The AI Notetaker in Your Next Meeting Could Be a Wiretapping Problem

AI notetakers have moved from a novelty to a default in the space of about two years. Otter, Fireflies, Granola, Read AI, and the built-in AI companions now shipped with Zoom and Microsoft Teams sit in a large share of business video calls, quietly transcribing, summarizing, and in some cases recording the audio itself. Usually one person turns the feature on — a salesperson who wants a clean call summary, a manager who does not want to take notes during a client meeting — and everyone else on the call either notices a bot join or does not notice anything at all. That casualness is exactly what a wave of 2026 lawsuits is now testing, and the early results suggest businesses have been treating a legal question as a workflow preference.

The core issue is consent, and it is more specific than most people assume. A meaningful number of U.S. states, and Canadian provinces with their own wiretap-style statutes, require every participant on a call to consent before it can be recorded — not just the person who scheduled it. A line in a calendar invite noting that "this meeting may be recorded" has long been treated by many businesses as sufficient cover. This year's litigation is testing that assumption directly: a proposed class action filed against Granola this past summer argues that its notetaker captures system audio without ever joining the call as a visible participant, so other people on the line have no real notice a recording is happening at all. Otter.ai is defending several consolidated suits making a similar argument, and Fireflies.ai is fighting separate biometric-privacy claims in Illinois tied to how it processes voice data. None of these cases has fully resolved, but together they signal that courts are willing to look past a boilerplate disclaimer and ask whether a participant actually understood and agreed to being recorded.

For a business, the exposure is not really about the notetaker vendor losing its own lawsuit. It is about what happens once a client call, a job interview, or an internal strategy discussion has been transcribed and stored by a third party the business does not control. Many of these tools retain transcripts by default, and several have faced allegations that they use recorded conversations to train their underlying models unless a user actively opts out — meaning a client's confidential terms, a candidate's medical accommodation request, or an executive's frank assessment of a competitor can end up inside a dataset the business never agreed to contribute to. A transcript is also a permanent, searchable record in a way a person's memory of a call never was, which changes its value as evidence if a dispute or a lawsuit follows later. A privileged legal conversation captured by a notetaker built to summarize and share is a privilege problem waiting to be discovered, often by the business's own lawyer, after the fact.

This is a genuinely different risk profile than most of the AI tools a business has already had to think about, because the exposure does not depend on the notetaker doing anything wrong. It works exactly as designed, and the designed behavior is what creates the liability. A chatbot that gives a wrong answer is a quality failure with an identifiable cause. A notetaker that faithfully records a conversation nobody agreed to have recorded is a consent failure that exists the moment the tool is turned on, regardless of how accurate the transcript turns out to be.

The fix is not banning the tools, which would throw away a real and popular productivity gain for a problem that has a much narrower solution. It starts with the notice itself: a visible bot that announces itself and a spoken or explicit in-meeting acknowledgment holds up far better than a line of calendar text nobody reads, particularly on calls with people outside the business who never saw the invite. It continues with the vendor's settings, not just its marketing — check specifically whether transcripts are used for model training by default, how long recordings are retained, and whether that retention can be shut off at the account level rather than left to each employee's individual preference. And it means drawing an explicit line around which conversations a notetaker should never sit in on: calls involving legal counsel, sensitive personnel matters, or anything where the discussion itself, not just the outcome, needs to stay off the record.

None of this requires a legal review before every call. It requires the same habit a well-run business already applies to any other form of recording a conversation with a client or employee: get real consent, know where the recording goes, and know how long it stays there. The businesses that get caught out by this wave of litigation will not be the ones that decided AI notetakers were too risky to use. They will be the ones that let a convenient default answer a consent question nobody had actually asked.