AI Notetaker tl;dv: Flaw Left 181,000+ Meetings Exposed
A researcher publishes his write-up of a severe tl;dv flaw: 181,874 meeting records from 80,000+ users were queryable — including government calls.
Illustration · AI-generated (AI IN LIFE)
At a glance
- 181,874 meeting records from 80,000+ users were queryable
- Root cause: missing tenant isolation in the Cloud Firestore database
- 1,000+ meetings had publicly exposed transcripts and invitee emails
- .gov meetings from 23 countries were among those affected
- Discovered late January 2026; per Dark Reading, the flaw stayed open for months
Security researcher “BobDaHacker” has published his full analysis of a severe vulnerability in the AI meeting notetaker tl;dv — and the details are ugly: 181,874 meeting records from more than 80,000 users could be queried through the service's Google Firebase backend, including timestamps, recording status and organizers' email addresses.
The root cause is mundane: missing tenant isolation in the “meetings” collection of the Cloud Firestore database. Any authenticated user could read other users' records. More than 1,000 meetings had publicly exposed transcripts and invitee emails; in tests, the researcher managed to join live private calls in roughly 80 percent of attempts.
Particularly sensitive: affected meetings included .gov addresses from 23 countries — government and agency communication. Dark Reading reported in early August that the flaw had remained open despite repeated disclosure attempts since its discovery in late January.
The case joins a growing series of security failures in AI productivity tools that sit deep inside calendars, conferences and internal communication — while backend configuration fails to keep pace with growth. The fix would have been a few lines of Firestore security rules.
If you deploy AI notetakers at work, you are inviting a third party into every confidential meeting. Before rollout, data processing agreements, data residency and the question “who can technically read these transcripts?” belong on the table — under GDPR, a leak like this is a reportable incident.
FAQ
What exactly was exposed?
Meeting metadata such as timestamps, recording status and organizer emails; 1,000+ meetings also exposed transcripts, and joining live calls was often possible.
How did this happen?
The Firestore database lacked tenant isolation — any logged-in user could query other users' records. A few lines of security rules would have prevented it.
What should companies do now?
Check whether tl;dv is in use, involve privacy teams, assess affected meetings — and security-review any AI notetaker before rollout.
