Is Otter AI Safe for Confidential Meetings? (2026 Audit)

In August 2025, a class-action lawsuit accused Otter.ai of recording confidential workplace conversations and using them to train its AI, without the consent of everyone in the room. If you’re asking whether Otter AI is safe for confidential meetings before a board session, legal review, or patient consultation, you’re asking the right question. And in August 2026, a federal judge let the core wiretap claims in that case proceed.

This audit rests on Otter’s own privacy page, help center, and terms, the court record, legal analyses, and the vendor documentation for the alternatives. Live account tests were not run for this update, and where a claim rests on a help page rather than a hands-on check, that’s noted. Where Otter does something well, that’s noted too.

The Short Answer

Is Otter.ai safe for confidential meetings? Not by default. Otter uploads audio to cloud servers, shares data with third-party sub-processors, and on standard plans may use de-identified recordings for AI training unless you opt out. A federal court has let core wiretap claims proceed. For legal, medical, or HR discussions, local transcription is safer.

That’s the verdict. The rest of this post covers the reasoning, the nuance, and what to do if you’re stuck with Otter anyway.

How Otter.ai Actually Handles Your Meeting Audio

You can’t judge the risk without knowing where your audio goes.

What Happens When You Hit Record

You press record, or Otter’s meeting bot joins through your calendar. (Otter renamed OtterPilot to Otter Notetaker, so older guides use either name.)
Audio streams to Otter’s cloud. Otter stores data on AWS in a US West region, with server-side AES-256 encryption.
Otter’s engine transcribes the audio.
AI features generate summaries, action items, and speaker labels, using third-party LLM providers.
The recording, transcript, and AI outputs sit on Otter’s servers.

Nothing in that chain happens on your device. Closing your laptop doesn’t matter, because the audio is already elsewhere.

Recording modes differ. The Notetaker bot joins Zoom, Google Meet, and Teams as a visible participant. Otter’s desktop app adds a bot-free option. That mode is still cloud-based, but other people on the call get no obvious cue that anything is capturing them.

What Otter Collects Beyond the Audio

Audio of every speaker, not just the account holder
Speaker-identification data — Otter learns to recognize recurring voices, which is how “Speaker 2” stays the same person next week
Meeting metadata: titles, attendees, calendar invites, timestamps, platform
Contact and location details, per the privacy policy

The speaker-recognition point is the one people overlook. A model of how a specific person sounds is biometric-adjacent data, and that person may never have made an Otter account. This is now a live legal issue: the Illinois biometric privacy voiceprint claims survived the motion to dismiss in the Otter litigation.

Otter.ai’s Encryption and Security Infrastructure

Otter isn’t sloppy. Here’s the fair assessment.

What Otter Gets Right

SOC 2 Type II attestation, meaning an auditor tested controls over a period, not just on paper
TLS in transit and AES-256 at rest
2FA for all users; SSO and SCIM on Enterprise
Human-access limits — Otter says staff need explicit customer consent before manually reviewing a recording, and that training runs automatically without human review
No training at the LLM vendors — Otter states its third-party AI providers may not train on customer data

That’s a solid baseline for a cloud SaaS product. If your worry is an outside attacker, these controls address it.

What’s Missing

For confidential work, the gaps are about who can see your data by design, not whether it gets hacked.

No end-to-end encryption. Otter’s servers must read your audio to transcribe it. Otter, and any authorized sub-processor, can technically access your content.
SOC 2 isn’t a guarantee. It shows controls exist and operated, not what Otter is permitted to do with your data.
De-identification is Otter’s own definition. Otter describes a “proprietary method,” with no public technical documentation of how it works.
No public audit artifacts. Ask for the SOC 2 report and the latest penetration-test summary under NDA.
US-only storage. More on that in the GDPR section.

Does Otter.ai Use Your Meetings to Train Its AI?

For standard accounts, yes by default. This is the biggest anxiety driver for readers, and it matters more since August, as you’ll see below.

What the Policy Says

Otter trains its models on user data after applying its own de-identification method. Consent comes through the signup flow, and two problems follow:

It’s opt-out, not opt-in on standard plans.
The signup flow only binds the account holder. Everyone else on the call never saw it.

Otter’s position is that this is disclosed, and it is. But a policy that exists isn’t the same as a participant who knows about it.

How to Opt Out

The answer depends on your plan, and sources conflict on the details.

Enterprise: Otter’s help center says Enterprise workspaces are opted out of AI training by default, enforced through the agreement and workspace configuration, changeable only by your account manager.
Other plans: Reviews describe a toggle at Account → Settings → Data Controls that can revert after subscription changes, or an opt-out by request with no audit trail. Check your own account and screenshot what you see, with the date.
Retroactivity: There’s no clear Otter statement that opting out removes previously ingested data. Ask in writing.

Otter uses third-party LLM providers for summaries and chat. One analysis of Otter’s March 2026 sub-processor list names AWS, OpenAI, Anthropic, and Google Cloud, all US-based. Verify against the current list. Each sub-processor adds a contract, a jurisdiction, and a possible breach surface.

Most posts say “get consent” and move on. The structural problem is deeper.

When someone adds Otter’s Notetaker to a call, that person has accepted Otter’s terms. Everyone else has accepted nothing — not its privacy policy, its sub-processors, or its training clause. A join banner isn’t agreement to have your voice uploaded to a third party’s cloud.

Calendar auto-join makes it worse. Once Otter is linked to a calendar, it scans upcoming events for meeting links and joins them to record. The meeting organizer may never have decided to record.

One-party consent: one participant’s agreement is enough.
All-party consent: everyone must agree. California and Illinois are the commonly cited US examples, and GDPR consent rules impose comparable requirements in the EU.

Federal wiretap law and state equivalents sit on top. This is where the safety question stops being a technical one. A tool can be well secured and still expose you to liability if the recording itself was unlawful.

Real-World Consequences

In September 2024, an AI researcher received an Otter transcript that included hours of investors’ private talk after he had left the call. The email went to everyone on the invite, and the deal didn’t happen. This wasn’t an attack — it was the default behavior doing what it was set to do.
A journalist wrote about interviewing an activist with Otter, then realizing the company shares data with third parties. Otter says it doesn’t share data with foreign governments or law enforcement.

The Otter.ai Class-Action Lawsuit, Explained

What Was Alleged

Filed August 15, 2025, in the Northern District of California
Original plaintiff Justin Brewer, later joined by others in three related suits
Consolidated October 22, 2025, as In re Otter.AI Privacy Litigation, No. 5:25-cv-06911, before Judge Eumi K. Lee. The consolidated complaint followed on December 5, 2025.
Core theory: OtterPilot recorded and transcribed conversations without all-party consent, then used them to train Otter’s AI

The August 2026 Ruling

On August 13, 2026, Judge Lee granted Otter’s motion to dismiss only in part. The federal Wiretap Act claim, the California Invasion of Privacy Act claim, the Illinois biometric privacy voiceprint claims, unjust enrichment, and the unfair competition claim survived. Other claims were dismissed with leave to amend.

The detail worth your attention is the reasoning: one analysis describes the court as finding it plausible that Otter is a third-party eavesdropper because it uses recordings for its own benefit. That links the training default directly to wiretap exposure. Training isn’t only a privacy preference — it may be the legal hook.

Otter’s Defense and What This Does Not Prove

Otter’s stance is that users consent through signup, that its privacy policy discloses training, and that no unlawful interception occurred. The ruling accepts the complaint’s allegations as true for now. It makes no finding that Otter violated any law, and Otter denies it.

Current Status

As of August 20, 2026, reporting showed plaintiffs had until roughly August 27 to file an amended complaint, with none docketed yet. Discovery and a class-certification fight come next. Check the docket before relying on any status line, including this one.

For your organization, the practical point is simple: your counsel now has a public example of how a cloud transcription default can be framed as a consent failure. If you use Otter, they should know about this case.

HIPAA

Otter announced HIPAA compliance in July 2025, but only on the Enterprise plan, with the Business Associate Agreement process starting through your account manager.

Basic, Pro, and Business users can’t use Otter for protected health information in a compliant way. Full stop.
Otter’s own page says the customer is responsible for controlling when PHI enters the environment, including features like Notetaker that auto-join scheduled meetings.
A signed agreement makes the disclosure lawful — it doesn’t remove it. Patient audio still travels to Otter’s cloud.
Ask in writing whether the agreement covers the AI summary and chat features, not just transcription.

Residency: Otter discloses no EU or EEA data residency option, so European customers’ audio is stored in the US.
Transfers: Otter’s privacy policy commits to the EU-U.S., UK, and Swiss data privacy frameworks. Earlier transfer frameworks were struck down twice, so many EU teams still pair it with Standard Contractual Clauses and a transfer impact assessment.
CLOUD Act: US authorities can compel a US provider to produce stored data. EU regulators have flagged this.
Review the data processing agreement terms before signing.

Other Frameworks

CCPA: Otter states it doesn’t sell user data.
FERPA: Relevant for lectures and student meetings. No Otter-specific FERPA agreement was found, though a competitor publishes one.
PCI: Spoken card data in recorded meetings is outside what Otter is built for.
Attorney-client privilege: Third-party access can complicate a privilege claim. Ask counsel, not a blog.

Who Should NOT Use Otter.ai for Confidential Meetings

“Use caution” is useless advice. Here’s a definite list.

Lawyers and legal teams — case strategy and work product shouldn’t pass through a cloud transcriber without an enterprise agreement and client consent
Healthcare providers — any PHI conversation needs an executed agreement, which means Enterprise
Therapists and counselors — session content is among the most sensitive data there is
Financial advisors and firms — client financials and material non-public information carry regulatory exposure
HR teams — investigations, terminations, and accommodation requests are discoverable and damaging if leaked
Journalists — source-protected interviews, especially with whistleblowers or at-risk activists
Boards and M&A teams — deal terms and competitive strategy

For anyone on this list, the default answer is no. The exceptions require Enterprise, contractual protections, and legal sign-off.

If You Must Use Otter: A 7-Step Hardening Checklist

Sometimes the decision was made above your head. Here’s how to shrink the exposure. Verify each path in your own account, since menu names change.

Confirm the plan, and get terms in writing. Pro and Business tiers cap conversation length; Enterprise is custom-quoted, and it’s the tier with the agreement and default no-training terms.
Handle AI training. On Enterprise, confirm the default opt-out in your agreement. On other plans, check the data controls setting, screenshot it with the date, and ask in writing about retroactivity.
Switch off calendar auto-join. Choose manual meeting selection. Events toggled individually keep their old setting, and changes made shortly before a meeting may not take effect — review your calendar after changing it.
Shorten retention. Admins can set a custom retention period through their account manager. Expired conversations disappear for every collaborator.
Turn on 2FA, and SSO where available. Centralize access through your identity provider.
Kill auto-share. Auto-share and pre-recording notification toggles matter — auto-share is how the 2024 investor transcript reached everyone on the invite.
Run a consent protocol. Announce recording at the start of every meeting, get agreement, and document it in all-party consent jurisdictions.

Bonus: request and archive Otter’s sub-processor list, your data processing agreement, and any signed BAA. Keep dated screenshots of every setting.

The failure to watch for is a lopsided rollout — teams handle the quick settings and skip the ones that need habits, so the bot keeps joining meetings nobody chose to record.

Someone Else Brought Otter Into Your Meeting. Now What?

Almost every article speaks to the person choosing the tool. Here’s the view from the other end of the recording.

How to Tell If Otter Is Recording You

Check the attendee list for “Otter Notetaker” or a similar name.
Watch for recording notices, which depend on the host’s settings and the platform.
Know the blind spot — the bot-free desktop mode may show nothing at all.
Ask: “Is anyone recording or transcribing this?” costs nothing.

How to Make It Stop

Any participant, including people without an Otter account, can type “stop otter” in the Zoom, Google Meet, or Teams chat to remove all Notetakers immediately, including multiple bots at once.
The host can also remove Notetaker like any other attendee, and the meeting continues unrecorded by Otter.
Removing the bot ends the recording — it doesn’t delete what was already captured. Ask the account owner to delete it.

Your Rights

All-party consent jurisdictions: you can refuse recording, and recording over your objection may be unlawful.
GDPR: you can request access, erasure, and object to processing.
Otter itself: you can request deletion under the data-subject-rights section of its privacy policy.

For IT Admins

Move from allow-all to an approved list of notetaker apps in Zoom, Google Workspace, and Microsoft Teams, and use waiting rooms so hosts approve every participant. Menu names differ by platform and change often, so check each admin console’s current documentation.

Safer Alternatives to Otter.ai: A Privacy-Architecture Comparison

The most useful distinction isn’t a feature list. It’s where the audio goes.

	Otter.ai	Zoom AI Companion	Teams Copilot	Fireflies.ai	Fellow	Local (Whisper etc.)
Architecture	Cloud	Cloud (native)	Cloud (native)	Cloud	Cloud	On-device
True end-to-end encryption	No	No	No	No*	No	N/A
Trains on your data	Standard: yes, de-identified, opt-out. Enterprise: off by default	Check terms	States no	States no	States no	N/A
BAA available	Enterprise only	Healthcare plans; confirm tier	Eligible plans, under Microsoft’s BAA	Enterprise + Private Storage + signed BAA	Check docs	N/A
Consent model	Visible bot; any participant can remove	Platform-native notice	Platform-native notice	Bot joins	Check docs	You control
Residency	US (AWS)	Varies by plan	Tenant geography	US default; Private Storage on Enterprise	Check docs	Your device
Audio leaves device	Yes	Yes	Yes	Yes	Yes	No

*Fireflies markets AES-256 and TLS as “end-to-end.” That’s standard encryption at rest and in transit, not true end-to-end, because the service still reads your audio.

Fireflies’ HIPAA configuration requires an Enterprise plan, Private Storage, and a signed BAA all at once, and it stops applying if any one is removed. Microsoft says Copilot prompts and responses aren’t used to train its foundation models. Cells marked “check” couldn’t be confirmed from vendor documentation, so verify them before relying on the table.

The cloud vs. local divide matters most. Every cloud tool, however well secured, sends audio to a third party by design. On-device tools built on OpenAI’s Whisper and run on Apple Silicon or a local GPU keep audio on your machine — no upload, no sub-processor chain, no training risk, and no third-party consent problem, because there is no third party.

The trade-off is real. Local tools give up shared workspaces, searchable cloud archives, and polished live multi-speaker transcription. If you need collaboration, you’ll feel it. For a privileged legal call, you probably won’t care.

For teams that must stay in the cloud, platform-native tools keep data inside an ecosystem you may already have contracts with. That’s a smaller footprint, not zero risk.

If you’re choosing between the two most popular cloud note takers, our Otter vs Fireflies comparison breaks down their 2026 pricing, free-plan limits, language support and the lawsuits each one faces.

The Verdict, by Use Case

Non-sensitive internal meetings (standups, brainstorms, project updates): Otter works fine. The convenience is real and the risk is low when nothing privileged or regulated comes up.
Moderately sensitive meetings (client calls, internal strategy, hiring): use Otter only on Enterprise, with the training opt-out confirmed in your agreement, retention set, auto-join off, and a consent protocol in place. Document everything.
Highly confidential meetings (legal privilege, patient data, HR investigations, M&A, source-protected journalism, board discussions): don’t use Otter. The architecture sends audio to external servers and third-party processors by design, and a court has just allowed wiretap claims about that design to proceed. The only safe approach is a tool where audio never leaves your device.

The question isn’t whether Otter is a good transcription tool. It is. The question is whether your meeting can tolerate cloud upload, third-party processing, and, on standard plans, a default that allows AI training. For most confidential meetings, it can’t.

Frequently Asked Questions

Does Otter.ai record meetings without consent? The Notetaker joins as a visible bot and may trigger platform notices, but it doesn’t independently obtain consent from every participant. The host or account owner consents for the meeting, and in all-party consent jurisdictions that may not satisfy the law.

Can Otter.ai join my meeting without me knowing? Yes. If anyone on the invite has Otter linked to a calendar with auto-join turned on, the bot can enter automatically. The bot-free desktop mode shows no visible bot at all.

Does Otter.ai use my data to train its AI? On standard plans, yes by default, using de-identified data. Enterprise workspaces are opted out by default, though whether opting out removes already-ingested data is unclear.

Is Otter.ai HIPAA compliant? Only on the Enterprise plan with an executed Business Associate Agreement. Basic, Pro, and Business plans cannot be used for protected health information.

Does Otter.ai have end-to-end encryption? No. It uses TLS in transit and AES-256 at rest, but its servers must read the audio in order to transcribe it, so true end-to-end encryption is not in place.

How do I remove or block Otter.ai in a meeting? Any participant can type stop otter in the meeting chat, and the host can remove the bot from the attendee list. To prevent it entirely, set auto-join to manual and restrict third-party notetaker apps at the admin level.

What happened in the Otter.ai class-action lawsuit? The case was filed in August 2025 and consolidated in October 2025. On August 13, 2026, the court let Wiretap Act, California privacy, and Illinois biometric privacy claims proceed while dismissing others with leave to amend. This is not a finding of liability.

What are safer alternatives to Otter.ai for confidential meetings? On-device tools based on Whisper keep audio entirely on your machine. For cloud features, platform-native tools like Teams Copilot and Zoom’s healthcare offerings sit inside ecosystems that already have business associate agreements in place.