The AI notetaker that helps you in the meeting testifies against you in discovery.

Through 2025 the AI notetaker legal risk was a footnote in a security review. In 2026 it has a case number. The Otter.ai wiretap litigation put a price on the recording itself — and reframed the transcript as something a plaintiff’s lawyer will one day ask you to produce. The record you never made is the only one that can’t be subpoenaed.

The lawsuit that put a number on it

In August and September 2025, four putative class actions were filed against Otter.ai. On December 5, 2025 they were consolidated as In re Otter.AI Privacy Litigation, No. 5:25-cv-06911-EKL, before Judge Eumi K. Lee in the Northern District of California. The core allegation is blunt: Otter’s Notetaker “recorded and transcribed meeting participants without the consent of all parties” and used that content to help train its models.

The claims are not exotic. They are the oldest wiretap statutes on the books, pointed at a new device:

• California Penal Code § 632 — recording a confidential communication without all-party consent.
• California Penal Code § 631 — wiretapping.
• California Penal Code § 637.2 — the part that stings: statutory damages of $5,000 per violation, or three times actual damages, whichever is greater.
• The federal Wiretap Act, 18 U.S.C. §§ 2510–2522.

Otter argued the motion to dismiss on May 20, 2026. As of this writing no ruling has issued, and no court has held Otter’s practices lawful or unlawful. But the mechanism is already clear to every lawyer watching: § 637.2 prices the harm per recording. In a class action, “per recording” is a multiplier, not a rounding error. This is the same all-party consent map we walked through in Is It Legal to Record Client Calls? — except now it has a docket.

Discovery doesn’t care whether the summary was accurate

Wiretap exposure is the headline. The quieter risk is the one corporate counsel started writing memos about in mid-2026: the transcript is discoverable evidence. Multiple firms converged on the same phrase within weeks of each other — the meeting bot becomes Exhibit A.

Here is why that phrase spread. A human keeps curated minutes: decisions, action items, the sanitized record. An AI notetaker keeps everything — the offhand remark, the joke, the half-formed idea someone corrected thirty seconds later, in a meeting that never would have been recorded at all. Once that artifact exists, litigation rules attach to it:

• You may have to preserve it the moment litigation is reasonably anticipated.
• You may have to produce it in response to a discovery request.
• You may have to explain it — including the parts the AI got wrong and attributed to a named speaker.

An inaccurate summary is worse than no summary. It is a document with your name on it, produced under oath, that says something you did not say. The volume of AI-generated records can expand discovery disputes and weaken negotiating leverage regardless of the merits of the underlying claim. You are litigating over a record you never meant to create.

The record you never made can’t be produced

This is where architecture beats policy. A retention setting is a promise to delete something that existed. Fazit’s posture is that the something never exists in the first place.

• Audio is captured into a fixed-size ring buffer in RAM. It is never written to disk, never uploaded, and overwritten continuously as the call proceeds.
• Transcription runs on-device on Apple Silicon. No audio and no transcript reaches any vendor, so there is no processor to subpoena and nothing to train a model on. The mechanics — including why persisting audio would require a deliberate, reviewable code change — are in Why “Never Records” Is Not Marketing.
• What survives the call is a single Markdown note in your own Obsidian vault, stamped audio_retained: false. It is a file you hold, on your machine, under your own retention policy — the same reason deletion actually reaches it, which we covered in How to Delete Everything an AI Notetaker Knows About You.

Map that against the four records a cloud notetaker generates:

THE FOUR RECORDS AN AI NOTETAKER CREATES —
AND WHO CAN SUBPOENA EACH

1. The audio recording      → wiretap exposure (per-recording penalty)
2. The verbatim transcript  → discoverable; preserves every offhand remark
3. The AI summary           → discoverable; inaccuracies attributed to you
4. The distribution log      → who received it, when, on whose account

Fazit creates exactly one of these — the note — and it lives
in your vault, on your machine. Items 1, 3, and 4 do not exist.
Item 2 is never persisted; the audio is gone before the call ends.

There is no wiretap penalty for a recording that was never fixed to a medium. There is no third-party disclosure — and no privilege-waiver argument — when the audio never leaves the room, a distinction that matters most for the privileged conversations we covered in AI Notetakers for Lawyers. And the note that does exist is yours, in a format you can grep, on hardware you control.

What to do before your next call

You do not need to wait for Judge Lee’s ruling to reduce your exposure. Three moves:

• Inventory what your current tool retains. Audio? Transcript? Summary? For how long, on whose servers, and is any of it used for training? If you cannot answer in one sentence per line, that is the finding.
• Set a default of “no AI notetaker in sensitive meetings.” Board, HR, legal, deal, and investigation calls are where a stray transcript does the most damage. The safest default is no third-party recorder unless someone with authority signs off.
• Prefer capture that produces the fewest artifacts. The ideal notetaker leaves one record — the note — and no recording behind it.

The full security posture is documented at getfazit.com/security.

FAQ

Can an AI notetaker transcript be used as evidence in court?

Yes. Once a transcript, recording, or summary exists, it is generally discoverable — a party may have to preserve and produce it in litigation, including the portions the AI transcribed inaccurately. That is precisely why corporate counsel began restricting notetakers in sensitive meetings in 2026.

Is using an AI notetaker illegal?

Not inherently — but recording a conversation without the consent required in your jurisdiction can be. In all-party-consent states like California, recording a confidential communication without everyone’s consent can trigger statutory damages of $5,000 per violation under Penal Code § 637.2. A tool that transcribes without persisting audio changes that analysis; it does not remove the duty to tell participants.

What is the Otter.ai lawsuit about?

In re Otter.AI Privacy Litigation (N.D. Cal., consolidated December 2025) alleges Otter’s Notetaker recorded and transcribed meeting participants without all-party consent and used the content to train its models, citing California’s wiretap statutes and the federal Wiretap Act. A motion to dismiss was argued in May 2026; the case is undecided.

How does an on-device notetaker reduce discovery risk?

By generating fewer artifacts. Fazit keeps audio in RAM only, transcribes on-device, and writes a single Markdown note to your vault — no cloud recording, no verbatim transcript stored, no distribution log. There is less to preserve, produce, and explain because less was ever created.