The AI Notetaker Lawsuits Aren't About Recording. They're About Who Gets Told.

The AI notetaker lawsuits filed against Otter.ai and Granola this year are not arguments about whether recording a meeting is legal. That question was settled long before either company existed. They are arguments about two narrower design choices: whether the tool tells anyone it is there, and what happens to the words after the call ends.

Zoom's built-in recorder has survived legal scrutiny for a decade because it clears a low bar: a visible banner, a chime, and a stop when the meeting ends. An AI meeting notetaker that joins silently or keeps transcripts for other purposes clears a different, higher bar, or it doesn't, and 2026 is the year courts started checking.

What the complaints actually allege

In re Otter.AI Privacy Litigation is consolidated before Judge Eumi K. Lee in the Northern District of California. Otter asked her to dismiss the case in May. On 13 August she declined, letting claims proceed under the federal Wiretap Act, California's Invasion of Privacy Act, and an Illinois biometric-information statute (reported by Recording Law and Law360). That order does not decide whether Otter broke the law. It decides that the plaintiffs' version of events, if true, describes a real violation, and that the case gets to move forward instead of being thrown out.

Seventeen days earlier, on 30 July, a separate proposed class action landed against Granola in the same federal district. Chamberlain v. Granola makes a related but sharper claim: that Granola's notetaker records both virtual and in-person conversations without disclosing its presence to most participants by default, and that the resulting recordings and transcripts get used to train Granola's own models. Coverage of the wider pattern, including suits touching Fireflies, suggests this is now an industry-wide exposure rather than one company's bug.

The consent patchwork these tools ran straight into

The underlying rule predates any of this by decades. The federal Wiretap Act sets a floor: recording an oral or electronic call generally needs the consent of at least one party to it. Roughly a dozen states go further and require every party's consent, among them California, Illinois, Florida, Washington and Pennsylvania. Call a prospect in California from a one-party-consent state and, in practice, the stricter rule governs, because California courts have applied their statute to any call with a non-consenting Californian on the line, regardless of where the recording device physically sits.

None of that is new. What changed is volume and default behaviour. A Zoom call got recorded because a host remembered to press a button, on a call they had already sized up. A notetaker bot gets added because it is set to auto-join every calendar invite, which means far more all-party-consent-state calls are being captured without anyone re-checking the rule for that specific call, that specific week, with that specific external guest.

Otter, Granola, Fireflies: the same AI notetaker lawsuits, different exposure

The three companies named across the current wave of suits share a legal theory but not an implementation, and the differences are exactly what the complaints turn on.

Granola's design is the one drawing the sharpest language from plaintiffs' lawyers, because a tool that never appears in the participant list removes the one signal — "X's notetaker has joined" — that courts have previously treated as a form of notice. Otter and Fireflies at least show up. Whether showing up is enough to count as informed consent under a two-party-consent statute is precisely what the pending cases will decide.

The second front: what happens to the transcript afterwards

Recording consent is one claim. A separate one, present in each of the current suits, is about what the vendor does with the recording once it exists. Sending the host a summary is one use. Feeding the audio and transcript into training data for the vendor's own commercial model is a different use, and plaintiffs argue it needed separate disclosure that a checkbox about "generating notes" never provided.

This is where the Illinois biometric claim comes in. Voice carries measurable characteristics — pitch, cadence, spectral features — that can function as a biometric identifier once a system extracts and stores them, and Illinois's biometric-information law imposes its own notice-and-consent regime on top of ordinary wiretap rules, with statutory damages set per violation rather than per lawsuit. A company that only cleared the wiretap bar can still be exposed here.

Why the terms-of-service defence keeps losing ground

The standard corporate answer to all of this has been: it's in the terms of service, the host clicked accept, the vendor disclosed it. Courts hearing these cases so far have been unwilling to treat that as the end of the analysis, for a simple reason. Wiretap and privacy statutes protect the participants on a call, not just the account holder who signed up for the software. A sales rep's agreement to their employer's AI-notetaker vendor does not put the prospect on the other end of the line, who never saw any terms, on notice of anything.

That gap between who agreed to the tool and who is subject to the tool is the through-line connecting Otter's wiretap claim, Granola's invisibility claim, and the Illinois biometric claim. Each one is really the same question asked three different ways: who actually knew, and what proves it.

What this changes for a team running one of these tools now

• Check whether your notetaker visibly announces itself in the call, or runs silently in the background — that single fact is central to at least one live lawsuit.
• Find out whether transcript-based model training is opt-in or opt-out by default in your contract, not just in the public terms of service.
• Map which recurring external calls involve participants in two-party-consent states; a blanket auto-join setting doesn't distinguish them.
• Treat a host's acceptance of a vendor's terms as covering the host, not the other participants, until your own disclosure tells guests what's recording them and why.

None of this requires abandoning AI notetakers. It requires treating the two design choices this litigation is testing — visibility and downstream use — as facts to verify per vendor, rather than assumptions to inherit from whichever tool a team member installed first.

Where this heads next

Whatever the Otter and Granola cases resolve to, the design pattern being tested here is already being copied by newer entrants chasing the same feature set: join quietly, summarise well, improve the model on whatever came through. The vendors likely to come out ahead of the next wave of suits are the ones that make both choices visible on their own, before a judge makes them do it.