Does Your AI Notetaker Train on Your Data? 2026 Guide

The short answer: Yes, several popular AI notetakers train on your meetings by default. Otter.ai, Granola and Fathom use de-identified customer conversations unless you opt out (and Otter documents no self-serve opt-out), and Sembly and Wispr Flow do the same on their lower tiers. Fellow, Fireflies.ai, Zoom AI Companion and Hedy state they do not train on your content. Hedy goes one step further: speech recognition runs on your device, and with Local AI on supported hardware the AI does too, so nothing has to reach a vendor to be trained on.

Why this became a 2026 story

The training-on-your-data question stopped being theoretical when the litigation started. Otter.ai is the defendant in a consolidated federal class action, In re Otter.AI Privacy Litigation, in the Northern District of California. It alleges that Otter recorded conversations without all-party consent and kept them to train its models. On 13 August 2026 the court let most of the claims proceed, holding that a vendor that independently collects, retains and uses recordings for its own commercial purposes can be treated as a third-party eavesdropper. That is a pleading-stage ruling, not a finding of liability, and Otter has denied the allegations. Read the full analysis and the lawsuit tracker.

A separate thread runs through Fireflies.ai, which faces Illinois biometric-privacy litigation over how it captures voiceprints. That is a consent-and-biometrics problem, not a “trains-by-default” one, since Fireflies states it does not train on meeting content. It lands in the same place for a buyer: read the data terms before you connect a tool to your calendar.

The EU context adds weight. Article 50 of the EU AI Act, which covers transparency when people interact with AI, began to apply on 2 August 2026. The Digital Omnibus on AI (Regulation (EU) 2026/1744) moved the high-risk workplace-system obligations to 2 December 2027. Emotion recognition at work is a separate matter: it is prohibited outright, not merely high-risk, and has been since February 2025. See our EU AI Act guide.

Who trains on your data

Here is where the tools stand, based on their published terms as we checked them on 14 October 2026. Our fact sheet is the single source of truth and covers all 28 tools.

The pattern worth noticing: the strongest protections often cost something. Any Granola user can switch training off, but only its $35/user Enterprise tier has it off by default and enforceable by an admin. Notta lists “no AI training” as an Enterprise feature, which implies lower tiers may be used. So the people most likely to be training fuel, individuals and small teams on entry plans, are also the people with the least control. If you are on a free or starter plan with one of these tools, assume your meetings are improving someone’s model.

How to check and change your settings

You don’t have to take a marketing page’s word for it. Three concrete steps:

Open the data-controls section of your account. Most tools that train by default put the toggle under a “Data,” “Privacy,” or “AI improvement” heading in account settings, not in the main preferences. With Fathom the control lives in My Settings, and a Team Edition admin can opt out everyone. Otter’s privacy page describes no self-serve control, so ask in writing; its policy lists withdrawing consent and objecting to processing among your rights. Do this for every account in your organization, not just your own. Shadow accounts that staff set up are the real exposure.

Read the data-processing terms, not the homepage. “We take privacy seriously” is not a commitment. “We do not use customer content to train our models,” written into a DPA or BAA you can sign, is. Fellow, Fireflies.ai and Zoom make that commitment in their terms, which is a meaningfully stronger position than a settings toggle a vendor can change in a future update.

Match the tool to the sensitivity of the meeting. A weekly standup and a board session carrying material non-public information do not need the same posture. For the sensitive end, prefer tools where training is structurally impossible.

Where your data is stored

Residency is a separate question from training, and it matters for GDPR and sector rules. From the same 14 October check:

Our GDPR guide covers subprocessors and no-training commitments in more depth.

Why on-device tools sidestep the problem

The cleanest answer to “does this tool train on my data” is to use a tool that never has your data to begin with. On-device processing changes the question from a policy promise to an architectural fact. If the audio is transcribed and analyzed on your own machine and never uploaded, there is no vendor-side store to retain, no transcript to subpoena, and nothing to feed a training run.

MacWhisper is the strictest version of this: a Mac app running local Whisper models with no cloud component for transcription. It is transcription-first, so you trade away live summaries and coaching, but for a record-only use case it carries almost no vendor risk.

Hedy takes the on-device approach further while keeping the AI features. Its speech recognition always runs on your device. Since the May 2026 Local AI release, the summaries, notes, chat and live coaching can run on your device too, on Apple Silicon Macs, capable Windows GPUs, iPhone 15 Pro and newer, and Android phones with a supported GPU. If you choose Hedy’s cloud AI instead, the analysis runs in its EU or US cloud, and its AI providers are contractually prohibited from training on your content. Hedy completed SOC 2 Type I in April 2026 and signs a HIPAA BAA on its Enterprise plan (200-seat minimum). It sits at #1 in our privacy guide and our overall rankings.

None of this means cloud tools are unusable. Fellow, Fireflies.ai and Zoom commit not to train on your content, and Fathom, which has a strong free tier, lets you opt out in one setting. For many teams that is enough. The point is narrower: the default behavior of the category leans toward harvesting, the opt-outs are uneven and sometimes paywalled, and the only way to know where you stand is to check the data-controls settings yourself and read the terms you are agreeing to. For the meetings that would genuinely hurt to leak, an on-device or contractual no-training tool is the baseline.