and the notetaker questions went to a subsample of 167 people.

Microsoft's documentation describes a meeting policy called Manage external bots and their access to meetings. It sits in the Meeting join and lobby section of meeting policies in the Teams admin center. The default option is to require approval before a detected bot joins.

With that option on, participants detected as bots are placed in the lobby regardless of how the meeting's lobby is configured. An organizer has to admit them explicitly and is reminded of the risk before doing so. Microsoft recommends allowing only organizers and co-organizers to admit from the lobby, so that another presenter cannot wave a bot through.

Sources differ on whether Teams can block outright. UC Today reported in May 2026 that administrators can set a policy to block external bots automatically at the tenant level. The Microsoft page we read lists two options, detect with approval or do not detect, and no block option. Check your own admin center for what your tenant offers.

Microsoft lists two limits. Some external bots may not be detected, and a human participant may occasionally be labeled as a bot.

This article found no official Zoom support page describing a single account-wide switch that blocks every third-party notetaker. The controls that exist work at the door: a waiting room so the host admits each participant, and the host's ability to remove a participant. Cornell's guidance is to remove an uninvited bot as you would any uninvited guest.

A stronger option may be available through Zoom's support team. Virginia Commonwealth University announced that from 12 March 2026, Zoom support would allow only authorized applications into meetings hosted on its account and block all others. The university said the earlier approach of blocking individual apps could not keep up with the number of new ones.

Zoom's own AI features are handled differently. UC Today reports that Zoom's AI Companion must be started by a host or co-host and shows a visible indicator. Zoom's recording consent prompt, which participants accept or leave, applies to recordings made through Zoom. It does not appear for a third-party bot's separate recording.

UC Today reported on 25 March 2026 that Google Meet now sorts join requests into two queues. Requests Google flags as risky, including bots and people whose identity cannot be verified, go into one queue and are automatically denied unless the host intervenes. People on the invitation or in the host's organization go into the other.

For Google's own tools, an administrator can require explicit consent. A Workspace update dated 5 May 2026 describes a setting that makes participants actively agree before automatic note-taking, recording or transcription begins. It is off by default and can be enabled for a domain, an organizational unit or a group.

Google's help page says participants get a notification when a recording starts or stops. That notice covers Meet's built-in recording. A third-party bot that was admitted to the call records on its own system, outside that notice.

Some notetakers never join the meeting. They capture the audio on the participant's device. The law firm Baird Holm describes a proposed class action filed in July 2026 against Granola, which alleges the tool captures audio directly from people's computers. That case has no ruling in the sources reviewed.

A tool like that is invisible to the lobby. No participant appears, so there is nothing for Teams to label or for a Zoom host to remove. The same is true of a phone on a desk or a screen recorder.

The other gap is human. A lobby control only works if the person admitting participants reads the names. Microsoft's own best-practice list includes educating organizers to be careful with lobby admission decisions.

Blocking deals with the tools you do not want. The tools you do allow raise the question of consent. A July 2024 chart by the law firm Kilpatrick lists 13 states that require every party to a private conversation to agree before it is recorded.

A pending case shows how a court may look at it. In the litigation against Otter, the company argued that a participant had consented because he could see the notetaker in the meeting. On 13 August 2026 the court said that even if the notetaker appeared as another participant, it could not infer that he knew his communications were being recorded, transcribed and retained. The ruling came at an early stage and decided nothing on the merits.

The practical reading is that a name in the participant list is weak notice. Telling people in words, before the tool starts, and noting their answer is stronger.

A written policy reaches the device-level recorders and the human errors that platform settings cannot. The points below are a starting outline, and counsel should review any policy that touches recording law.

Keep it short enough that people read it. A one-page rule that staff follow does more than a long document nobody opens.

• List the notetaking and recording tools staff may use, and say that all others are not allowed in company meetings.
• Tell staff not to sign in to a notetaker with their work Microsoft or Google account, as Cornell advises.
• Require an announcement before any recording or AI transcription starts, including recording on a personal device.
• Say who admits participants from the lobby and that unknown bots are denied.
• Mark meeting types where no notetaker is allowed, such as legal, personnel or board discussions.
• Keep a dated note of who was told about a recording and what they said.