AI Meeting Assistant Privacy: What Happens to Your Audio
Follow audio, transcripts and shared results separately. See what stays on your device, what goes to AI or email providers, and what to ask before a meeting.
On this page
An AI meeting assistant can keep audio on your device while still sending the transcript to an online AI provider. It can also capture without a bot and process everything in the cloud. To understand privacy, follow the audio, the text and the shared result separately.
Meetings Brief processing choice: The device-only instructions below require On this device. Live notes also offer Online with Groq when available: selected audio leaves your device in short batches after your processing choice, with shared limits. Online is available for existing recordings: audio goes through private Supabase upload storage to the provider named before upload (Cloudflare Workers AI or Groq), subject to shared usage limits. Check permissions and privacy; AI briefs separately send transcript text to Anthropic.
This guide explains those choices and the actual Meetings Brief workflows. It is not a certification that a tool is suitable for a confidential, regulated or legally sensitive meeting. Use the recording-consent guide for questions to resolve before capture, and the privacy policy for our service's specific disclosures.
Four questions before you press Start
| Question | What to establish |
|---|---|
| What can the tool hear? | The selected tab, microphone, system audio or an uploaded file |
| Where is speech turned into text? | On the device, or at a named online speech provider |
| What happens to the transcript? | Local storage, AI processing, server retention and who can reopen it |
| Who receives the result? | You, invited participants, a shared workspace or anyone with a link |
An answer to one row does not answer the others. In particular, no bot, no signup and no audio upload describe different properties.
What happens in Meetings Brief
Live meeting transcription
On this device: At live transcription, the selected meeting audio is recognized on your device. Where Chrome provides the chosen on-device language, it uses that language pack; otherwise the website uses a downloaded Whisper model. Those speech routes do not upload the audio.
The shared meeting tab supplies the other participants' sound. Allow your microphone to include your own voice. In Chrome's on-device mode, those inputs can be marked You and Others; that is not individual speaker identification or a claim to identify people by voice.
Finishing a live meeting can send text automatically. A completed transcript of at least 150 words starts a brief request, which sends transcript text to Anthropic through our service. A confirmed email and automatic email enabled allow the brief and transcript attachment to be sent through Resend. Turning off automatic email does not turn off the on-page AI brief.
Short or incomplete captures wait for a deliberate review and brief request. If recognition stops with draft words, they are saved in a marked Unconfirmed words section. Review it rather than treating provisional text as a quotation. Daily usage limits and service errors can prevent AI generation or delivery.
Transcribing an existing recording
On this device: At recording transcription, the file is read and processed on your device. Finishing transcription alone does not upload its text or request an AI brief. Choosing Get the brief passes the transcript into the live-results workflow for online AI processing.
If your requirement is to transcribe without sending meeting content to an AI service, stay with the file-transcription result and its text/subtitle downloads. An account-free AI summary is still an online request.
Notebook dictation, Aria and saved audio clips
The notebook app's microphone is not the same feature as the live-transcription tool. Notebook dictation uses the browser's speech service; Aria uses on-device recognition when available and otherwise can use that service too. In the online route, the browser maker receives audio even though our servers do not. The Web Speech API documentation distinguishes online recognition from on-device language packs.
Notebook AI actions also send note text to Anthropic. Some run automatically, including action-item extraction when a Home-screen note is saved and summaries as you work on meeting notes. Audio clips deliberately recorded in a notebook meeting's Audio tab are stored locally in the browser. See the policy before treating any of these features as a purely offline notebook.
The Chrome extension
The extension page shows its publication status. Its audio recognition uses Chrome's on-device route; sending its completed transcript to the website can lead to an AI brief and email under the conditions above.
New extension meetings keep a protected local copy using a key managed by the browser profile, without asking you to invent a password or recovery key. This protects stored copies, not text on an unlocked screen or a compromised device. It does not sync or provide a backup. Older protected meetings may still need their original recovery key.
Where saved results and shared copies go
Live transcripts and recording results are saved in the browser you used. They do not automatically appear on another device, and clearing browser data can remove them. Keep a download or delivered email if you need a durable copy.
Some information also reaches our service: for example, brief-delivery records and certain notebook action items or Aria conversations. A finished brief or long text-tool result is held temporarily in private server storage so the page can retrieve it after an interruption. Confirmation that the browser saved it triggers deletion; a cleanup process handles older unconfirmed results. Outages can delay cleanup. The policy gives the retention details and exceptions; local transcription does not mean the entire service has no server storage.
A share link carries its brief in the part of the URL after the # sign. Anyone with the full link can read it. Treat it as a copy you are handing over: a forwarded message, screenshot or saved result is outside the original browser's controls. Sending through Gmail, Outlook, WhatsApp or your email app also involves that service.
Bot-free capture does not decide privacy
A visible bot is one way to receive meeting sound. A desktop app, browser extension or screen-sharing picker can receive sound without adding a participant. Any of those can be followed by online processing.
For example, Otter documents a desktop capture route, while Fireflies documents a Google Meet extension. Check the privacy terms for the actual route and plan you will use. A bot ban is not permission to use an unapproved alternative.
On our website, capturing another browser or installed app depends on whether your browser and operating system offer sound for the chosen window or screen. Screen-sharing permission does not guarantee an audio track. Choose the narrowest source that works, and avoid including unrelated audio in the captured surface.
Explain the full workflow to participants
Before using the live tool, a plain-language explanation might be:
For On this device sessions only. If you later transcribe a recording Online, its notice must name the processor shown before upload (Cloudflare Workers AI or Groq) and explain that audio leaves the device:
“I'd like to transcribe this meeting on my computer. The audio stays here, but the text will go to Anthropic to create a summary. If I email it to myself, Resend will deliver the summary and transcript. I plan to share it with [intended recipients]. Is that acceptable?”
Replace the recipients and workflow with what you actually intend to do. This is an explanation to discuss, not wording that by itself establishes valid consent in every jurisdiction. Resolve applicable rules and organizational requirements first, offer an alternative if someone objects, and stop if you do not have the required permission.
For recording-only transcription without an AI brief, say so instead. Do not describe a live meeting as “nothing leaves my laptop” when its text will be sent to create the result.
Questions to put to any provider
Ask for answers tied to the feature and plan you are considering:
- Which company processes audio, which processes text, and where?
- How long are recordings, transcripts, failed jobs and backups kept?
- Does deleting a meeting remove each of those copies?
- Who can access meeting content, including support staff and other workspace members?
- Is content used for training, and what account settings or contract terms control that?
- Does speaker labeling use stored voice characteristics, or another method?
- Who gets automatic recaps, and can the organizer change that before the meeting?
- What export, deletion and incident-support options are available?
Do not infer a provider's answers from its price, a no-signup start or an attractive privacy badge. Match written terms and controls to your meeting's needs.
Frequently asked questions
Does Meetings Brief upload my meeting audio?
The live-transcription and file-transcription tools process audio on your device. Notebook dictation and Aria have different speech routes and can send audio to the browser maker's online service. AI summaries send text to Anthropic, and email delivery uses Resend.
Does disabling automatic email keep the live transcript offline?
No. After a completed live transcript of at least 150 words, the on-page brief still sends text for AI processing. File transcription does not send the transcript for a brief unless you choose Get the brief.
Is bot-free transcription more private?
The absence of a bot does not establish where audio is processed, who receives text or how long results are kept. Check the actual capture, processing and sharing settings.
Does on-device transcription remove the need for permission?
No. It changes where speech is processed. Recording and transcription still need the permissions and approvals applicable to that conversation.
Are meeting notes always encrypted on my device?
No. New extension meetings use browser-managed protection; website-only meetings, recording transcripts and notebook notes are not covered by that extension encryption. Downloads, shared links and emailed copies have their own handling. The privacy policy describes the boundaries.
Does a personal Gmail address prevent native Google Meet transcription?
No. Eligibility depends on the host subscription: Google's supported editions include Workspace Individual as well as listed business and education editions. Free personal hosting alone does not include native transcripts.
For a meeting happening now, open live transcription. For a file whose text should stay local, transcribe the recording and download the result without requesting an AI brief.