AI Notetakers in Court: What the Granola and Otter Lawsuits Mean for Your Meetings

Two of the best-known AI meeting notetakers are now defendants in federal privacy class actions. The cases — against Otter.ai and Granola — are not about a data breach. Nobody hacked anything. They are about something more basic: who agreed to be recorded, and what the vendor did with people’s words afterwards.

If your team uses an AI notetaker — or sits in meetings where someone else does — these cases are worth understanding. Below is what has actually been filed and decided so far, what is still only alleged, and the questions every buyer should ask before rolling a notetaker out to a whole company.

A note on fairness: a lawsuit is a set of allegations, not a verdict. Nothing described below has been proven in court unless we say a court has ruled on it. We will be precise about which is which.

The Otter case: a judge let the core claims proceed

In re Otter.AI Privacy Litigation, No. 25-cv-06911-EKL (U.S. District Court, Northern District of California), consolidates four class actions first filed in 2025, beginning with Brewer v. Otter.ai. The plaintiffs — people who were recorded by Otter’s notetaker — allege that Otter recorded and transcribed their conversations without the consent of everyone on the call, and that it used conversation data for its own purposes, including improving its machine-learning models.

On August 13, 2026, Judge Eumi K. Lee ruled on Otter’s motion to dismiss. The court granted the motion in part and denied it in part, and the core claims survived:

• Surviving claims include the federal Wiretap Act (ECPA), the California Invasion of Privacy Act (CIPA §§ 631 and 632), the Illinois Biometric Information Privacy Act (BIPA) — based on allegations about voiceprints and speaker-identification profiles — plus unjust enrichment, declaratory judgment, and California’s Unfair Competition Law.
• Dismissed claims, most with leave to amend, include the federal and California computer-access statutes (CFAA/CDAFA) and the Washington Privacy Act, along with most common-law privacy claims.

The reasoning that matters most: the court held that the plaintiffs had plausibly alleged Otter is a “third-party eavesdropper” under CIPA § 631 — rather than merely an invited participant or a passive tool of the meeting host — because Otter allegedly “independently collects, retains, and uses communications for its own commercial purposes.” In other words, the allegation that the vendor uses recordings for itself — including to improve its own models — is what converted a recording tool into a legal eavesdropper in the court’s analysis.

Two limits to keep in mind. First, this was a pleading-stage ruling: the judge found the claims plausible enough to proceed to discovery, not that Otter broke the law. No court has found Otter liable, and no class has been certified. Second, the case is ongoing — but it has already survived the stage at which weak cases get thrown out, and the statutory exposure pleaded is significant (CIPA allows $5,000 per violation; the ECPA allows the greater of $10,000 per violation or $100 per day).

The Granola case: sued for having no bot at all

Chamberlain v. Granola, Inc., No. 3:26-cv-07926 (N.D. Cal.), was filed on July 30, 2026, before Judge Edward M. Chen; an amended complaint followed on October 19, 2026. The plaintiff is a meeting participant — not a Granola user — who says she never knew the tool was running.

Granola’s design is the opposite of Otter’s: no bot joins the meeting, nothing appears in the participant list. It captures system audio on the user’s own device. The complaint alleges that this invisibility is the problem, pointing to Granola’s own marketing — the product has been promoted with lines to the effect of “other people in the room won’t know it’s there” — and alleging that in-meeting notification and watermarking features ship off by default. The complaint brings claims under the federal Wiretap Act and CIPA §§ 631 and 632, and also alleges a second, separate issue:

Default AI training. The complaint alleges Granola uses captured communications for AI model training by default. Granola’s own published security documentation confirms the shape of this: Granola trains on anonymized data, individual users can opt out, org-wide opt-out is an Enterprise feature, and training is off by default only on Enterprise plans — meaning on the standard plans, it is on unless someone turns it off. Granola also acknowledges that once data has been incorporated into a model, it cannot be extracted — and that it cannot guarantee data was not used before an opt-out took effect.

This case is at an early stage. As of early November 2026, Granola had not responded on the merits, the plaintiffs face real legal hurdles (federal wiretap law’s one-party consent framework chief among them), and none of the allegations have been proven. A parallel proposed class action has also been filed against Fireflies.ai on similar theories.

The pattern: two defaults are doing the damage

Step back from the dockets and the two cases allege the same two product decisions:

1. Capture that other people can’t see. A bot that joins without clear agreement, or no bot and no signal at all. Either way, the people with the most at stake — the ones being recorded who never chose the tool — are the last to know.
2. Training on by default. The moment a vendor reserves the right to learn from your conversations, it stops being just your tool in the eyes of plaintiffs’ lawyers — and, in the Otter ruling, in the eyes of a federal judge at the pleading stage. The “eavesdropper” theory only works because the vendor allegedly keeps and uses the data for its own benefit.

Neither case is really about transcription quality. They are about defaults — what happens to the people on the other side of the call, and what happens to the data after the meeting ends.

Five questions to ask any AI notetaker vendor (including us)

1. Does the tool train on my meetings? Get the answer in the privacy policy, not the marketing page. Check whether “no training” is contractual or just a blog post.
2. If training exists, what is the default? “You can opt out” and “we never do it” are very different answers. Is opt-out per user or org-wide? Does it apply retroactively?
3. How do other participants know they’re being recorded? A visible bot, an announcement, a watermark — and are those notifications on by default, or buried in settings?
4. Who are the subprocessors? Can the vendor name every company that touches your audio and transcripts?
5. What happens when I leave? Can you export everything, delete everything, and get a straight answer on how long deletion takes — including backups?

Where Laxis stands

We will keep this short, because this post is about the public record, not a sales pitch. Laxis commits — in our Privacy Policy and our Terms of Use — that we do not use your recordings, transcripts, notes, or dictation to train AI models without your consent. The details, and our reasoning, are in our companion post: Your Meetings Are Not Training Data: The Laxis No-Training Pledge. On the other question these cases raise, Laxis is bot optional: record on your own device, or turn on a visible AI Assistant Bot that joins the call. And because consent duties sit with the person recording, we have also published a plain-English guide: Is It Illegal to Record a Conversation?

The courts will decide the Otter and Granola cases. But the questions those cases are asking — did everyone agree, and whose data is it afterwards? — are questions you should ask before your next meeting, not after your first subpoena.

Sources and further reading

• In re Otter.AI Privacy Litigation, No. 25-cv-06911-EKL (N.D. Cal.), Order on Motion to Dismiss, Aug. 13, 2026 (Dkt. 68); docket on CourtListener.
• Chamberlain v. Granola, Inc., No. 3:26-cv-07926 (N.D. Cal., filed July 30, 2026; amended complaint Oct. 19, 2026).
• Barnes & Thornburg LLP, “AI Note-Taking Tools, Silent Recording, and the Consent Problem” (2026).
• Granola, model training documentation (training defaults and opt-outs by plan).

This post summarizes public court filings and vendor documentation for general information. It is not legal advice, and allegations in pending litigation are not findings of fact.