Recording a Call Is Now a Consent Decision

On 13 August 2026 a federal judge in the Northern District of California declined to throw out the central claims in a privacy case against Otter.ai. The lead plaintiff, Chaka Theus, had been on a video call with a medical professional. Otter's notetaker was on that call as well, and her complaint says the company kept the recording and folded it into the data it uses to train the systems it sells. Judge Eumi K. Lee allowed Theus and most of her co-plaintiffs to continue, in In re Otter.AI Privacy Litigation (5:25-cv-06911).

The order is one of three separate cases against notetaker vendors: Otter, Granola and Fireflies. None of them has ended in a finding of liability, and none of them is about a podcast. They are about meetings in which software recorded people who had not agreed to it, and about a question that has become practical for anyone who records conversations: who in the room can agree on behalf of everyone else?

The mechanisms differ, and so do the weak points. A visible bot at least announces itself. An app that runs on someone else's laptop does not, and Granola's own marketing said so before the lawsuit did: other people in the room will not know it is there. Read AI has not been sued at all. Chapman University banned it in 2025, and the University of Washington blocked its access to Zoom, citing a tool that could join, transcribe and summarise meetings even when participants were not present, and without their awareness or consent.

Otter's defence was that it only did what it was told: somebody invited the bot, somebody pressed record, and whatever consent that person gave should cover the software as well. The court did not accept that at this stage, because the complaint plausibly alleges that Otter independently collects, retains and uses communications for its own commercial purposes. That is the whole dispute in one sentence. A tool that only carries out an instruction sits in the host's hands. A service that keeps the conversation for its own ends is treated as another party in the room, and another party needs its own consent.

None of the changes below waited for a judgment. Microsoft Teams now detects outside note takers and puts them in the meeting lobby regardless of how the lobby is configured, so an organiser has to admit them explicitly, and the tenant policy ships with require approval when detected as the default. Zoom needs a host or co-host to start its own AI Companion before it can join a meeting on Teams or Google Meet, and admins can switch that off. Google Meet's note taking needs the same host approval and can be limited by department. The decision moved from the person who clicked record to the admin console.

1. 01

   Name everyone who will be recorded

   That includes the co-host, the colleague listening in and the guest who joins ten minutes late. A recording that starts before someone knows about it is the pattern in all three lawsuits.

2. 02

   Say it on the recording, in one sentence

   We are recording this call and both of us agree to it. A sentence in the file is the cheapest evidence that nobody was surprised, and it survives every later argument about who said what.

3. 03

   Ask for a yes and wait for it

   Silence is not agreement. Under the laws in play here, an answer is what separates disclosure from consent, and written notice is what some of them ask for.

4. 04

   Keep the file where the conversation happened

   A copy on the host's account with a retention rule and a deletion date is easier to explain than a copy that lives in a vendor's training set. Decide who may delete it before the take, not after.

Airtape is not a bot that joins a call. The host starts the take, the studio counts down, and while the recording runs an ON AIR bar with a running timer is visible to every participant. Each person's camera and microphone are captured locally on their own device and uploaded into the host's session, so a dropped connection costs an upload instead of the file. Guests open a link and type a name: no installer, no account, and nothing that transcribes quietly in the background.

• The start is an event in the room, not a setting somebody switched on months ago.
• Every participant sees the same ON AIR bar, so the timer is the disclosure.
• Each speaker keeps a named, separate track, which is also the record of who was present.
• Deleting a session deletes its files, and the host decides when that happens.

The practical rule that comes out of these cases is not complicated. A recording should begin because someone in the room chose it, and everyone in the room should be able to see that it began. Everything after that, what you do with the file, how long you keep it, which service builds a voiceprint, is a decision you can write down and defend. What gets companies sued is the part that happens without anyone choosing it.