AI Notetakers: What the Otter Case Actually Decided

The story circulating is that a court ruled against an AI notetaker on wiretap grounds. What happened is narrower and more useful. In August, a federal judge in California declined to throw out wiretap and biometric privacy claims against Otter.ai, which means those claims proceed to be argued rather than that anything has been decided about them.

What the August order did

Four class actions filed in 2025 were consolidated as In re Otter.ai Privacy Litigation before Judge Eumi K. Lee in the Northern District of California. On August 13, 2026, the court granted Otter's motion to dismiss in part.

Read that bottom row carefully before repeating the headline version. A motion to dismiss asks whether a claim could succeed if everything alleged turned out to be true, so the plaintiffs' account is assumed rather than tested. Otter may well prevail on the facts. What the order establishes is that the legal theory is not obviously wrong, which is a lower bar and still a meaningful signal.

The argument the court did not adopt

Otter argued that its Notetaker joins meetings as an invited participant and operates as the host's recording tool, effectively an extension of the customer who deployed it. The court did not accept that framing at this stage, and the reasoning carries further than this one case.

Because plaintiffs allege the service retains the conversations and the voiceprints and uses them to improve its own models, the court found they had sufficiently pleaded that the vendor acted as a third party listening in rather than as a tool held by a participant. Appearing in the attendee list was not treated as settling consent.

Plenty of organisations currently rely on exactly that assumption. The bot shows up in the participant list, nobody objects, and the recording is treated as consented to. That position is weaker than it looks, and it costs nothing to replace with an actual announcement.

Going bot-free does not settle it either

A second proposed class action, Chamberlain v. Granola, was filed in the same court in July 2026 and concerns a notetaker that captures audio directly from one participant's machine without appearing in the meeting at all. The complaint alleges participants had no notice an AI notetaker was present, and that meeting content was used by default for commercial purposes including model training unless a user changed a setting.

Nothing there has been decided and Granola may prevail. The reason it is worth knowing about is that the two cases close off opposite escapes. A visible bot was argued to supply consent and that argument did not succeed at dismissal; an invisible tool avoids the bot problem and raises the notice problem instead.

The Granola complaint also points at a question aimed squarely at customers rather than vendors. Where a tool offers consent features that the customer has to switch on, leaving them off is a decision the customer made, and the commentary around these cases is asking what that means for the deploying organisation.

Where this leaves a business in Canada

These are US cases under US statutes, and Canadian law approaches recording differently, so none of this applies directly. Two things make it relevant anyway.

Your calls include people elsewhere. Roughly a dozen US states require consent from every participant rather than just one, California among them, and the applicable standard tends to follow where the participants are rather than where you are. A sales call with a client in California is the ordinary case, not the exotic one.

And recording a meeting collects personal information, which brings your own privacy obligations into play regardless of what any US court decides. The sensible posture is to meet the strictest standard likely to apply to the people on your calls, which in practice means asking everyone. Treat this as general information rather than legal advice, and the broader policy frame is in whether your business needs an AI policy.

What to change this month

Announce it out loud. A sentence at the start of the call saying the meeting is being recorded and transcribed by an AI assistant, with a pause for anyone to object. That is more than a participant-list entry provides and it takes four seconds.

Find out what is actually running. Most organisations have more notetakers in use than they have approved, added by individual employees on personal accounts under terms nobody at the company has read. That is the same free-plan gap described in what to check in AI terms of service.

Check the training setting, in writing. Whether your conversations are used to improve the vendor's models is a clause, not a preference, and it should be off by default on a business plan rather than off because somebody toggled it.

Consider consolidating onto the platform you already contract with. The notetaker built into your conferencing tool sits under terms your organisation reviewed once, which is a smaller unexamined surface than several separate vendors added individually. The productivity case for these tools is real and unchanged, as covered in AI meeting notes and the end of post-call admin.

Set a retention period. Transcripts of every meeting your company has ever held are discoverable, and keeping them indefinitely is a choice rather than a default, which is the same argument as in always-on AI and devices that record everything.

Frequently Asked Questions