Are AI Notetakers Safe? An Honest Look at Privacy, Consent, and Data Practices [2026]

An AI notetaker in the participant list is ordinary now. But since the tool stores recordings on  servers you don’t control, it raises new questions about AI notetaker privacy and security. The short answer is that AI notetakers are safe when the tool holds SOC 2 Type II certification, contractually bars training AI models on customer data, and is used with the consent your jurisdiction requires. That's a conditional answer, and the good news is each condition is verifiable before your first meeting.

This guide covers the real privacy risks, what to look for in a trustworthy tool, and how Fireflies addresses each one.

What Are the Real Privacy Risks of AI Notetakers?

Privacy concerns about AI notetakers trace back to five specific things about how these tools capture, store, and share what people say in a meeting. Each one is documented in statute, in published privacy policies, or in how the products are built.

1. Recording without everyone's knowledge or consent

Some meeting bots join calls automatically, and participants do not always notice when one arrives, especially in a large meeting. That mechanism is where the consent question starts, because recording rules in many jurisdictions require every participant to agree before a conversation is captured. In 10 US states, recording someone without every participant’s agreement carries criminal penalties, and civil liability in most of them.California, Florida, Illinois, Maryland, Massachusetts, Montana, New Hampshire, Pennsylvania and Washington require consent from everyone on the call. In Oregon, the law asks only that everyone be told. 

2. Data storage and access by outside parties

Meeting audio, transcripts, and summaries sit on the company’s servers, and often on cloud infrastructure that company rents from someone else. Every additional party in that chain is another set of credentials and another access policy, which means another place a recording of your meeting can be reached by someone who was never in the room.

3. AI training on customer data

Some companies reserve the right to use customer content to train or fine tune their models, and others use broader language about improving the service, which can extend to human review of meeting samples. Meeting audio also usually passes through at least one outside model provider, and that provider may retain the content under its own terms, which is a separate question from whether the vendor itself trains on it.

4. Voiceprints and biometric privacy law

An AI notetaker that identifies who is speaking may generate a voiceprint, a mathematical pattern of a person's voice distinctive enough to identify them again in a different recording. That puts a voiceprint in the same legal category as a fingerprint or a face scan, because unlike a password, nobody can change their voice after a breach. Several US states regulate biometric identifiers directly. Illinois goes furthest under its Biometric Information Privacy Act, which requires written notice and consent before collection and is the only such law that lets individuals sue the collector themselves. A 2024 amendment limited recovery to one violation per person for repeated collection by the same method, which reduced the exposure per person without changing the consent requirement. Texas, Washington and Colorado have their own versions, among others, enforced by state attorneys general. In Colorado, district attorneys can enforce them as well. This area of law is still moving.

5. Where the transcript goes after the meeting

Some tools send the transcript to every participant automatically, and some make every recording visible to anyone in the workspace. Either way, a conversation that four people had becomes a searchable document that people who were never in the meeting can read months later.

Consent and Recording Law

Recording a meeting with an AI notetaker is legal in most jurisdictions, including across the United States and the EU, when participants are informed and the applicable consent standard is met. What varies is the standard itself. US federal law permits recording when one party consents, states can be and often are stricter, and the EU requires a documented lawful basis before anyone hits record.

Which law applies is the harder question, and your company's headquarters does not answer it anywhere. Under Article 3, GDPR reaches any organization processing data about people located in the EU, wherever that organization sits. US courts have split on whether the controlling law is the one where the recording happens or the one where the recorded person is sitting, and in Kearney v. Salomon Smith Barney (2006), California's supreme court applied California law to a firm recording from Georgia.

One-party vs all-party consent in the United States

Most states follow one-party consent, meaning the recording is lawful as long as one person on the call knows about it. In practice, that person is the one who started it. Roughly a dozen states apply an all-party consent standard instead, where everyone has to agree before the conversation is recorded. The exact count depends on how each statute is read.

A call with people in three states can answer to three standards at once. Following the strictest is a risk management convention, and no court has announced it as a rule. For the full breakdown, see call recording laws by state.

GDPR consent requirements for recording in Europe

Under GDPR, recording someone in the EU requires a lawful basis under Article 6, usually consent or legitimate interest, and legitimate interest only holds after a balancing test that weighs your purpose against the participant's privacy. Telling people you are recording is separately required under Articles 13 and 14. Notice on its own is not a lawful basis.

A second layer applies once an AI notetaker identifies individual speakers. Processing audio to uniquely identifymay constitute biometric data under Article 9, which generally requires explicit consent. The European Data Protection Board draws that line at the point where a recording is technically processed to identify someone.

Organizations should confirm their own obligations with counsel, since a general summary will not cover them. For the operational side, see GDPR call recording best practices.

Notify before you record

Whatever law applies, telling everyone before recording starts is the approach that holds up, because the announcement doubles as the moment you ask for consent.

AI notetakers capture meetings in more than one way. A meeting bot shows up in the attendee list, which is real notice for anyone who looks at it, and not everyone does. Without something said in advance, those participants can reasonably feel they never agreed to be recorded. Some tools skip the bot and record the call directly, and those usually include a notification feature that tells participants recording has started.

What Makes an AI Notetaker Safe? A Checklist

Nine criteria determine whether an AI notetaker is safe enough for a specific meeting. Most of these are published on the tool’s trust center, in its privacy policy, or on its plan comparison page.

1. SOC 2 Type II certification

System and Organization Controls (SOC) 2 Type II is an attestation report showing that security controls actually run in an organization. An external, independent auditor tested it by observing how the tool operates over a period of months. Normally, the certification status is published on the trust center or security page. The full report is usually available under NDA, which matters for a procurement review.

2. GDPR compliance

The General Data Protection Regulation (GDPR) is a privacy and security law that applies to any tool processing data from people in the EU. It gives the people you record the right to have that recording erased. The privacy policy or a dedicated GDPR page should link the data processing agreement, the subprocessor list, and the data residency options.

3. HIPAA compliance with a Business Associate Agreement

Health Insurance Portability and Accountability Act (HIPAA) is a US federal law that protects the privacy and security of patient medical records and health information. HIPAA only protects you if the company signs a Business Associate Agreement (BAA), which makes it legally accountable for the health information on your calls. HIPAA compliance is usually limited to specific paid plans. The tool's pricing page shows which tier includes the BAA, and it is rarely the free one.

4. A contractual ban on training AI models with your data

Without this ban, your meetings can become training data for a model other people query later. The commitment has to reach the tool’s own AI suppliers, because that is where meeting audio actually travels. Look in the privacy policy and terms of service under data use and subprocessors.

5. Encryption at rest and in transit

Encryption at rest protects recordings sitting on a server, and encryption in transit protects them while they cross the network. You shouldn't have to choose. You need both because they are equally important. A recording that is encrypted on the server but travels unencrypted can be intercepted in transit, while the reverse leaves it readable to anyone who accesses the storage. It is common practice for AI notetakers to list these two encryption methods separately on security pages, usually AES 256-bit for storage and TLS for transmission.

6. Retention and deletion controls

Deletion you control yourself is the difference between owning your data and asking someone else to remove it, and the retention window decides how long a recording exists to be breached or subpoenaed. Both live in the product settings, documented in the help center under data retention.

7. A bot that joins visibly

The most common way an AI notetaker records a meeting is by joining the call as a bot. A tool that puts user privacy and security first notifies participants before it joins, so everyone in the meeting knows the call is being recorded. For the most current and accurate information on how a bot joins, check the company’s help center. Most AI notetakers document their default joining behavior there.

8. Consent notification

Notification features let you meet an all-party consent standard without relying on someone remembering to say something. Look in the meeting or compliance settings, documented in the help center under recording notifications.

9. Access controls inside your workspace

Access controls decide whether a transcript is readable by the four people who attended or by everyone with a login. They sit in the admin panel, and the plan comparison page usually shows which tier unlocks them.

How Fireflies Addresses AI Notetaker Safety

Meeting the standards above costs a company money and forecloses options, which is why few tools meet all of them. Fireflies documents its answers publicly and shows workspace compliance status inside the product, on a Security Checklist in Team Settings that updates in real time.

Certifications and compliance

Fireflies holds SOC 2 Type II certification, audited independently every year. GDPR compliance covers EU data protection requirements. Both apply on every plan including the free one, which matters because compliance that begins at the paid tier leaves your smallest teams uncovered.

HIPAA compliance is Enterprise only and takes two things together, Private Storage enabled and a signed Business Associate Agreement. Removing either one disables it automatically. FERPA compliance works the same way for student education records, Enterprise only, requiring Private Storage and a signed Data Sharing Agreement.

The infrastructure underneath is audited as part of SOC 2. Meeting content is encrypted with AES at 256 bits while stored and TLS 1.2 or higher while moving. Vulnerability scanning runs continuously alongside a bug bounty program on HackerOne, and system backups cover configuration and metadata while excluding meeting content itself. Full documentation sits on the Fireflies security page, and HIPAA-compliant private storage covers the healthcare setup in detail.

AI training data policy

The training data policy applies on every plan, including the free one. Fireflies.ai does not use customer data to train any AI models. Your personal data is never used to train AI models. Users own their data. We impose a zero-data retention policy for meeting content with our AI vendors.

That last commitment has three parts. AI vendors do not store meeting content after processing, do not retain access to it once the service is delivered, and never use it to train models. Those agreements extend to the transcription providers and model suppliers Fireflies actually uses, including OpenAI and Anthropic. You can check the full list on Fireflies’ Trust Center page. Data received through Google Workspace APIs falls under Google's Workspace developer policy, which bars using it to train or improve any AI model beyond that specific user's own.

Fireflies acts as a data custodian. Internal teams have no access to meeting content by default, and any access requires the user's explicit permission, usually for support.

Visible bot and consent

The Fireflies AI Notetaker joins as a participant everyone on the call can see in the participant list, under a display name attendees can read. It does not join silently.

Which meetings it joins is configurable. Auto-join and recording rules decide which calls get captured, and a host can keep it out of a meeting entirely.

Visibility is not consent, and Fireflies does not treat it as consent. The person who invites Fireflies to a meeting is responsible for telling the other participants and for meeting whatever standard their jurisdiction sets. The notice methods described earlier apply, because a name in a participant list only does the minimum.

User data controls

Users can delete a recording and its transcript at any time, and that deletion is a control inside the product. Visibility is set from the Meeting Notepad, where a meeting can be limited to participants, made private, or shared by link. On Business and Enterprise plans, external sharing can use a password protected link with an expiration period you set.

Business and Enterprise admins manage data retention, workspace access, and user permissions through Groups. Enterprise plans add custom data retention, the Super Admin role for workspace wide governance, and the Rules Engine, which applies recording, transcription, and retention policies across an organization automatically, plus Private Storage for holding meeting data in dedicated infrastructure.

Fireflies holds SOC 2 Type II certification and is GDPR compliant. Fireflies.ai does not use customer data to train any AI models. To review the full security practices before signing up, visit the Fireflies security page. Read the Fireflies Security Overview.

Frequently Asked Questions

Are AI notetakers legal?

AI notetakers are legal in most jurisdictions when participants are informed and the applicable consent standard is met. In one-party consent states, one participant's knowledge is enough. In all-party consent states, everyone has to agree. Calls crossing state or national borders answer to every applicable standard at once, and following the strictest is the lower risk path. Consult counsel for your specific situation.

Can AI notetakers hear confidential conversations?

AI notetakers hear everything said on a call they have joined, including anything confidential that surfaces mid conversation. The controls that matter are the ones deciding which meetings the tool joins at all. Recording rules and auto-join settings can exclude meetings by title keyword, and a host can remove the bot at any point during a call.

Do AI notetakers share your meeting data?

AI notetakers share meeting data in two directions worth checking separately. Inside your organization, sharing defaults decide whether a transcript reaches every participant, every workspace member, or only the people you name. Outside it, meeting audio passes to the transcription and model suppliers the tool uses, and the terms governing those suppliers are set out in the privacy policy.

Do AI notetakers train AI models on your conversations?

Some AI notetakers train on customer data and some contractually prohibit it, so the privacy policy is the only reliable place to check. Fireflies states the commitment directly. Fireflies.ai does not use customer data to train any AI models. Your personal data is never used to train AI models. Users own their data.

Do other meeting participants know when an AI notetaker is recording?

Meeting participants know an AI notetaker is recording when it joins as a visible bot, because it appears in the attendee list under a name they can read. Tools capturing audio through a browser extension or desktop recorder never join the call and may leave no visible trace. Anyone who spots an unfamiliar participant should ask the organizer what it is.

What is the difference between a bot-based and bot-free AI notetaker for privacy?

A bot based AI notetaker joins the call as a visible participant, which gives everyone present notice that recording is happening. A bot free tool captures audio through a browser extension or desktop app without joining, so other participants may see nothing at all. That difference shifts the entire notice obligation onto whoever set the tool up. Transcript extensions in particular leave no trace in the call itself.

What compliance certifications should an AI notetaker have?

An AI notetaker should hold SOC 2 Type II certification, meaning an outside auditor tested its security controls over months, and GDPR compliance if anyone on your calls is in the EU. HIPAA compliance with a signed Business Associate Agreement is required when calls touch protected health information, and FERPA applies to student education records.

What AI notetaker features protect privacy?

AI notetaker features that protect privacy include SOC 2 Type II certification, a contractual ban on training AI models with customer data that extends to subprocessors, encryption at rest and in transit, configurable retention, permanent deletion controls, admin limits on who can open a transcript, and a bot that joins visibly under a readable name.

Can I use an AI notetaker for HIPAA-covered conversations?

You can use an AI notetaker for HIPAA covered conversations only if the company behind it signs a Business Associate Agreement. A HIPAA compliant AI notetaker has that contract in place, and a compliance badge on a website is not one. With Fireflies, HIPAA compliance is available on Enterprise plans and requires both Private Storage and a signed BAA.

Does Fireflies record without consent?

Fireflies does not record silently. The Fireflies AI Notetaker joins as a visible participant everyone on the call can see in the attendee list, and auto-join and recording rules control which meetings it enters at all. Informing participants and meeting the consent standard in their jurisdiction is the responsibility of the person who invited Fireflies.

Is Fireflies GDPR compliant?

Yes. Fireflies is GDPR compliant, and that compliance applies on every plan including the free tier. Fireflies also holds SOC 2 Type II certification, audited independently every year. Organizations with specific GDPR obligations, including data processing agreements and subprocessor disclosures, can review the documentation on the Fireflies security page.

Conclusion

AI notetakers carry real privacy exposure. Consent rules vary by state and by country, meeting content sits on someone else's servers, and not every tool commits to keeping that content out of its training data. The tools worth trusting answer all three with outside certification, written policy, and controls you can operate yourself. Fireflies documents each on its security page.