The transcript that kept running after the meeting
In September 2024 machine learning engineer Alex Bilzerian joined a Zoom call with a venture capital firm. After the call he automatically received the transcript by email, as Otter.ai users do. Except this transcript contained several additional hours of conversation recorded after the call had officially ended, including the investors' internal discussion. Bilzerian posted about it publicly, and that incident grew into one of the most consequential legal fights in the AI industry.
Otter.ai is now defending a consolidated class action in California that turns on a single question: may an AI notetaker listen in when only the host agreed? The case does not decide whether meeting bots are allowed. It decides how a bot is supposed to get into a meeting.
One incident became four lawsuits
On 15 August 2025 Justin Brewer filed the first class action against Otter.ai in the U.S. District Court for the Northern District of California, and three more suits followed by September. On 22 October 2025 Judge Eumi K. Lee consolidated them: In re Otter.AI Privacy Litigation, No. 5:25-cv-06911-EKL. The consolidated complaint was filed on 5 December 2025.
The composition of the plaintiff group is the telling part. Brewer, Jasper Walker, Michael Walker, Chaka Theus and Nadine Winston say they were not Otter customers at all. They sat in meetings that somebody else had brought the Notetaker into. The people suing are not the paying users, they are the people on the other end of the call.
What the plaintiffs are claiming
The central claims are under the federal Wiretap Act as part of the Electronic Communications Privacy Act (18 U.S.C. sections 2510 to 2522) and the California Invasion of Privacy Act with Penal Code sections 631 and 632, which require the consent of every party to a conversation. On top of that come claims under the Computer Fraud and Abuse Act, the Unfair Competition Law and common law intrusion upon seclusion. The second major allegation: Otter allegedly failed to disclose that the recordings fed the training of its automatic speech recognition and machine learning models.
The numbers explain why the whole industry is watching. The California Invasion of Privacy Act provides for 5,000 US dollars per violation or three times actual damages, and the federal Wiretap Act for the greater of 10,000 US dollars per violation or 100 dollars per day.
Otter's defence and where the case stands
Otter.ai filed a motion to dismiss on 20 January 2026. Its core argument: the Notetaker is not an independent eavesdropping third party but an extension of the host who is running the meeting anyway, so nothing is intercepted. CEO Sam Liang told TechCrunch in October 2025 that the company is "on the right side of history", summarising the principle as: if you want AI to help, you need to put AI in the meetings.
The meeting bot is not the problem
The quick takeaway in a lot of commentary is that meeting bots are a legal liability, so get rid of them. That is the wrong takeaway. Otter is not accused of having a bot, it is accused of having a bot in meetings where not everybody had allowed it. That is a consent problem, not a technology problem.
A visible bot is more honest than an invisible recording
A bot that shows up as a participant has an underrated advantage: everybody in the room can see that the conversation is being captured. The name sits in the participant list, and it is often clear who invited it. Tools that tap system audio in the background appear in no participant list at all, so whoever sits at the other end has no way of knowing the conversation is being recorded. Otter's bot, incidentally, was visible, which is why the lawsuit is not about visibility.
The allegation targets the defaults, not the technology
What the complaint describes is a chain of default settings that worked against the other participants. The Notetaker joined calendar meetings automatically as soon as the host had an Otter account, and it did not ask the other attendees for permission. People without an Otter account, according to the complaint, got little or no notice that the recording was also used to improve the product. And in the incident that started it all, the recording kept running after the meeting was over.
Every one of those points is a product decision, not a technical necessity. Neither California nor Germany bans recording a conversation, both ban recording it without the knowledge and consent of the people involved. So the relevant question is never "bot or no bot", it is: did everybody know, and did everybody agree?
Why bot-free tools only move the problem
Since the lawsuits started, several vendors market "no bot in your meeting" as a privacy feature. That is marketing, not law. The recording happens either way, only the visibility disappears. The reverse is also true, though: a visible bot does not replace the announcement. A participant named "Notetaker" in the list does not prove that everyone noticed it and accepted it.
What the case means outside California
The lawsuit runs under California law, but the principle is not a California quirk. Roughly a dozen US states require all-party consent, and in Europe the bar sits higher still, with criminal law and data protection law stacked on top of each other.
Germany treats secret recording as a criminal offence
Under section 201 of the German Criminal Code, recording, using or passing on someone's non-public spoken word without their consent carries up to three years imprisonment or a fine, and it targets the individual who did it, not just a company. A dictaphone would be covered exactly like a bot. Consent is the decisive element: if participants are clearly informed beforehand and then continue to take part, that is generally read as implied consent. Internally a clear notice at the start is usually enough, with external participants you want explicit agreement. Our guide on recording meetings in Germany has the wording and the practical steps.
The GDPR asks for more than a bot in the participant list
Voices are personal data and a transcript is processing, which brings in Art. 5(1)(a) on transparency, Art. 13 on information duties, Art. 6 on a legal basis and Art. 28 on data processing agreements. The point Otter, by its own account, leaves to the host is precisely the point supervisory authorities tend to look for at the controller: a contract that pushes the entire consent burden onto the user does not automatically clear the vendor. Our article on Otter.ai and GDPR goes through how the product is set up.
Training data and hosting location
The training allegation has the longer tail. Once recordings flow into model training, the content leaves the purpose it was collected for, which needs its own legal basis and regularly collides with confidentiality duties from contracts or professional rules. For lawyers, tax advisors and doctors it is a straight disqualifier.
Then there is the jurisdiction you buy along with the vendor. Otter states that it processes primarily on US servers, and after Schrems II the United States is not treated as a safe third country, while the CLOUD Act opens access routes that European contracts cannot switch off. We covered why the server location of a meeting tool is more than a line in a spec sheet.
How to recognise a clean setup
The table separates what is being litigated in California from what nobody argues about. It works as a review grid for any tool, not just Otter.
| Criterion | Risky | Unproblematic |
|---|---|---|
| Joining the meeting | Bot joins every calendar entry automatically | Bot only joins when explicitly invited |
| Visibility | Background capture, no entry in the participant list | Named participant, visible to everyone |
| Consent | Only the host agreed | Announced at the start, external parties agree explicitly |
| End of recording | Capture continues after the meeting | Recording ends with the meeting |
| AI training | Recordings improve the product by default | Training on customer data contractually excluded |
| Hosting location | United States, CLOUD Act access possible | Germany or EU, data processing agreement in place |
| Transcript distribution | Automatic email to every address on the invite | Access governed by roles and permissions |
Six rules for working with AI notetakers
None of these needs a legal project. Together they cost about two minutes per meeting and take you out of the risk zone currently being argued in California.
- Announce it before the bot starts: "I am letting our AI assistant take notes, and if anyone would rather not, I will switch it off." Ten seconds that turn a secret recording into an open one.
- Get active agreement from externals: with customers, candidates and works councils, silence is not a defensible basis, because you cannot show the notice landed.
- Keep the bot visible: visibility is not a weakness, it is the evidence that nobody was misled.
- Verify when recording stops: the Otter incident happened after the meeting. Anything that appears after the goodbyes is a failure you were warned about.
- Exclude AI training in writing: ask about telemetry, feedback loops and review pipelines, not only foundation model training, and get the answer into the data processing agreement. Our GDPR checklist for AI meeting tools lists the questions where vendor answers get vague.
- Leave sensitive conversations out: terminations, health matters, investigations. A tool does not have to run in every meeting to pay for itself.
How Sally avoids the conflict
Sally is an AI meeting assistant built by Aliru GmbH in Mannheim, Germany, and it works on an invitation principle. The bot appears in Google Meet, Zoom, Microsoft Teams and Webex, but only when you bring it in, for example by inviting invite@sally.io to the meeting, and it sits visibly in the participant list. From the transcript Sally produces a summary, decisions and tasks, across 99+ languages and with connections to 8,000+ tools. On the privacy side the differences sit exactly where the California fight is: hosting is exclusively in Germany, training on customer data is contractually excluded, and a data processing agreement is standard. The details are on the pages about GDPR and security and AI meeting transcription.
And because honesty matters here: Sally does not make the announcement for you. No vendor can collect your counterpart's consent on your behalf. German hosting protects you from the third country risk, not from criminal recording law. For in-person meetings the capture runs through participants' phones lying visibly on the table, which according to Sally reaches up to 98.8 % accuracy with several devices.
Conclusion: invite the bot, do not sneak it in
The Otter class action is not an argument against AI notetakers, it is an argument against notetakers that run without the knowledge and consent of everyone involved. The gap between a clean setup and a risky one comes down to three decisions: the bot joins only on invitation, the recording is announced, and the recordings never become training material. In fairness, Otter's recognition is strong and the product is mature, the dispute is about defaults and jurisdiction. If that is your reason to move, our comparison of Otter.ai alternatives lists the candidates with European hosting.
Get those three points right and you keep the upside without the headlines: minutes that write themselves, tasks that do not get forgotten, and knowledge that stays inside the company. You can try Sally free for 30 days, with plans starting at 8 euros per month.
