The August 13 Otter.ai Ruling: Why Vendor Training on Your Meetings Just Became a Wiretap Question

Published September 03, 2026

September 3, 2026 · 11 min read

On August 13, 2026, U.S. District Judge Eumi K. Lee of the Northern District of California issued the first federal ruling in In re Otter.AI Privacy Litigation, No. 5:25-cv-06911, and it did not go the way Otter.ai wanted. The court granted the company's motion to dismiss only in part, letting the federal Wiretap Act, California Invasion of Privacy Act §631, and both Illinois Biometric Information Privacy Act claims proceed toward discovery. According to Bloomberg Law's reporting on the order, the plaintiffs adequately pleaded claims under ECPA, CIPA, BIPA, and California's Unfair Competition Law, along with unjust enrichment.

Buried inside the ruling is a holding that will reshape how every AI notetaker vendor gets underwritten, sold, and reviewed: the court accepted the theory that when a transcription vendor retains conversations and uses them to train its own machine-learning models, that independent commercial use plausibly transforms the vendor from "the meeting host's tool" into a distinct third-party eavesdropper under wiretap law. The revenue model became an element of the claim.

What Judge Lee Actually Held

The order is a mixed decision, but the surviving claims are the ones with teeth. As Recording Law's case tracker summarized, the federal Wiretap Act (ECPA) claim, the CIPA §631 claim, both Illinois BIPA voiceprint claims, unjust enrichment, and the UCL claim all survive and head toward discovery. Dismissed with leave to amend: both Computer Fraud and Abuse Act counts, the California CDAFA claim, the Washington Privacy Act claim, and most intrusion-upon-seclusion claims.

Coverage from UC Today's account of the order emphasizes that the ruling does not determine whether Otter broke the law. It finds only that the plaintiffs have pleaded enough facts for the central allegations to proceed beyond the motion-to-dismiss stage. Plaintiffs have 14 days from the August 13 order to submit an amended complaint on the dismissed counts; Otter then has 21 days to respond.

An Otter.ai spokesperson declined to comment on the ruling. The company has denied wrongdoing throughout the litigation.

The Consolidated Case: How Four Suits Became One

The action bundles four putative class suits filed between August and September 2025. As the CIPA case tracker at Recording Law documents, plaintiff Justin Brewer filed the lead complaint on August 15, 2025, alleging that Otter's "Otter Notetaker" and "OtterPilot" tools join meetings on Zoom, Google Meet, and Microsoft Teams, transmit the audio to Otter's servers in real time, and transcribe what every participant says — including participants who are not Otter users and were never asked for consent. Three related complaints followed; Judge Lee consolidated them on October 22, 2025.

The consolidated complaint asserts claims under the federal Wiretap Act, CIPA §§631 and 632, the Illinois Biometric Information Privacy Act, the CFAA, and several common-law theories. Otter moved to dismiss all of them. The August 13 order is the answer.

The Third-Party Eavesdropper Question

The doctrinal fight at the heart of the ruling was whether OtterPilot is a "party" to the conversation (which would defeat wiretap liability under one-party consent) or a "third-party eavesdropper" (which would trigger liability). Otter argued that its user invited the assistant into the meeting, so the user consented on the tool's behalf. Under CIPA §631, if the notetaker is treated as a mere recording tool operated at the host's direction, the wiretap claim collapses.

The court rejected that framing at the pleading stage. As ID Tech Wire reported, the court found Otter was adequately alleged to be a third-party eavesdropper under California law rather than a tool operating for the meeting host. The wiretap claim also proceeded on an alternative theory: a party to a conversation loses the statute's protection when the interception serves a tortious purpose, and the plaintiffs plausibly alleged that theory too.

Why Training on Customer Data Was the Pivot

The most portable piece of the reasoning is where the court drew the line between "tool of the host" and "third-party listener." The line is drawn at whether the vendor retains and reuses the audio for its own purposes. The Lawsuit Intelligencer analysis of the CIPA holding puts it plainly: because plaintiffs alleged Otter retains recordings and uses them to train its own models, the vendor moved outside the line of cases treating a recording service as an extension of its client. That framing makes the revenue model itself an element of the claim. Any AI vendor whose terms permit training on customer content sits inside the holding's logic — the product category is irrelevant.

The BIPA Voiceprint Claims That Survived

The Illinois biometric claims are their own exposure track, and they survived on facts specific to speaker identification. The Illinois plaintiffs allege Otter captured their voiceprints during Zoom and Microsoft Teams calls. Otter argued they had not alleged the voiceprints could identify anyone. The court pointed to allegations that the Otter Notetaker tags speakers in real time using Zoom participant names or data clients enter manually, building a speaker identification profile so it can recognize the same person in later conversations.

That factual pattern — a consistent speaker label attached to a voice pattern across sessions — is exactly the kind of "biometric identifier" that Illinois's Biometric Information Privacy Act was designed to reach. BIPA authorizes statutory damages per violation, and every meeting with a new participant can generate a new violation. For a widely deployed enterprise tool, the exposure math is not subtle.

Cloud AI Notetaker vs. On-Device Transcription: What Actually Differs After August 13

The most useful way to read the ruling is architecturally. The court's third-party-eavesdropper theory rests on specific facts about where audio goes after capture. When those facts change, the analysis changes.

Two things about that table matter. First, on-device architecture does not eliminate consent obligations. In an all-party consent state like California or Illinois, the person doing the recording still needs to notify and obtain consent from every participant. That is a statutory duty the technology cannot discharge. Second, on-device architecture does remove the specific factual predicate Judge Lee relied on — that a separate commercial actor received and reused the audio. There is no separate recipient when the audio never leaves the device.

The Ripple Effect: Granola, Fireflies, and the Rest

The Otter ruling is being read as a template. On July 30, 2026, plaintiffs filed Chamberlain v. Granola, Inc. in the same Northern District of California. Reporting from HR Executive on the Granola complaint notes it accuses the venture-backed notetaker of recording conversations without telling most participants and feeding those recordings into its own AI model training by default. As Tool Directory's tracker of the AI notetaker lawsuits observes, plaintiffs' counsel reached for Granola's own website copy — a line promising other participants "won't know it's there" — as evidence of intent. A sentence written as a product benefit became the basis for a concealment argument.

Fireflies faces a parallel BIPA class action, Cruz v. Fireflies.AI Corp., No. 3:25-cv-03399, in the Central District of Illinois. The National Law Review's analysis from privacy specialists at Sheppard Mullin notes that AI notetakers that process voice characteristics to identify speakers may be generating voiceprints — biometric identifiers regulated under BIPA.

All three cases rest on the same federal wiretap theory. A ruling in the Otter case is now the closest thing to a live precedent, and it landed on the plaintiffs' side of the ledger. For a deeper look at the earlier stage of the Otter litigation, see our plain-English guide to the pre-ruling posture of In re Otter.AI Privacy Litigation.

What the Ruling Does Not Decide

It is worth being precise about what August 13 did not do, because the difference matters for how customers and employers respond.

• It did not find Otter liable. A motion-to-dismiss ruling accepts the complaint's factual allegations as true for the limited purpose of testing legal sufficiency. Otter still denies interception.
• It did not certify a class. The proposed classes remain uncertified. Class certification is a separate fight that will follow discovery.
• It did not set damages. No damages are awarded at the motion-to-dismiss stage. ECPA and CIPA both provide statutory damages per violation, so the exposure math becomes relevant only if the case proceeds to judgment.
• It did not clear other vendors. The ruling's logic is portable but its holdings bind only this case. Every other AI notetaker litigation will be decided on its own record.

The Employer Exposure Question

Employers who deployed OtterPilot into workplace meetings are not defendants in In re Otter.AI, but they sit inside the same fact pattern. As HR Executive's coverage of the employer risk picture laid out earlier this year, the federal Wiretap Act allows private plaintiffs to seek statutory damages calculated as the greater of a per-day amount or a minimum statutory award. BIPA authorizes statutory damages for improper collection of biometric identifiers — including voiceprints used to identify individual speakers. A single training session, all-hands meeting, or applicant interview can generate dozens of alleged violations.

Beginning August 2026, the EU AI Act adds another layer. HR Executive reports that AI systems used for worker monitoring and management may be classified as high-risk, a category that could encompass tools offering sentiment analytics or productivity scoring alongside transcription. In co-determination countries such as Germany and France, deploying an AI notetaker may also require works council consultation before rollout.

The interview-recording pattern in particular carries elevated risk. Our earlier analysis of AI notetakers in job interviews and HR compliance walks through why external candidates, protected characteristics, and post-hoc discovery collide in interview transcripts specifically.

How Basil AI Solves This

The theory of harm Judge Lee accepted is fact-specific. It depends on the vendor being a distinct actor who receives, retains, and reuses the audio. Basil AI is built on a different architecture, and the difference maps directly to the elements of the surviving claims.

Basil AI is a fully on-device AI meeting transcription app for iOS and Mac. Audio is captured and transcribed locally using Apple's Speech Recognition framework, which runs on the Apple Neural Engine. There is no vendor cloud that receives the audio in the first place. There is no separate commercial recipient to characterize as a third-party eavesdropper under CIPA §631, because the audio never leaves the user's device. There is no vendor training pipeline consuming customer conversations, because the vendor never sees them. There is no subpoena surface at Basil for meeting content that Basil never held. Apple's own privacy commitments underpin the platform's on-device processing model.

That architecture does not make recording lawful in every jurisdiction. The user still needs to comply with the consent statutes of the states whose residents are on the call. But it removes the specific factual predicate — vendor retention and vendor training — that the Otter court relied on to move the case past dismissal. For a broader comparison of the category, see our privacy comparison of Granola, Otter, and Basil and the deep dive on bot vs. botless AI notetakers for client-facing meetings.

What to Do This Week

For deployers of any cloud AI notetaker, three items belong on the near-term list:

1. Read the terms. Identify whether your vendor's terms of service permit training on customer content, whether that setting is on by default, and how transcripts and recordings are retained. Vendor training was the fact the court leaned on.
2. Fix the consent flow. In all-party consent states, obtain affirmative consent from every participant before a bot joins. "They can leave the meeting" is not consent. Written notice with an opt-in is closer to the standard courts are gesturing at.
3. Audit the biometric surface. If your notetaker performs speaker identification that persists across meetings, you may be inside BIPA's voiceprint framework in Illinois. That is a separate statutory-damages track from the wiretap claims.

Then decide whether cloud transcription is the architecture you want at all. The Otter ruling did not close the door on cloud AI notetakers, but it opened one on the specific business model that pays for them.

Meeting notes that never leave your device

Basil AI runs 100% on-device on Mac and iPhone. No cloud upload. No vendor training. No third-party eavesdropper problem.

This article is general information, not legal advice. In re Otter.AI Privacy Litigation remains pending. Consult counsel licensed in your jurisdiction about your specific situation.