Threat · curated 4 Aug 2026
AI Notetaker Lets Hackers Spy on Government, Corporate Video Calls
First reported darkreading.com
Coverage timeline
Single-source incident — first reported, latest, and curated coincide.
Why it matters
tl;dv's autonomous notetaking agent joins, records, and transcribes video calls by default, so a broken access-control flaw exposes sensitive government and corporate conversations to eavesdroppers at scale.
A Google Firebase misconfiguration in tl;dv, an AI meeting notetaker used by over two million users including government agencies and firms like Salesforce and Cloudflare, lets any user query other users' meeting information and join their calls. Researcher BobDaHacker demonstrated the flaw in late January by accessing the back-end environment and joining calls hosted by government agencies and large organizations.
Summary
The AI meeting assistant tl;dv, which claims more than two million users and is used across dozens of government agencies, universities, and major corporations, contains a Google Firebase misconfiguration that lets any authenticated user query other users' meeting information from its Cloud Firestore database.[0]
Security researcher BobDaHacker demonstrated that the exposed 'meetings' collection lacks tenant isolation, allowing enumeration of live calls, metadata, and creator email addresses, and that impersonating an AI notetaker enabled joining private calls roughly 80% of the time. The back end exposed over 180,000 call records from more than 80,000 users, and over 1,000 sampled meetings leaked invitee emails and transcripts publicly.[0]
BobDaHacker attempted to report the flaw to tl;dv without success and then notified Dark Reading; the issue was still live as of the August 4, 2026 publication date and tl;dv did not respond to press inquiries.[0]
Attack chain
- Authenticated access: A user signs up or signs in to tl;dv and is assigned a session ID in the Firebase back end, which grants query access to the Cloud Firestore database.[0]
- Data enumeration: Exploiting missing isolation on the 'meetings' collection, the user queries every live conference call into which tl;dv is invited, obtaining metadata such as timestamps, recording status, and the creator's email address.[0]
- Meeting intrusion: Using the harvested meeting information, the researcher impersonated an AI notetaker and requested to join private calls, succeeding roughly 80% of the time; some calls were left open to the public.[0]
- Bulk data scraping: The researcher sampled more than 27,000 meeting IDs and scraped them, finding over 1,000 that exposed invitees' emails and call transcripts on the open internet.[0]
Disclosure timeline
| Date | Event |
|---|---|
| Late January | BobDaHacker discovered that any tl;dv user can access the Firebase back end and query other users' meeting information.[0] |
| After discovery | The researcher attempted to report the issue to tl;dv without success, then notified Dark Reading.[0] |
| August 4, 2026 | Dark Reading published the report; tl;dv did not reply to press and marketing contacts and the issue remained live.[0] |
How it works
When a user signs up or signs in, tl;dv assigns a session ID in its Firebase back end that permits querying the app's Cloud Firestore database. Tenant isolation generally prevents users from seeing other users' transcripts, recordings, and chats, but the 'meetings' collection lacks that isolation, so any authenticated user can query every live conference call into which tl;dv is invited and retrieve metadata (timestamps, recording status) and the creator's email address.[0]
Per the researcher, Firestore security rules are the first configuration Google recommends and the default rules warn that they are open and must be locked down. The fix is a few lines of security rules that scope reads to the authenticated user's organization.[0]
A separate internal tl;dv World Cup bracket game ('Too Long; Didn't Score') has an unauthenticated player-data API endpoint where a simple GET request returns all 42 participating employees, including personal email addresses, usernames, and the game admin identity.[0]
Affected versions and patch status
| Product | Affected | Patch status |
|---|---|---|
| tl;dv AI meeting assistant (Google Firebase / Cloud Firestore back end) | Live production environment; 'meetings' Firestore collection lacking tenant isolation; exposed 180,000+ call records across 80,000+ users | Unpatched — issue reported as still live as of the August 4, 2026 publication date; vendor did not respond to inquiries[0] |
| tl;dv internal 'Too Long; Didn't Score' World Cup bracket web app | Unauthenticated player-data API endpoint exposing 42 employees' data | Unpatched — game remained live weeks after the World Cup concluded[0] |
Key takeaways
- A single missing set of Firestore security rules on one collection was enough to expose live meeting data for 80,000+ users and enable an attacker to join private government and corporate calls about 80% of the time.[0]
- AI notetakers hold deep access to communication layers yet receive little security scrutiny, and users often forget they are running silent, high-permission participants in meetings.[0]
- User-side privacy defaults materially reduced exposure — only ~1,000 of ~70,000 meetings leaked content — underscoring that default-private settings are a meaningful control while the underlying platform flaw remains unpatched.[0]
Defensive actions
- Treat AI notetakers as active participants and scrutinize unexpected bots in a call's participant list.: The researcher joined a Malaysian government meeting where the tl;dv bot appeared in the participant list and 157 people did not question it; an unexpected notetaker is a red flag for eavesdropping.[0]
- Configure meeting privacy settings to default-private, proportionate to content sensitivity.: Of roughly 70,000 meetings checked, only about 1,000 had public sharing enabled and those exposed transcripts and invitee emails; the other 69,000 were protected from content exposure by privacy settings alone.[0]
- For developers using Firebase, scope Firestore security rules to the authenticated user's organization/tenant for every collection.: The exposure resulted from missing isolation on the 'meetings' collection; a few lines of properly scoped security rules would remediate it, and Firestore's default rules warn they are open.[0]