AI Notetakers on Your Board Calls: Which Ones Train on Your Data by Default, and How to Turn It Off

A board call used to have the people on the invite and nobody else. Now it often has a bot or two as well: the investor's Otter, your CFO's Fathom, the Zoom or Teams assistant someone switched on. Each one records, transcribes and stores the conversation in its own cloud.

Most owners worry about the obvious part, who can read the notes. The quieter question is whether the vendor uses your board conversation to improve its own AI. For some tools the answer is yes unless you switch it off. For others it is no, full stop. Here is which is which, straight from each vendor's own pages.

The short answer, tool by tool

Otter. Otter trains its own models on user recordings and transcripts after removing anything that identifies the user, and says this training runs automatically [1]. Enterprise workspaces are the exception: they are opted out of model training by default, and opting in goes through your Otter account manager [2]. We could not find a self-serve switch for this training on Otter's own pages for the individual and business plans, so if Otter is on your board calls, ask Otter in writing.

Fathom. Fathom uses de-identified customer data to improve its own models. You can switch that off in My Settings, and on Team Edition an admin can switch it off for every user at once in Organization Settings [3]. Fathom says its AI providers (Anthropic, OpenAI and Google) are not allowed by contract to train on your data [3].

Granola. By default Granola may use anonymized data to improve its models. The switch is in Settings, Preferences: turn off "Use my data to improve models for everyone." Enterprise workspaces are opted out by default and admins can enforce it for the whole company [4].

Fireflies. Fireflies says customer data is never used to train AI models, its own or anyone else's, and that its third-party vendors keep no copy of the data after processing [5]. Nothing to switch.

Zoom AI Companion. Zoom says it does not use your audio, video, chat, screen sharing, attachments or other meeting content to train its own or third-party AI models [6]. Nothing to switch.

Microsoft Teams with Copilot. Microsoft says prompts, responses and the data Copilot reads through your Microsoft 365 account are not used to train the large models behind Copilot [7]. Nothing to switch.

Google Meet with Gemini. Google says Gemini in Meet and the rest of Workspace does not use your content to train or improve Gemini or other generative AI models [8]. Nothing to switch.

Read. Read uses meeting data to improve its features only if you join its Customer Experience Program, which is off unless you opt in. You can opt in or out in your account settings [9].

Two things to keep in mind. "De-identified" and "anonymized" mean names are stripped, not that the content is gone, and board talk is often identifying on its own (a deal size, a product name, a hire). And the setting that matters is the one on the account of whoever brought the bot, which on an investor call is usually not you.

Turning it off is the easy part

A training opt-out does not change who can read the notes, where they are stored or how long they are kept. It also does nothing about consent.

Recording law is where most of the risk sits. A group of US states, including California, Florida, Illinois, Pennsylvania and Washington, require everyone on the call to agree to a recording, and an AI notetaker is a recording. Courts are taking that seriously: on August 13, 2026 a federal judge in California let the core wiretap and privacy claims against Otter go forward, finding the plaintiffs had plausibly alleged Otter acts as a third-party eavesdropper [10]. That ruling decides nothing about guilt yet, but it tells you the "a bot joined, nobody objected" approach is now a legal question.

Professional bodies are moving too. The New York City Bar's ethics committee says lawyers must tell clients and get their consent before an AI tool records, transcribes or summarizes their calls [11]. And Microsoft now detects outside bots trying to join Teams meetings, holds them in the lobby and makes the organizer admit them on purpose; admins can also block them outright [12].

What it does to the room

There is also a softer cost. When people know a bot is listening, they talk differently. They hedge, skip the honest version and save the real conversation for a side call. On a board or investor call, the candid part is the whole point. If you have ever had a director say "let's take this offline" the moment the recorder pinged, you have seen it.

A four-rule policy for small businesses

You do not need a long document. You need four decisions, written down and shared.

One tool, one owner. Pick a single notetaker for the company and set its training switch yourself, using the list above. Anyone recording a work meeting uses that tool or none.

Consent on the invite and out loud. The invite says the meeting will be recorded and transcribed by AI. The host says it again in the first thirty seconds. If anyone objects, the bot leaves.

Board, investor and client calls start with bots off. Turn one on only after everyone on the call agrees, and ask outside guests what their bot does with the recording before you admit it.

A bots-off label for candid work. Strategy sessions, one-on-ones and personnel talks get a plain "no bots" tag on the invite, so people stop moving the real conversation somewhere else.

Where your phone calls fit

Many board check-ins and investor updates still happen by phone. If your business phone system records and transcribes calls, the transcript stays inside your own account with no outside bot in the middle. That is how it works on Tonet: recordings and transcripts are never used for Tonet advertising or shared with ad platforms, recording is a per-number setting you control, and consent to record is still yours to get [13].

What does your team do today when an investor's bot shows up in the lobby? If the answer is "admit it and move on," this is a good week to write the four rules down.

Sources:

[1] Otter.ai: Privacy & Security. Otter's statement that it de-identifies user data before training its models and that training is automatic.

[2] Otter.ai Help Center: Enterprise Admin Controls Overview. Enterprise workspaces are opted out of AI model training by default.

[3] Fathom Help Center: Is Fathom secure?. Use of de-identified data for Fathom's own models, the My Settings and Organization Settings opt-out, and the contract terms with its AI providers.

[4] Granola Docs: Models and training. Default use of anonymized data, the Settings > Preferences switch, and the Enterprise default.

[5] Fireflies Guide: Responsible and secure meeting notetaking with Fireflies. Customer data is never used to train AI models; third-party vendors keep zero data.

[6] Zoom: An IT leader's guide to Zoom AI Companion. Zoom does not use customer audio, video, chat or other communications to train its own or third-party AI models.

[7] Microsoft Learn: Data, Privacy, and Security for Microsoft Copilot. Prompts, responses and data accessed through Microsoft Graph are not used to train foundation models.

[8] Google Meet Help: How Gemini in Workspace protects your data. Gemini does not use your Workspace content to train or improve Gemini or other generative AI models.

[9] Read AI: Privacy Policy. The opt-in Customer Experience Program and how to opt in or out.

[10] In re Otter.ai Privacy Litigation, N.D. Cal. (2026), via FindLaw. The August 13, 2026 order on Otter's motion to dismiss.

[11] New York City Bar, Formal Opinion 2025-6. Ethical issues in using AI to record, transcribe and summarize conversations with clients.

[12] Microsoft Teams Blog: Introducing smarter bot protection in Microsoft Teams meetings. Lobby detection of external bots and the admin policy to manage them.

[13] Tonet. Call recording and transcription inside your own account.