Chamberlain v. Granola: The "Invisible" AI Note-Taker Wiretap Case

The Granola lawsuit moves the privacy-litigation story off the website and into the meeting room. Granola is an AI note-taker: it listens to a call or a meeting and produces a tidy summary. The complaint in Chamberlain v. Granola alleges that it does this by silently capturing the conversation through the user's device — without the little "so-and-so's notetaker has joined" bot that participants are used to seeing — so the other people on the call may never know software is recording them. Under California's all-party-consent recording law, recording people who haven't agreed is exactly the conduct the statute was built to punish. Different technology, identical principle to the website cases: something captured a communication before everyone consented.

What the case is about

Filed July 30, 2026 in the Northern District of California, Chamberlain v. Granola, Inc. brings claims under the federal Wiretap Act (ECPA) and the California Invasion of Privacy Act. The core allegation is that Granola's AI notepad "recorded and used private conversations without everyone's consent." According to the complaint, the app operates by silent capture — it listens through the device of the person using it, rather than joining the call as a visible participant — which means others on the call "may never learn that software is capturing what they say."

The complaint goes a step further than a typical recording claim. It alleges that Granola "uses captured communications to train AI models by default," without clear per-participant permission, and that once a conversation has been absorbed into a trained model, that data "cannot be cleanly extracted or deleted." The named plaintiff, a Florida resident, seeks to represent a nationwide class of people whose conversations were allegedly captured. The case is new; Granola had not yet filed a response as of publication.

The detail that makes it matter

The feature at the center of the case isn't a bug — it's a selling point. "No bot in the call" is marketed as a benefit: your meeting looks normal, no awkward robot participant, just clean notes afterward. The lawsuit reframes that same design as the problem. If the other participants can't see that recording is happening, they can't consent to it — and in an all-party-consent state, that absence of visible notice is precisely what creates liability.

That's a uncomfortable pattern for a whole category of products. The "invisible," "frictionless," "no one has to know" design that makes AI note-takers pleasant to use is the same design that strips away the notice consent law depends on. The second wrinkle — training models on captured audio by default, irreversibly — turns a one-time recording problem into a permanent-data problem. You can delete a recording; you can't easily un-train a model.

The all-party-consent theory

California is a two-party (all-party) consent state: under CIPA, recording a confidential communication generally requires the consent of everyone party to it, not just the person doing the recording. Cal. Penal Code §632 prohibits recording a confidential conversation without all-party consent; §631 addresses interception. The complaint pairs these with the federal Wiretap Act (ECPA, 18 U.S.C. §2511), which prohibits intercepting the contents of a communication without proper authorization.

The logic maps cleanly onto an AI note-taker. If software captures the content of a meeting — the actual words spoken — and the participants on the other side never agreed, that is the classic recording-without-consent scenario, just with an AI transcription layer on top. The "user consented by installing the app" argument doesn't resolve it, because the user is only one party; the rest of the room didn't install anything.

The model-training allegation adds a dimension older recording cases didn't have. A traditional unlawful recording is, at least, a discrete file you can locate and delete. The complaint's claim that captured conversations are used to train AI models "by default," and then "cannot be cleanly extracted or deleted," describes harm that doesn't reverse. If true, a conversation someone never agreed to have recorded doesn't just sit in storage — it becomes part of a model's learned behavior, beyond the reach of a normal deletion request. That permanence is likely to feature in how courts weigh these cases, and it's a sharp reminder for any business evaluating an AI tool: the question isn't only "does it record?" but "where does what it records end up, and can that ever be undone?"

Part of a bigger wave

Granola isn't alone. A cluster of 2026 cases targets the same "invisible AI note-taker" design across different products — Otter.ai has faced wiretap allegations over its meeting assistant, and the category as a whole is drawing scrutiny for recording and, in some suits, voiceprint/biometric concerns. The common thread is that the convenience feature — automatic, silent, always-on capture — collides with consent laws written for a world where recording was a deliberate, visible act.

For businesses, this matters in two directions at once. If you build a tool that captures conversations, the design question "can every participant tell this is happening, and did they agree?" is now a litigation question. And if you use an AI note-taker in your own meetings — sales calls, interviews, client sessions — you may be the one doing the recording in the eyes of the law, which means the consent obligation can land on you, not just the vendor.

There's a reason this category is attracting suits now rather than two years ago: adoption exploded. AI note-takers went from a niche productivity tool to a default setting in sales teams, recruiting pipelines, and customer-success orgs — often rolled out company-wide without anyone asking whether the people on the other side of those calls had agreed to be recorded and transcribed. Every one of those calls is a potential all-party-consent event, multiplied across every rep and every meeting. That volume is exactly the dynamic that made website tracking attractive to plaintiffs: a single, repeated, automated behavior that touches an enormous number of people the same way. The litigation tends to follow the adoption curve, and AI capture tools are now well up that curve.

Why it belongs on a CIPA tracker

At first glance an AI note-taker case looks out of place next to website-pixel settlements like Forbes or the healthcare portal cases. It belongs there because the underlying legal engine is the same CIPA machinery, and because it shows where the theory is heading.

Tracking the Granola case alongside the website cases is how you see the trajectory: the "capture before consent" theory started with cookies and pixels, and it's now reaching any technology that quietly ingests a communication. A business that only thinks about consent on its website is defending one door while another opens.

What this means for you

The direct lesson depends on which side of the tool you're on, but the principle is one sentence: if something you run captures a communication, every party to it needs to know and agree.

Frequently asked questions