AI Note-Taking Tools, Silent Recording, and the Consent Problem

AI note-taking tools are now embedded across enterprise workflows, but a growing line of class action litigation is exposing a foundational legal problem: most of these tools capture the communications of people who never consented, and some are designed so those people will never know.

Background: The Granola Lawsuit and Granola’s Note-Taking Tool 

In Chamberlain v. Granola, Inc., filed in the U.S. District Court for the Northern District of California, a putative class of plaintiffs targets Granola, an AI notepad provider that syncs into your calendar and takes notes during meetings. Granola’s AI note-taking tool is unique; it never announces itself in a meeting. In other words, unlike its competitors, no bot joins the meeting and no notification appears for meeting participants when someone in the meeting uses Granola’s AI note-taking tool. Instead, Granola uses system audio and microphone input to capture and transcribe all participants’ communications in real time on Google Meet, Zoom, Microsoft Teams, and other virtual meeting platforms. Granola states that it does not record or save audio, and that it does not access stored recordings. 

Chamberlain v. Granola, Inc., which also names Granola Labs Ltd., alleges that this design is intentional. The complaint emphasizes that Granola actively markets the absence of a visible meeting bot as its core differentiator. Plaintiffs even quote Granola’s own website: “Other people in the room won’t know it’s there.” This distinction, plaintiffs argue, “matters most.” Granola’s anonymous nature is emphasized by its optional transparency features that default to off. While Granola offers a chat-message notification and a video watermark and users “can enable one or both,” neither is enabled by default. And Granola expressly states it “does not configure these settings on behalf of customers.” 

Plaintiffs bring a variety of privacy-related claims against Granola for its practices:  

1. Common law intrusion upon seclusion  
2. Violation of the ECPA, 18 U.S.C. §§ 2510 et seq.  
3. Wiretapping under the California Invasion of Privacy Act (CIPA), Cal. Penal Code § 631;  
4. Recording confidential communications under CIPA, Cal. Penal Code § 632 
5. Violation of the California Comprehensive Computer Data Access and Fraud Act (CDAFA), Cal. Penal Code § 502 
6. Violation of the California Unfair Competition Law (UCL), Cal. Bus. & Prof. Code § 17200 et seq. 
7. Unjust enrichment

All of these claims are premised on Granola’s ability to run and record in meetings without any notification to meeting participants. The lawsuit proposes two classes: a nationwide class of all U.S. persons whose communications were recorded by Granola and a California subclass. These are allegations only; Granola has not yet responded on the merits. 

The ECPA Claim’s Structural Vulnerability 

The federal wiretap count faces a significant hurdle. Under ECPA’s one-party consent framework, an interception is lawful if one party to the communication, here, the Granola user, consents. To overcome this, plaintiffs invoke the statutory exception for interceptions conducted for a “criminal or tortious purpose.” But the Ninth Circuit has interpreted that exception to require a purpose other than the interception itself, and the torts plaintiffs allege are essentially the recording. If the exception fails, so does the federal count, leaving plaintiffs with California state statutes, a Florida-based named plaintiff, and a considerably harder path to nationwide class certification. 

Granola’s practices create two additional problems. First, Granola shifts legal obligations to the user, stating that customers “remain responsible for determining what notice or consent is required for their use case and jurisdiction.” Plaintiffs frame this as an attempt to disclaim liability for a product that is unlawful by design. Second, Granola allows for artificial intelligence training by default. On free and business plans, anonymized data is used for model improvements unless each user individually disables the feature. Even after opt-out, Granola states it “cannot guarantee that anonymized data wasn’t used before you changed the setting.”

The Line of Cases: Otter.ai, Fireflies.ai, and now Granola

The Granola class action joins other pending class actions against AI note-taking providers. In re Otter.AI Privacy Litigation, No. 5:25-cv-06911 (N.D. Cal.), consolidates four suits filed between August and September 2025 and is pending in the same district. Fireflies.ai faces biometric privacy suits in Illinois, including Cruz v. Fireflies.AI Corp., No. 3:25-cv-03399 (C.D. Ill. Dec. 18, 2025); a separate action, Parrinello v. Fireflies.AI Corp., No. 3:26-cv-02479, is pending in the Northern District of California.

In the Otter.ai and Granola cases, the plaintiffs allege ECPA violations, CIPA violations, and intrusion upon seclusion. The Fireflies.ai suits in Illinois proceed on a different theory, under the Illinois Biometric Information Privacy Act (BIPA). In each case, the plaintiffs are challenging these providers’ ability to capture and use participants’ communications during virtual video conferencing. No court has ruled on the merits in any of them; Otter.ai’s motion to dismiss has been argued and remains pending.

The Otter.ai case is particularly important to watch. It is pending in the same district and is likely to produce the first ruling on whether a transcription service constitutes a third-party eavesdropper or is merely a tool of the person who activated it. That framing will directly affect Granola’s exposure, but the precedent may not be symmetrical. Granola’s facts are worse than Otter.ai’s on the notice question, because Otter.ai at least sends a visible bot into the meeting. An Otter.ai loss (finding that even a visible AI notetaker is an unlawful eavesdropper) would be a very bad sign for Granola. But an Otter.ai win would not necessarily save Granola, because Granola’s entire value proposition is that no one knows it’s there. A court could distinguish the two on that basis alone.

Notably, Otter.ai works differently. It is not stealth. It joins meetings as a visible bot, but it presents a different containment problem. When a single user grants Otter.ai access (sometimes via a Single Sign-On (SSO) click if the system administrator has not blocked the application), Otter.ai embeds in that user’s calendar, mail, Teams, and chat, auto-joins all meetings and invitations (including declined ones), and then propagates by emailing attendees and sending notes on behalf of the credentialed user, prompting others to grant access themselves. Once inside an organization, Otter.ai is nearly impossible to remove, and even after a system administrator revokes access, data already captured is not deleted from Otter.ai’s systems.

Fireflies.ai, by contrast, faces claims specifically grounded in biometric privacy: the Illinois complaints allege that its speaker-recognition feature generates voiceprints of participants who never consented.

In other words, all three cases target the same fundamental design choice: capturing the communications of non-consenting participants and using that data for commercial purposes. The difference is mechanism. Granola operates on stealth by design, Otter.ai operates on viral credential propagation, and Fireflies.ai on biometric data collection.

Granola stands out because the other tools, Otter.ai and Fireflies.ai, announce themselves. Granola is the first major product in this litigation line that is designed never to be disclosed at all. The absence of notice isn’t a user error or a configuration oversight; it is the advertised value proposition. This matters for two reasons. Legally, it eliminates an argument that the non-user participant had constructive notice or an opportunity to object or leave the meeting. Commercially, the product’s competitive positioning is itself the alleged violation.

Notice and Consent in the Age of Always-On Recording

Privacy laws, such as the ECPA and CIPA, were built around a world where recording required an affirmative act with a visible device. We are now in a world where ambient, continuous, invisible capture exists.

The broader litigation trajectory reflects a shift in the underlying legal fight. Early disputes in this space concerned whether AI recordings were permissible at all. The current wave, including Granola, Otter.ai, and Fireflies.ai, has moved past that threshold question. The fight is now about how these recordings are made, what rights attach to the recorded parties, and especially what uses the AI model makes of captured communications. The Granola case sharpens this further because it involves not just implied consent but essentially no consent; participants have no way to know the tool is active.

This raises problems for companies employing tools like Granola: how do you provide meaningful notice when the product is invisible by design? Do you send a chat message after recording starts? Do you use a video watermark? Do you send an email after the meeting? Do any of these options even give the non-user participant a meaningful opportunity to withhold consent? And what constitutes valid consent?

What Companies Should be Doing Now

Companies utilizing an “invisible” AI service, such as Granola, should do the following:

• Conduct an immediate inventory of which AI note-taking / transcription tools are active in the organization’s environment, including shadow-IT deployments.
• Review SSO and application-access permissions, particularly for tools like Otter.ai that auto-propagate through credential chains. System administrators should proactively block unauthorized applications and audit whether existing OAuth grants have already been exploited. Because revocation of access does not result in deletion of data already captured by the vendor, early intervention is critical.
• Establish and enforce a policy on meeting recording and AI notetaking that addresses:
  • Which tools are approved
  • What notice must be provided
  • Whether recording requires all-party consent or one-party consent depending on jurisdiction; and 
  • How to handle meetings with participants in two-party-consent jurisdictions.
• Evaluate vendor contracts and DPAs: Does the vendor use meeting data for model training? Is that use disclosed to meeting participants? Does the contract give you the right to require deletion, and does the vendor’s architecture actually permit deletion once data has been incorporated into model training? Granola’s own acknowledgment that trained-in data “cannot be extracted” suggests that contractual deletion rights may be unenforceable in practice for some vendors. 

The Regulatory and Litigation Trajectory 

The Granola, Otter.ai, and Fireflies.ai class actions are likely just the beginning. Currently, our existing privacy law structure is inadequate for products whose primary privacy impact falls on non-purchasers. Until the law catches up, litigation will fill the gap. 

The Otter.ai motion to dismiss, still pending, will likely be the first substantive ruling in this space. Companies should track it closely. A ruling that transcription services constitute third-party eavesdropping, rather than a tool of the consenting user, would validate the theory underlying all three cases and significantly increase exposure for every company in this market. Conversely, even a ruling favorable to Otter.ai would not necessarily protect tools like Granola that provide no notice whatsoever. 

Because there is a gap in our privacy law structure, congressional interest in AI training data provenance and consent is escalating. Companies should continue to monitor legislative developments as well as the pending motions in the Northern District of California.