In re Otter.ai: an AI notetaker that keeps conversations for its own models is pleaded as an eavesdropper

Read your AI vendor’s data-use terms before you route conversations through it. A tool that only records for you is treated like your tape recorder. A vendor alleged to keep the content and train its own models is treated, at least at the pleading stage, as a third party listening in. That distinction kept the CIPA claim alive against Otter.ai.

US District Court, Northern District of California · CIPA §631, Wiretap Act, BIPA · Order on motion to dismiss, 13 August 2026

The problem

Otter’s "Otter Notetaker" joins Zoom, Microsoft Teams and Google Meet calls as a "silent participant" and records and transcribes them in real time. The plaintiffs say it did so without the consent of every participant, in calls that included medical, financial and work discussions. They allege Otter kept the conversations, voiceprints and other data "for its own commercial use, including to train its automatic speech recognition and machine-learning models" (Order at 1).

Case record

Court: US District Court, Northern District of California, San Jose (Judge Eumi K. Lee)
Case number: 5:25-cv-06911-EKL; related cases include Winston v. Otter.ai, No. 25-cv-07712 ([referral order](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-0.pdf))
Parties: Seven participants from California, Illinois and Washington v. Otter.ai Inc. ([order at 2](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))
Counsel: Not stated in the cited record
Laws: CIPA §§631 and 632, federal Wiretap Act, Illinois BIPA, Washington Privacy Act, CFAA, CDAFA, UCL, privacy torts and others
Stage: Motion to dismiss granted in part, denied in part; leave to amend
Precedent: District court order; not binding on other courts
Amount: None
Source review: 17 October 2026

The problem
Case record
What was alleged
Technical issue
Timeline
Outcome
What it cost
Why it matters
How Inth catches this
Checks for your own systems
Sources and limits

What was alleged

What was alleged
Claim	What the law requires	Facts alleged	Result
CIPA §631(a)	A third party reading or learning contents in transit without all-party consent	Otter records and transcribes in real time and retains content to train its models	Survives for all three California plaintiffs ([at 16–17](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))
CIPA §632	Recording a confidential communication without all-party consent	Calls with a medical professional; "sensitive" work calls	Plausible only for the plaintiff who spoke with a medical professional ([at 17](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))
Federal Wiretap Act	Contemporaneous interception; one-party consent unless for a criminal or tortious purpose	Real-time capture; use of data to train models "for its own pecuniary gain"	Survives ([at 10–11](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))
Illinois BIPA §15(a), (b)	Written notice, release and retention policy before collecting biometric identifiers	Otter captures and stores voiceprints to identify speakers in later meetings	Survives ([at 6–9](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))
Intrusion and California constitutional privacy	Reasonable expectation of privacy; highly offensive intrusion	Recording of medical and work calls	Survives for the medical-call plaintiff only ([at 13–15](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))
CFAA; CDAFA; Washington Privacy Act	Access to a computer with loss; private communication	Bot joins meetings	Dismissed with leave to amend ([at 11–13, 17–18, 21–22](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))
UCL; unjust enrichment; declaratory relief	Economic injury; benefit unjustly kept	Market value of conversational data	Survive ([at 18–21](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))

Plaintiffs withdrew their conversion, trespass, larceny, CIPA §635 and CIPA §638.51 counts (at 22). From 1 January 2027, SB 690 (Stats. 2026, ch. 976) leaves §638.51 claims over website and app conduct to the Attorney General alone (bill text).

Technical issue

Tools or vendors: the Otter Notetaker, used with Zoom, Microsoft Teams and Google Meet.
Where: live virtual meetings, not a website. Otter is the vendor and the defendant.
What data: audio, real-time transcripts and voiceprints. Depending on the version, Otter tags speakers automatically from Zoom participant names or from data its customers enter, and builds a "speaker identification profile" to recognise people in later calls (at 8).
When: during the call, without consent from every participant.
Secondary use: alleged retention and use of conversations to improve Otter’s machine-learning models (at 1, 16).
How it came to light: the complaints. No evidence has been tested.

Inth’s technical interpretation: a meeting bot is a participant client: it joins through the platform, receives the mixed audio stream, sends it to the vendor’s speech pipeline and gets text back within seconds. The legal question is not where the audio goes, because it must go to the vendor for transcription. It is what the vendor keeps and uses afterwards. The same pattern sits inside many web products today: a support chat or AI assistant that forwards each message to a model provider. If that provider retains content or trains on it by default, the Otter theory reaches you.

Illustrative pattern, not the defendant’s code. Conversation text forwarded to an AI vendor on default terms:

// support-assistant.ts (generic AI vendor SDK)
import { aiVendor } from './ai-vendor';

export async function answer(message: string, userId: string) {
  return aiVendor.complete({
    input: message,
    user: userId,
    // vendor default: retain inputs and use them to improve models
  });
}

Illustrative pattern, not the defendant’s code. Notice in the interface and retention switched off in the request:

import { aiVendor } from './ai-vendor';

// UI above the input: "Messages are processed by our AI provider to answer you.
// They are not used to train its models."
export async function answer(message: string) {
  return aiVendor.complete({
    input: message,
    store: false,          // vendor option: do not retain this request
    trainingOptOut: true,  // contractual setting confirmed with the vendor
  });
}

Timeline

Timeline
Date	Event
2025	Brewer v. Otter.ai filed as No. 5:25-cv-06911; related suits follow
11 September 2025	Winston v. Otter.ai referred to Judge Lee as possibly related ([order](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-0.pdf))
After consolidation	Consolidated complaint (Dkt. 35) and Otter's motion to dismiss (Dkt. 42) ([order at 1](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))
15 May 2026	Deadline to seek leave to amend pleadings ([order at 22](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))
13 August 2026	Motion to dismiss granted in part, denied in part ([order](https://www.govinfo.gov/content/pkg/USCOURTS-cand-5_25-cv-06911/pdf/USCOURTS-cand-5_25-cv-06911-1.pdf))

Outcome

On §631, Otter argued it was the meeting host’s recording tool. The court accepted that a vendor acting only as an extension of its customer is not a third-party interceptor, but held: "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" (at 16).

On standing, the court distinguished Popa v. Microsoft: the alleged intrusion was "materially more invasive than the monitoring of browsing activity in Popa", because it involved whole conversations on sensitive topics (at 5). On the Wiretap Act, Otter’s commercial motive did not defeat the crime-tort exception at this stage, because "committing a tort and seeking a profit are not mutually exclusive" (at 11, quoting Riganian v. LiveRamp).

Specific facts mattered. Plaintiffs who called their meetings only "private" or "sensitive" lost their intrusion and §632 claims; the one who described a call with a medical professional kept them (at 13–14, 17). This is a pleading decision: the allegations are not findings.

What it cost

What it cost
Item	Record
Damages or settlement	None
Fee award	None
Defence costs	Not stated in the cited record

Why it matters

It marks the line for vendors. Gutierrez v. Converse and Torres v. Prudential were lost for lack of evidence that a vendor read anything for itself. Otter is alleged to have kept and used content for its own models, which is the fact pattern those cases lacked.
It reaches chatbots and AI features. Any product that forwards user conversations to an AI provider should know whether that provider retains or trains on them. See Session replay and chat recording.
It survives the standing line. Popa, Cook v. GameStop and In re BPS Direct all turn on sensitivity; recorded conversations about health or money sit on the sensitive side.
See CIPA §631, CIPA §632 and Federal Wiretap Act.

How Inth catches this

Code audit: spots a new AI, transcription or chat vendor dependency as it enters the product, and maps the conversation content it receives to the vendor record and the policy that covers it.
Website audit: records a new AI chat or assistant script or domain appearing on your pages, with the time, region and consent state.
Consent banner: where the policy is opt-in, chat waits behind the banner. Inth Consent has no role in a meeting bot inside Zoom or Teams.
Evidence: finding history shows when the vendor was added, by whom, and the approval rationale.

Inth cannot tell whether a vendor trains on the data it receives; that is in the vendor’s contract and settings.

Checks for your own systems

In the repository

List every AI, speech and chat SDK and every server call to a model provider, and what user content each sends.
Check request options for retention or storage flags and record what each is set to.

In the browser and network

Use the AI or chat feature with a test message and note every domain the message reaches, including subprocessors.

With the people responsible

Ask each AI vendor in writing whether it retains inputs, for how long, and whether it trains on them; confirm opt-outs are in the contract.
If staff use meeting bots, check that every participant is told the call is recorded and transcribed, including external guests.

Sources and limits

Source review: 17 October 2026. This page describes the cited records and procedural stage. It does not audit Otter.ai’s current systems. Inth wrote the technical checks; they cannot decide which laws apply or prove compliance.