AI meeting minutes and LGPD: recording meetings legally in Brazil

Search "can I record a meeting without telling anyone" in Brazil and you get law firms — none of them explain how to turn the recording into usable minutes. This guide joins both ends: what the LGPD expects from anyone recording work meetings, a copy-paste disclosure script, and the practical record → transcribe → minutes-with-citations flow, with no bot joining the call.

The LGPD (Law 13,709/2018) has no article dedicated to "meeting recordings" — but participants' voices, names, and opinions are personal data, and recording plus transcribing is data processing. That requires a lawful basis from Article 7: the two most cited in this context are consent (item I) and legitimate interest (item IX), which demands a documented balancing between your interest and the data subjects' rights. Brazilian courts have accepted recordings made by one of the parties to the conversation in specific contexts, but that is case law about evidence — not a free pass to record colleagues and clients without transparency. This is not legal advice: for sensitive situations (HR, health, minors' data), consult a lawyer.

Even where legitimate interest might carry the recording, announcing it remains the practice that settles everything: it satisfies the transparency duty of Article 9, avoids the surprise that turns into an ANPD complaint, and preserves trust with clients and candidates. Rule of thumb: if you would hesitate to say "I am recording this", do not record.

Say the disclosure at the start of the call, with the recording already running — that way the agreement is captured in the transcript itself. A template that covers purpose, tool, and the right to object:

Adapt it to the context: for external meetings, put the notice in the calendar invite as well; in job interviews, wait for an explicit "yes" before continuing; for sensitive topics, prefer consent over legitimate interest.

Yes — the transcript and the minutes still contain personal data, so the obligations follow the file, not just the audio. When an AI service processes the recording on your behalf, it acts as an operator (processor), and the data-processing agreement is the instrument that formalises that relationship. In practice, the checklist your DPO will ask for:

• A lawful basis, chosen and recorded (consent captured in the transcript, or a documented legitimate-interest balancing)
• A clear notice before recording — who records, why, and with which tool
• A DPA with the AI vendor, plus the guarantee that your data never trains models
• Minimisation: record what you need, set retention, delete what you no longer need
• Controlled access: who can see the minutes and the transcript (per-note and per-folder permissions)

With the legal side handled, the practical flow is short — and this is where AI pays for itself: minutes stop being a 40-minute chore after the meeting and become an automatic by-product of it.

1. Record locally, without a bot: Reline captures your microphone and system audio directly on your computer (macOS, Windows; Linux in beta) — no extra participant joins the call, on any meeting platform.
2. Transcribe with speaker separation: the transcript arrives in real time, separated into Me vs Other, in 60+ languages — including meetings that switch between Portuguese and English.
3. Generate the minutes with citations: the AI turns the transcript into a structured summary — decisions, actions, deadlines — and every claim links back to the exact transcript segment, so you can verify before distributing.

What good minutes must contain (and what the AI fills in by itself from the transcript):

• Date, participants, and the meeting's purpose
• Decisions taken — with who decided and the segment that proves it
• Agreed actions, each with an owner and a deadline
• Open items and next steps
• A link to the full transcript, with controlled access

The tool choice changes the shape of the data processing. A recorder bot is a visible third participant: it joins the call, shows up in the participant list, usually requires calendar access, and transcribes from the first second. Bot-free local capture removes that third party from the room — which reduces the exposure surface, but does not replace your duty to disclose.

The honest reading of the table: bot-free is the architecture with fewer exposed parts — not a legal shortcut. Transparency and lawful-basis obligations apply equally to both models; what changes is how many extra eyes and systems take part in the processing.

Reline was designed for exactly this flow: bot-free local capture, a speaker-separated transcript (Me vs Other), and AI summaries where every claim cites the exact transcript segment — all inside a Notion-style collaborative notepad with five permission levels per note and folder. The interface is fully available in Portuguese, and it runs on the web, macOS, and Windows (Linux in beta). The honest caveat: only capture is local — transcription, AI, and storage run in our cloud under a DPA, and your meetings are never used to train models.

On cost: Reline starts free — 1 hour per note, 10 hours of transcription per month, unlimited notes, no card required. Professional is $15/user/month ($12 annual) and removes the recording cap. Enterprise — SSO, audit log, DPA, SLA — is $32/user/month ($26 annual).