What Atlas knows about your meetings.
And what it does not. Plain English on what Atlas can see, what it stores, who can access it, and how to take it out of a meeting.
What Atlas can see
During a meeting, Atlas hears the audio of the call. It does not see your screen, your other tabs, or any documents you have not given it. It does not record video.
With your permission, Atlas can read your calendar events to know when meetings are scheduled. It uses narrowly scoped write access only to create a calendar event when you explicitly schedule a meeting from Atlas — it never creates, changes, or deletes events on its own.
What Atlas stores
For each meeting you keep, Atlas stores: the transcript, the summary, decisions, action items, and the artifacts it drew. Audio is processed as a live stream and is not recorded or stored. Transcripts, notes, and boards are stored so you can revisit them.
Email trust@atlasmeeting.com and we delete a meeting and everything attached to it — transcript, summary, notes, and artifacts.
Who has access
Your meetings are visible only to you. The only thing that leaves your account is the recap email, and it goes to the recipients you choose.
We do not read your meeting content. Engineering can access aggregate system logs to diagnose issues, never the content.
How to remove Atlas from a meeting
In the meeting itself: remove Atlas as a participant, the same way you would remove any other guest. You can also pull Atlas out of a live call from that meeting’s page in your dashboard.
Before the meeting: turn off auto-join for any calendar event from your dashboard.
Encryption & infrastructure
Data is encrypted in transit and at rest. Atlas runs on AWS in US-East.
We use Anthropic’s Claude models through Amazon Bedrock in Atlas’s AWS environment or Anthropic’s enterprise API for summary and artifact generation, depending on the Atlas feature and runtime configuration. AWS does not train on Bedrock inputs, and Anthropic does not train its models on Atlas API data.
Subprocessors
The third-party services Atlas relies on to run: AWS (compute, storage, and Claude via Amazon Bedrock), Anthropic (Claude models through Bedrock or Anthropic’s enterprise API), Google (Gemini voice models), Deepgram (speech-to-text), Clerk (authentication), Vercel (frontend hosting), Resend (recap email delivery), Exa (web search when the agent researches).
A full list with locations and data categories is available on request: trust@atlasmeeting.com.