
AI meeting notetakers: privacy and confidentiality risks
AI notetakers join calls, record everything, and send summaries to everyone on the invite. They save time, but they also create a new copy of every confidential conversation, often in a service nobody approved. Here is how to think about consent, storage, privilege and controls.
Reviewed by the ITRS security team · Last updated
How AI notetakers work
There are two broad kinds. Built-in features, such as transcription, recording and AI summaries inside Microsoft Teams, Zoom or Google Meet, run inside a platform your organization already licenses and administers. Third-party notetakers are separate services that a user connects to their calendar; a bot then joins meetings as a participant, records audio, produces a transcript and summary, and often emails it to attendees or stores it in the user's account with the vendor.
The second kind is where most of the risk sits. A single employee can connect one with a few clicks, sometimes granting it access to their whole calendar, and from then on it may join every meeting they are invited to, including meetings with clients who never agreed to it.
Consent and notice
Recording a conversation collects personal information: voices, names, opinions and whatever is discussed. Under PIPEDA, organizations need meaningful consent for the collection, use and disclosure of personal information, subject to limited exceptions. The Office of the Privacy Commissioner of Canada's guidelines for obtaining meaningful consent set out seven guiding principles, including emphasizing key elements, giving people clear options to say yes or no, and making consent a dynamic and ongoing process.
For meetings, that translates into practical habits:
- Say at the start that the meeting will be recorded or transcribed, and by what tool
- Put it in the invitation for external meetings, so people can object in advance
- Have a real way to opt out: turn the bot off, or the person leaves the transcript
- Do not let notetakers join meetings automatically where the organizer has not chosen to use one
- Apply stricter rules for meetings involving health, financial or HR matters
Where transcripts are stored and for how long
Ask three questions of any notetaker: where is the recording and transcript stored, who can access it, and when is it deleted? For Microsoft Teams, Microsoft's documentation states that transcripts are stored together with meeting recordings in OneDrive and SharePoint, which means your existing retention policies, holds and access controls can apply to them.
Third-party services typically store data in their own cloud, sometimes outside Canada, under their own retention terms, and some reserve the right to use customer content to improve their models. That copy sits outside your retention schedule and legal holds and, if the employee leaves, may stay in an account you do not control. The Canadian Centre for Cyber Security's generative AI guidance (ITSAP.00.041) advises selecting vendors carefully and being careful what information you provide to AI tools; a notetaker is that guidance applied to every word spoken in a meeting.
| Question | Built-in (licensed platform) | Third-party bot |
|---|---|---|
Where stored | Your tenant (e.g. OneDrive/SharePoint for Teams) | Vendor's cloud, location varies |
Retention | Your policies can apply | Vendor's terms or user settings |
Legal hold | Can be included | Usually outside your holds |
Access on staff departure | Managed with the account | May remain in a personal vendor account |
Use for model training | Check your licence terms | Check vendor terms; sometimes on by default |
Confidential and privileged meetings
For lawyers, and for anyone discussing legal advice with them, a third-party transcript is a disclosure of the conversation to another service. Whether that affects privilege or confidentiality obligations is a legal question, but the cautious position is simple: do not use third-party notetakers in privileged meetings, and think carefully before using any transcription in them.
The same caution applies to board meetings, HR investigations and disciplinary meetings, settlement discussions, and meetings involving health information. A transcript is also a record that can be requested in litigation or through an access request, and AI summaries can misstate what was said in ways a participant may later have to explain.
Setting organizational rules
Approve tools
Name the transcription and summary tools staff may use, normally the ones built into platforms you already administer.
Ban unapproved bots
No third-party notetakers connected to company calendars or accounts without approval.
Define no-record meetings
Privileged, HR, board and health-related meetings are not transcribed unless specifically authorized.
Announce and allow opt-out
Notice at the start of every recorded meeting; external participants told in advance.
Retain deliberately
Transcripts follow your retention schedule; delete those you do not need.
Handle external bots
Tell staff they may ask a client's or vendor's notetaker to leave, and should for sensitive topics.
Admin controls
Policy needs technical backing. In the Teams admin center, Microsoft documents transcription as a meeting policy setting, under Meetings > Meeting policies, which can be turned on or off for everyone or for specific users. Recording, Copilot and meeting lobby settings are also managed through meeting policies.
To stop third-party bots, look at how they get in. Most connect through an app consent in Microsoft Entra ID or Google Workspace; restricting user consent to third-party apps (in Entra, under Identity > Applications > Enterprise apps > Consent and permissions > User consent settings) means staff cannot connect a notetaker to their calendar without admin approval. Review the apps that already have consent and remove the ones you have not approved. Meeting lobby settings can keep unexpected participants, including bots, out of meetings until admitted.
We help organizations set notetaker policies, configure Teams and Google Meet to match, and find the bots already connected.
Related reading.
- GuideAI acceptable use policy: what to includeStaff are already using generative AI tools. A short, clear acceptable use policy tells them which tools are approved, what information must never go in, who is…
- GuideFinding the AI tools your staff already useBefore you can write a sensible AI policy, you need to know what is already in use.
- GuideControlling OAuth app consent in Microsoft 365Consent phishing doesn't steal a password. It tricks a user into clicking 'Accept' on a real Microsoft permission prompt for an attacker's app, which then reads mail and…
- ServiceAI governance and responsible-use policyYour staff are already using generative AI. Governance decides which tools are allowed, what data can go into them, who approves new uses and how leadership keeps…
- PracticePrivacy, Data Governance & AIPIPEDA and Quebec Law 25 readiness, data mapping, breach notification and AI governance.
Not sure what your obligations require of your systems?
Speak with our privacy practice. We are staffed 24/7. Active incident? Call (289) 582-9930, 24/7.
