AI Notetaker Suits Leave Meeting Hosts Holding the Risk

A San Jose federal judge ruled on August 13, 2026 that an AI notetaker can count as a third-party eavesdropper. Judge Eumi K. Lee let federal wiretap, California privacy, and Illinois biometric claims against Otter.ai move into discovery.

The captions name software vendors. The people with something to lose are often the ones who never opened an account: the host who flipped on calendar auto-join, the employer that blessed the tool, and the guest in California or Illinois whose voice went to a vendor server.

A Federal Judge Called the Bot an Eavesdropper

Otter, which claims 35 million users, is the furthest along. Four proposed class actions filed in August and September 2025 were consolidated on October 22, 2025 as In re Otter.AI Privacy Litigation, case 5:25-cv-06911-EKL, in the Northern District of California. The consolidated complaint followed on December 5, 2025. Lee heard the dismissal motion on May 20, 2026.

Seven named plaintiffs from California, Illinois, and Washington say Otter’s Notetaker joined Zoom, Microsoft Teams, and Google Meet calls as a silent participant, then recorded and kept what was said. The Illinois plaintiffs say the tool also captured voiceprints. The medical, financial, and work talk they describe is the kind of speech people still treat as private even on a video call.

Otter argued the users who turned the bot on had consented, and that the guests had not shown a concrete injury. Lee rejected the standing argument at the pleading stage, finding the alleged interception close enough to the old tort of intrusion upon seclusion. On the merits, the computer-hacking counts fell. The wiretap and privacy counts did not, in an order letting the wiretap claims proceed.

Because Plaintiffs plausibly allege that Otter independently collects, retains, and uses communications for its own commercial purposes, they have sufficiently alleged that Otter is a third-party eavesdropper under section 631.

Eumi K. Lee, U.S. District Judge, August 13, 2026 order

That sentence is the hinge. If the bot is only a tool of the host, one person’s consent can cover a federal wiretap claim. If the vendor keeps the audio for its own models, the vendor looks like a stranger on the line. Lee also found the Illinois plaintiffs had plausibly alleged a voiceprint collection under BIPA section 15(b). Otter got 14 days to face an amended complaint. The surviving claims go to discovery.

Granola Marketed a Notetaker Nobody Else Could See

OtterPilot at least shows up in the participant list. Granola built the opposite product. Tarra Chamberlain, a Florida resident, sued Granola, Inc. and Granola Labs Ltd. on July 30, 2026 in the same district, case 3:26-cv-07926-EMC, before Judge Edward M. Chen. The 38-page complaint that calls the product spyware says Granola intercepts every voice on a Teams or Zoom call without telling the other people it is there.

This case concerns spyware. Granola designed and created an AI notetaking product that surreptitiously intercepts, records, and interprets the communications of every participant in virtual meetings without their knowledge or consent.

Tarra Chamberlain, class-action complaint, July 30, 2026

Counsel pointed at Granola’s own site, which sold the idea that other people in the room would not know the tool was running, and called that gap the distinction that matters most against bot-based rivals. The off switch for model training, the complaint says, belongs to the Granola user. Guests never see a setting to refuse. Granola’s privacy policy, quoted in the filing, says data cannot be pulled back out of a trained model without retraining the whole system.

Brian McGinnis, a partner at Barnes & Thornburg who co-chairs the firm’s data-security practice, has been telling clients the same design choice is what makes the newer cases sharper than a bot that at least has a name on the screen. Notice features can be turned on, he said, including audio or video watermarking, but his understanding is that the product can ship with those off, which leaves the individual user to invent a compliance process the software did not.

The filing names a nationwide class and a California subclass, claims more than $5 million in controversy, and lists more than 100 putative members. Granola customers named in the complaint include Brex, Vercel, and Vanta. That has not been tested in court. Chen’s case is still at the response stage, and the parties have already stipulated more time.

The Class Is Everyone Who Never Clicked Accept

Each of the four live cases is built around people who were simply on the call. That is the stakeholder the product pages do not sell to, and it is the stakeholder the damages statutes actually pay.

THE FOUR CASES STILL OPEN

Fireflies, which claims more than 20 million individuals and 1 million organizations as customers, is accused of building voiceprints through speaker recognition when its bot joined a nonprofit meeting that Illinois resident Katelin Cruz attended. She was not a Fireflies user. BIPA still wants a written release from each person whose voiceprint is taken, plus a public retention and destruction policy.

Microsoft is in a different posture. Basich v. Microsoft Corporation, filed February 5, 2026 before Judge John H. Chun in Seattle, targets live transcription inside Teams rather than a third-party bot. The proposed class is people who were in or living in Illinois from March 1, 2021 onward when transcription ran, except those who had enrolled an Intelliframe voice profile. Microsoft says routine transcription is not a voiceprint. The feature is not on by default; someone has to enable it. A motion to dismiss was filed May 22, 2026 and remains pending.

What a Voiceprint Costs Under Illinois Law

Federal law is the easy baseline. The Electronic Communications Privacy Act generally allows a recording when one party consents. A dozen states, including California, Florida, Illinois, Pennsylvania, and Massachusetts, require every participant to agree. California’s Invasion of Privacy Act is the statute Lee kept alive, and it is the one Granola faces as well.

WHAT THE STATUTES PAY

• California CIPA: Injured people can seek $5,000 per violation without proving harm, or three times actual damages, whichever is greater.
• Illinois BIPA: Voiceprints sit in the statute’s list of biometric identifiers beside fingerprints and face geometry. Damages are $1,000 per negligent violation and $5,000 per intentional or reckless one, plus fees.
• No extra injury required: Illinois courts have treated the statutory violation itself as enough to sue. A 2024 amendment, applied retroactively by the Seventh Circuit on April 1, 2026 in Clay v. Union Pacific, limits repeat collections of the same identifier from the same person to one recovery.
• All-party states: Headquarters is irrelevant. A single guest sitting in California or Illinois can pull the stricter rule onto a call hosted somewhere else.

McGinnis put the biometric piece in plain terms: a dumb recording is still a recording, but once the software starts identifying people, you are in voiceprint territory, and that is treated as sensitive. Plenty of companies already avoid Illinois because of BIPA. A customer that cannot see who is on a call, or where they are sitting, does not have that luxury.

Calendar Integrations Put Employers in the Chain

None of the four complaints names an employer as a co-defendant. That fact does not clear the host. The account that connected OtterPilot to a work calendar, or that let Fireflies auto-join, is almost always a company account. Guests who later become class members were on that company’s call.

McGinnis said clients keep asking because they want the productivity and they keep hearing the other story too: someone joins a meeting, never sees a recorder, then gets an email with a transcript that misheard half the lines. He will not tell them the tools are legal as they ship. He will tell them there is a way to use them that is much less likely to produce a suit, and it is more than the default settings.

WHAT A HOST HAS TO ADD

• All-party consent: Get it from everyone on the call, in writing if you can, and verbally at the start if you cannot. One host clicking yes is the theory now under attack.
• A named tool list: Treat notetakers like a BYOD rule. Approve specific products, require the notice and watermark features to be on, and ban the rest.
• Limits on the output: Say what staff may do with a transcript, especially in hiring, firing, medical, or legal meetings, and how long the file may live.
• Guest geography: Assume a California or Illinois participant is on the line unless you have checked. Play to the stricter statute.

A visible bot name in the participant list is not consent under California law, which is why the “just look at the roster” defense is a thin one. Candidate interviews are a particularly bad place to improvise, because a voiceprint claim does not care whether the person was hired. The further a company moves from written opt-in toward silent capture, McGinnis said, the easier the challenge becomes.

New York City Lawyers Are Told to Keep Recorders Off

One profession has already been told, in writing, that it is the hidden party. The New York City Bar Association issued Formal Opinion 2026-2 in August 2026 for calls with people who are not clients: opposing counsel, witnesses, prospective clients, co-counsel, investigators. Consent of every party is required before a recording starts. The default practice should be not to record unless there is a good reason in that instance, because the ethical and tactical risks usually outweigh the convenience.

As is true with clients, an attorney should obtain consent of all other parties to a call before recording it, and should consider whether recording, transcribing and summarizing is tactically well-advised in the particular circumstances.

New York City Bar Association, Formal Opinion 2026-2

Privilege is the extra problem lawyers carry that a sales standup does not. A transcript of a witness call can be discoverable. A mediation discussion can lose its protection once it is sitting on a vendor’s server. Some notetakers, the opinion notes, lock a participant out if they refuse to be recorded, which makes free consent harder to claim. An earlier 2025 opinion already told lawyers to warn clients who want to run their own AI recorder on a privileged call.

That advice travels. HR, in-house counsel, and anyone on a regulated line is in the same seat even if no bar opinion has their name on it. The vendor will point at the user. The user works for a company. The guest who never clicked a box is the one the class papers describe.

Zoom and Teams Now Make the Bot Wait Outside

Meeting platforms have started treating third-party notetakers as uninvited people rather than as features. That is a practical answer to the consent hole, and it is also an admission that a bot posing as a participant was always a bad object.

WHEN THE PLATFORMS CHANGED THE DOOR POLICY

1. January 21, 2025: The University of Washington’s IT office blocks Read AI from Zoom and Teams integrations.
2. August 13, 2025: Chapman University’s IS&T department prohibits Read AI after finding it could auto-join from a calendar even when the user was absent.
3. October 7, 2025: UC Riverside restricts non-native AI bots across its video tools. Harvard’s IT group separately tells staff to skip AI meeting assistants unless they sit under an enterprise deal with real privacy terms. UCSB’s rule is simpler: if the meeting might hold sensitive data, do not allow third-party bots.
4. March 13, 2026: Microsoft Message Center notice MC1251206 says Teams will detect external meeting bots, label them Unverified, and make the organizer admit them. Admins can auto-block them at the tenant level.

Zoom’s own AI Companion already needs a host or co-host to start it, shows a sparkle when it is on, and asks each participant. Native tools can put a notice on the screen because they own the screen. A third-party bot that joins as a fake attendee was always borrowing someone else’s consent UI. Shopify chief executive Tobi Lutke drew that line in public last year, while still backing recording that people actually agree to.

I’m actually very pro meeting recording and ai summarization. But I’m not ok with bots joining as fake humans accomplish this. It’s a meeting between you and me. Not you and me and some startup’s viral growth strategy.

Granola is great. Gemini does this well in Google Meet.… https://t.co/0Z1JiZp02t pic.twitter.com/nec8Xm81nh

— tobi lutke (@tobi) October 30, 2025

A visible bot is rude, and it is also the one signal that lets someone object. Granola’s pitch was to remove that signal. McGinnis has already connected the next step: Apple Watch apps, clip-on recorders such as Plaud Note, and glasses that listen on a sidewalk, where there is no lobby and no pop-up. The meeting cases are the ones with a docket number. The physical versions will not have a participant list to kick.

Judge Chun still has Microsoft’s dismissal motion. Granola has not yet answered on the merits. Otter is the case that is now in discovery, on the theory that a productivity bot which keeps the tape for itself is an eavesdropper. The host who invited it remains the person who has to live with that tape.

Disclaimer: This article is news reporting and analysis of pending civil cases and published ethics guidance. It is for information only and is not legal advice, is not a prediction of how any court will rule, and is not a recommendation to use or ban any product. Readers who record meetings, buy these tools, or write workplace policy should consult a qualified attorney licensed in the relevant states before changing how they record or transcribe calls. Case status, product settings, and damages figures reflect the court papers, statutes, and statements cited here as of the dates on those documents and can change as the cases move.