Overview
Otter.ai transcribes meetings. Its notetaker bot reads a calendar, joins the Zoom, Teams or Google Meet call at the appointed time, records it, and files a transcript and summary in a searchable workspace. Otter says it closed 2025 at $100 million in annual recurring revenue, with more than 25 million users and a headcount reported at under 200.
What the company sells has moved. Through 2025 and 2026 Otter repositioned itself from notetaker to what it calls a Conversational Knowledge Engine, and the headline product is now a set of agents rather than a transcript. The Meeting Agent joins calls whether or not you turn up, answers questions out loud during the call by searching the organisation’s accumulated meeting archive, drafts follow-up email and books tasks. Sales and SDR variants of the same agent target revenue teams, with live objection coaching and, per Otter’s own marketing, running product demos.
That shift is why this entry sits in enterprise rather than assistant. The product being bought is no longer a record of one meeting. It is a permanent, cross-team, cross-year corpus of everything anyone said, and the person who decides it exists is rarely the person most exposed by it.
What It Knows About You
The capture is broad by design: full meeting audio, the transcript derived from it, speaker labels, participant lists, calendar metadata, and anything uploaded for transcription after the fact.
Two details do most of the work in the privacy score. First, Otter’s privacy policy states that the company trains its own speech models on de-identified customer audio and transcriptions, and it acknowledges that material used for training may still contain personal information. The de-identification method is proprietary and not publicly described, which means an outsider cannot check the claim that the two statements sit comfortably together. Training on customer content is the default on the consumer tier and applies to business-tier data as well; the HIPAA configuration on the Enterprise plan is where Otter says it restricts use of customer data for model training.
Second, the system builds a speaker profile. Court filings describe the notetaker tagging speakers in real time from participant names or manually entered data, so that the same voice can be recognised in later, unrelated conversations. That is the fact the Illinois biometric claim turns on.
Otter’s compliance posture is not thin, and this Index should say so. It holds a SOC 2 Type II attestation, announced HIPAA compliance in July 2025 for Enterprise, supports SAML SSO with Okta and Microsoft Entra ID, SCIM provisioning, workspace-wide two-factor enforcement, and admin-set retention windows that auto-delete conversations after a chosen duration. If your organisation buys the top tier and configures it, the controls exist.
The gap is that almost nobody on the call is your organisation. The other seven people on that Zoom are recorded, transcribed and speaker-profiled under a consent that one participant gave on their behalf.
The Real Risks
The litigation is consolidated, live, and past the first hurdle. Four suits filed against Otter between August and September 2025 were consolidated in the Northern District of California as In re Otter.AI Privacy Litigation. On 13 August 2026, Judge Eumi K. Lee granted Otter’s motion to dismiss in part but let the central claims through: the federal Wiretap Act, California’s Invasion of Privacy Act, Illinois BIPA, unfair competition and unjust enrichment. Claims under the Computer Fraud and Abuse Act and California’s computer access statute were dismissed, as were a Washington Privacy Act claim and some intrusion claims, several with leave to amend. The reasoning that matters commercially is the court’s willingness to treat the bot, at the pleading stage, as a third-party eavesdropper rather than a participant in the call, on the theory that it captures the conversation for the vendor’s own benefit. Nothing here is a finding that Otter broke the law. It is a ruling that the allegations are plausible enough to litigate, and that alone reshapes the risk of deploying any notetaker.
Voiceprints carry per-violation damages. BIPA treats a voiceprint the way it treats a fingerprint, with liquidated damages of $1,000 per negligent violation and $5,000 per intentional one. Applied to a product whose whole function is to identify who is speaking, across millions of meetings, the arithmetic is the exposure.
The bot has outlasted the meeting. In September 2024, engineer Alex Bilzerian finished a call with a venture firm, logged off, and later received an Otter transcript containing hours of the investors’ private conversation after his departure, including discussion of internal problems. The investors apologised. The deal did not survive. Earlier, in 2022, a Politico journalist who transcribed an interview with a Uyghur rights activist received a survey from Otter asking about the purpose of the interview, which is a reminder that the transcript is not only sitting on your laptop.
The compliance exposure lands on the person who invited it. Otter’s default flow does not collect consent from other attendees. In jurisdictions requiring all-party consent to record a conversation, that obligation does not disappear because a vendor automated the recording; it attaches to whoever brought the bot into the room. Recording rules vary by state and country and this is not legal guidance, but the structural point is simple: the risk and the convenience are held by different people.
Alternatives
- Your meeting platform’s own recording and transcription. Zoom, Teams and Google Meet all transcribe natively. The data is no less sensitive, but it stays inside a vendor your organisation already has a contract and a DPA with, and the recording indicator is the one participants already recognise. It does not solve consent, it just stops adding a party.
- Local, on-device transcription. Tools built on locally run speech models transcribe from a recording on your own machine without the audio leaving it. Slower and less polished, with no shared workspace and no searchable archive, which is exactly the trade.
- Otter Enterprise, configured, if the team is committed to it. The HIPAA-aligned setup restricts training on customer data, and admin retention limits and audit logging are real controls. This only helps if someone actually sets the retention window and enforces SSO; the defaults are not the protective configuration.
- One person takes notes and circulates them. Free, needs nobody’s consent, produces a record of what mattered rather than a record of everything, and leaves no archive for a future subpoena or breach to find. It costs one person’s attention per meeting, and that is a genuine cost, but it is worth pricing honestly against the alternative.
Our Verdict
D, and the shape of the scores matters more than the letter.
Privacy carries this rating at 8. A product that records people who did not agree, builds speaker profiles from their voices, trains on the audio by default outside the Enterprise tier, and has had claims under three separate privacy statutes survive dismissal is not a borderline case. Autonomy sits at 7 because the 2026 product is an agent that attends meetings you skip and acts on what it hears, and lock-in at 7 because export exists but the exits narrow inside a workspace: account deletion is unavailable to a workspace owner while other members remain, admins can disable transcript display in a way that removes the export option, and conversation ownership changes route through Otter support.
Two scores are deliberately unexciting. Job threat is 5. Automated transcription is displacing first-draft work, and the number of working court reporters in the United States has fallen by around 21 percent over a decade, but the reporting in 2026 describes a shortage of certified human transcribers alongside rising demand for people to correct machine output, not mass displacement. The SDR agent is the part of the product line that aims squarely at a paid role. Bias is 4, and that is the score with the least evidence behind it: no regulator finding or published audit of accent or dialect disparity specific to Otter surfaced in this research, so the score reflects the ordinary and well-documented unevenness of speech recognition rather than anything proven about this vendor. If an audit lands, this number moves.
What would raise the grade: consent collected from every participant by default rather than from the host alone, training on customer audio switched to opt-in across all tiers, a published description of the de-identification method that an outside auditor could test, and unblocked export and account deletion for anyone whose voice is in the archive. What would lower it: a certified class, or an adverse merits ruling on the wiretap or biometric claims.
As of September 2026, the litigation is at the amended-complaint stage. There is no certified class, no settlement, and no claim form, and anyone telling you otherwise is selling something.