Published October 15, 2026 · 11 min read · Compliance & Legal

Otter.ai Ruled a 'Third-Party Eavesdropper': What the August 2026 Order Means for Every AI Notetaker Buyer

Published October 15, 2026

Key takeaways

• Judge Eumi K. Lee held on August 13, 2026 that Otter.ai is plausibly a 'third-party eavesdropper' under CIPA § 631(a) because it uses meeting audio to train its own AI.
• Federal Wiretap Act, CIPA, Illinois BIPA, unjust enrichment, and UCL claims all survive and move to discovery; CFAA and Washington Privacy Act claims were dismissed.
• The reasoning applies to any cloud notetaker that stores audio or uses it for model training — not just Otter.
• Damages math is severe: $5,000 per CIPA violation and $10,000 per ECPA violation, multiplied across potentially millions of captured meetings.
• On-device transcription sidesteps the entire doctrine because no third party ever intercepts the audio in the first place.

Quick answer: On August 13, 2026, Judge Eumi K. Lee of the Northern District of California held that Otter.ai is plausibly a third-party eavesdropper under California Penal Code § 631(a) because it uses captured meeting audio for its own commercial benefit — training AI models. The ruling let federal Wiretap Act, CIPA, Illinois BIPA, unjust enrichment, and UCL claims proceed to discovery.

On August 13, 2026, Judge Eumi K. Lee of the U.S. District Court for the Northern District of California issued a 23-page order in In re Otter.AI Privacy Litigation that will shape every cloud AI notetaker purchase decision for the next twelve months. She granted Otter's motion to dismiss only in part, holding that plaintiffs plausibly alleged Otter acts as a third-party eavesdropper and allowing the federal Wiretap Act, California Invasion of Privacy Act, and Illinois Biometric Information Privacy Act claims to proceed toward discovery. If you buy, deploy, or allow a cloud notetaker in meetings you host, the reasoning in this order is now the operative legal framework you need to understand.

The Ruling in One Sentence

A federal judge held that an AI meeting assistant that captures audio and uses it to train the vendor's own models is not merely a "recording tool" for the meeting host — it is a separate listening entity, and that status is enough to put wiretap and biometric privacy claims in front of a jury. Bloomberg Law reported that plaintiffs "adequately pleaded claims of unjust enrichment and violations of the Electronic Communications Privacy Act, the wiretapping provisions of the California Invasion of Privacy Act, the Illinois Biometric Information Privacy Act, and the California Unfair Competition Law." Those five claims now move toward discovery.

What the Case Is, Procedurally

The case number is In re Otter.AI Privacy Litigation, No. 5:25-cv-06911-EKL (N.D. Cal.), before Judge Eumi K. Lee. It consolidates four putative class actions — Brewer, Walker, Theus, and Winston — filed between August and September 2025 and bundled together in 2026. The full text of the court's order on FindLaw shows plaintiffs asserted violations of ECPA, CFAA, CIPA, CDAFA, the California Constitution, UCL, Illinois BIPA, and the Washington Privacy Act. Otter moved to dismiss the entire complaint for lack of Article III standing and for failure to state a claim.

After three continuances of the hearing date, oral argument was ultimately heard on August 3, 2026 in San Jose, and ten days later the court issued the written order allowing the central claims under federal wiretap, California privacy and Illinois biometric-privacy law to move forward. This is a ruling on the pleadings, not a finding of liability — Otter denies wrongdoing and the merits remain to be litigated — but surviving a motion to dismiss is itself consequential because it unlocks discovery and all the document production, deposition, and expert work that follows.

Why 'Third-Party Eavesdropper' Is the Key Phrase

California Penal Code § 631(a) prohibits a person who, "without the consent of all parties to a communication," willfully reads or attempts to read the contents of a wire or electronic communication. Decades of case law have carved out an exception: a party to the conversation cannot wiretap itself. Otter's lead defense was that its service is simply an extension of the meeting host — the host's own ears, so to speak — and therefore not a separate interceptor.

Judge Lee rejected that framing. As the National Law Review analysis from Troutman Amin explains, the court held Otter is plausibly a third-party eavesdropper precisely because "it uses the recordings for its own benefit" — specifically to train its AI models. That commercial reuse is what breaks the "extension of the party" defense. A silent human note-taker working for the host would not have the same problem; a vendor that absorbs the audio into its own commercial product does.

What Survived, What Was Dismissed

The order was mixed, but the surviving claims are the ones with teeth. Here is the breakdown:

Claim	Statute	Status after August 13, 2026
Federal Wiretap Act	18 U.S.C. § 2511 (ECPA)	Survives
California wiretapping	Cal. Penal Code § 631(a) (CIPA)	Survives
Illinois biometric privacy	740 ILCS 14/15 (BIPA §§ 15(a), 15(b))	Both survive
Unjust enrichment	California common law	Survives
Unfair competition	Cal. Bus. & Prof. § 17200 (UCL)	Survives
Computer Fraud and Abuse Act	18 U.S.C. § 1030 (CFAA)	Dismissed
California Comprehensive Computer Data Access and Fraud Act	Cal. Penal Code § 502 (CDAFA)	Dismissed
Washington Privacy Act	Wash. Rev. Code § 9.73.030	Dismissed

Why the split matters: CFAA and CDAFA require concrete "damage" or "loss" in the computer-intrusion sense, which plaintiffs could not plead on these facts. The surviving claims — wiretap and biometric — don't require economic damage; they are statutory privacy torts with built-in per-violation penalties. That is the dangerous combination for the defendant.

The Damages Math

Once the "third-party eavesdropper" theory survives, the damages exposure is formulaic. CIPA § 637.2 provides $5,000 per violation or three times actual damages, whichever is greater. ECPA provides the greater of actual damages or $10,000 per violation. Illinois BIPA provides $1,000 per negligent and $5,000 per intentional violation of each subsection. The SolidAITech litigation summary notes that 12 US states require all-party consent and that penalties reach felony level in Pennsylvania. Multiply $5,000 by the number of allegedly captured meetings in California alone and you understand why Otter fought the pleading so hard.

Why This Reasoning Applies to Almost Every Cloud Notetaker

The order technically binds only Otter, but the AI Meeting Assistant Reviews analysis is blunt that the reasoning "applies to almost every meeting assistant on the market, not just Otter." The two factual predicates the court relied on — (1) a vendor server receives the audio, and (2) the vendor uses that audio for its own commercial benefit, including model training — are standard architecture across the cloud notetaker category.

The Granola copycat case

Just two weeks before the ruling, on July 30, 2026, a separate class action was filed against Granola in the Northern District of California. The complaint alleges that model training on user meetings is enabled by default on Granola's Free and Business plans, that each individual user must find and disable the setting personally, and that opting out only works going forward because "data already folded into a model cannot be isolated or removed without rebuilding the model from scratch." The Chamberlain v. Granola complaint brings claims under both the federal Wiretap Act and CIPA — the same statutes that just survived in Otter.

Fireflies and the biometric wing

Fireflies.ai faces separate biometric privacy suits in Illinois over voiceprint capture. Those cases now have a template: the Otter order confirmed that BIPA §§ 15(a) and 15(b) claims survive against an AI notetaker that extracts voiceprints without written consent. For a competitor policy comparison, Fireflies' privacy policy and Otter's privacy policy both reserve broad rights to process meeting audio on vendor infrastructure.

The Buyer's Problem: You Can Be the Named Defendant

If you host a Zoom with Otter running and the Chicago-based participant on the call never consented, Illinois BIPA liability can attach to the organization you work for — not just the vendor. Deploying companies have been named as co-defendants in parallel cloud-AI suits throughout 2026. If your general counsel has not yet asked which AI tools your employees are inviting into client meetings, the August 13 order is the moment to force that conversation. Our deeper look at deploying-organization liability walks through the specific vectors.

Checklist: Questions to Bring to Your Vendor's Next Call

1. Where is meeting audio processed — on the user's device, in your cloud, or routed through third-party sub-processors (Deepgram, AWS, OpenAI)?
2. Is audio retained? If so, for how long, and in what geographic region?
3. Is model training on customer audio enabled by default, and is opt-out per-user or per-tenant?
4. Once audio has been used to train a model, can it be removed from the model weights?
5. Do you extract voiceprints or any biometric identifier from the audio?
6. Do you provide a DPA that specifically addresses CIPA § 631 and BIPA § 15 compliance?
7. What notice does the tool present to non-account-holder participants before capture begins?
8. Will you indemnify the deploying organization for wiretap or biometric claims brought by participants?
9. How do you handle discovery subpoenas for stored transcripts in employment or commercial litigation?
10. What is your written response to the August 13, 2026 In re Otter.AI order?

Cloud Notetaker vs On-Device: The Architecture Comparison

Dimension	Cloud notetaker (Otter, Fireflies, Granola, Zoom AI)	On-device (Basil AI)
Where audio is processed	Vendor servers (US or multi-region)	User's iPhone, iPad, or Mac — never leaves the device
Third-party interception under CIPA § 631	Alleged basis of surviving claim	No third party involved in the capture path
Voiceprint extraction (BIPA)	Yes for some products; alleged basis of BIPA claims	No voiceprint sent off-device
Model-training reuse	Often default-on; opt-out per user	No vendor training pipeline — nothing to opt out of
Subpoena surface	Vendor can be compelled to produce stored audio and transcripts	Vendor has nothing to produce — only the user does
Offline support	Partial or none	Full — Apple Speech framework runs with no network
Data residency control	Vendor-defined region	User-defined device, in the user's physical possession

Regulatory Backdrop: What This Interacts With

The ruling does not sit alone. European regulators enforcing Article 5 of the GDPR have for years treated meeting audio as personal data subject to purpose-limitation and minimization requirements, and transfers to U.S. vendors for model training require a lawful basis most customers cannot establish. Under California's CCPA, voice data is "sensitive personal information," which triggers additional notice and opt-out duties. And HHS HIPAA guidance requires a BAA for any vendor that creates, receives, maintains, or transmits PHI — a bar most cloud notetaker vendors either do not meet or meet only on paid enterprise tiers with narrow scope.

For a deeper walkthrough of the two-party consent terrain the Otter case sits inside, see our compliance guide for AI notetaker users in two-party consent states.

What Judge Lee Did Not Decide

It is worth being precise about what has and has not happened. The court has not held Otter's practices unlawful. It has not certified a class. It has not reached the merits of any claim. What the order did was conclude that the plaintiffs' complaint, taken as true for purposes of the motion, states claims under five statutes that are legally viable. Those claims now proceed into discovery, where plaintiffs will subpoena Otter's engineering documents, training-data pipelines, and consent-flow screenshots. That discovery record is often what forces settlement — and settlement at the scale implied by per-violation damages would reshape the industry.

How Basil AI Solves This

Basil AI is architected so the "third-party eavesdropper" question never arises. Three concrete technical differentiators matter here:

1. Apple Neural Engine processing. Transcription runs on the device's Neural Engine using Apple's on-device speech recognition. Audio is converted to text on the same silicon that renders your lock screen. For the underlying framework see the Apple Speech framework documentation.
2. Zero network egress for audio. There is no Basil server that receives your meeting audio because there is no capture path to one. The vendor cannot be subpoenaed for recordings it never possessed, and cannot be a "third party" to a conversation it never joined. For the broader architecture, see Apple's privacy page.
3. No training pipeline on user content. Because the audio never reaches a Basil server, there is categorically nothing to feed into model training. The In re Otter.AI order's "commercial benefit" prong — the trigger that defeated Otter's extension-of-the-host defense — has no analog in Basil's architecture.

For a technical deep dive on how on-device transcription works end-to-end, see our walkthrough of what a bot-free meeting recorder actually is and our comparison of the Granola vs Basil AI bot-free architecture.

What to Do Monday Morning

If you are a buyer, GC, or security lead:

• Inventory which AI notetakers your organization currently permits — including shadow IT that employees install personally.
• Pull the vendor privacy policies and training-data clauses; the Otter policy and Zoom privacy statement are the two most relevant starting points.
• Prohibit cloud notetakers from any call involving California, Illinois, Washington, Florida, Pennsylvania, or other all-party-consent-state participants until written all-party consent is captured.
• For privileged, HR, M&A, or regulated conversations, require on-device-only capture or no AI capture at all.
• Document the policy. The In re Otter.AI plaintiffs' theory of deploying-organization complicity will be tested in parallel suits throughout 2027.

Bottom Line

The August 13, 2026 order is the first federal ruling to accept, at the pleading stage, that a cloud AI notetaker that trains on user audio can be a wiretap defendant in its own right. The reasoning is architecture-dependent, not vendor-dependent, which means every cloud notetaker on the market now operates in the shadow of this precedent. On-device transcription is not a workaround — it is the only architecture where the underlying predicate (a vendor "third party" possessing the audio) does not exist. For privacy-sensitive work, the simplest answer is also the most defensible one.