tl;dv AI Meeting Recorder Exposes 181,000+ Records via Unsecured Database — Unpatched for 6+ Months
Summary
- • tl;dv's Firestore database has no tenant isolation — any authenticated user can read all 181,874 meeting records across all accounts
- • Exposed data includes live joinable conference IDs for ~1,000 active calls at any moment, enabling silent meeting infiltration
- • Government meetings from 23 countries, major universities, and 35,000+ corporate domains are all affected by the flaw
- • Researcher reported the vulnerability January 28, 2026; six months later the database remains open and the CTO has not responded
Details
Firestore meetings collection has no per-tenant access control
Any authenticated tl;dv user can query the entire Firestore meetings collection across all accounts — a fundamental access control failure with no row-level or tenant-level restriction.
Firebase JWT exchange grants full cross-account Firestore read access
A Firebase JWT is exchanged at gw.tldv.io/v1/users/firebase/token for a token granting access to projects/lmi-store/databases/(default); the meetings collection has no row-level security or tenant filtering, making all records readable.
181,874 meeting records exposed across 84,312 users and 35,003 domains
The Firestore collection holds 181,874 meeting records belonging to 84,312 unique users spanning 35,003 email domains worldwide.
Approximately 1,000 live meetings have joinable conference IDs exposed at any time
Meetings with status 'recording' expose a real-time joinable conference ID. At any moment roughly 1,000 live meetings have their active call links visible in the collection.
Researcher successfully joined two live uninvited meetings as proof of concept
Researcher joined a Malaysian Ministry of Education presentation (157 participants) and a US university startup session (21 participants) by grabbing live conference IDs from Firestore — neither call had invited them.
Government meeting metadata exposed from 23 countries
Affected governments include the US, Brazil, Colombia, Malaysia, Ukraine, Japan, Qatar, Philippines, South Africa, Israel, and 13 others — all storing sensitive ministerial and agency calls on the platform.
Universities and corporations worldwide among the 35,003 affected domains
University domains include Berkeley, University of Tokyo, De La Salle, and Universidad Nacional de Colombia. Corporate users include HubSpot, Confluent, and Mitsui-Soko (484 meetings across four offices).
Peak month saw 43,209 meetings; busiest slot is Wednesdays 2pm UTC
July 2025 was the busiest month with 43,209 meetings recorded; the peak time slot is Wednesdays at 2pm UTC with 7,804 concurrent meetings in the exposed collection.
tl;dv has 2M+ users recording sensitive calls across sales, HR, and strategy
tl;dv is a VC-backed AI meeting recorder with over 2 million users that captures sales calls, job interviews, performance reviews, and internal strategy sessions.
Vulnerability reported January 2026 — database still open six months later with no vendor response
The researcher disclosed the flaw responsibly on January 28, 2026. As of July 2026, six months later, the Firestore database remains wide open and the CTO has not responded to any communications.
Each row is a distinct security finding, technical detail, or impact metric from the original researcher disclosure.
What This Means
This is one of the most serious AI SaaS security failures on record: a fundamental database architecture defect in tl;dv has exposed the meeting metadata and real-time call links of over 84,000 users across government ministries, universities, and global corporations for more than six months — with no fix and no vendor response. The flaw goes beyond a data leak: the ability for any free-tier user to silently join any live meeting on the platform transforms a metadata exposure into an active, real-time surveillance capability. As AI meeting recorders become standard infrastructure for sensitive business and government conversations, this incident is a critical reminder that proper tenant isolation and rapid vulnerability response are non-negotiable. Organizations using tl;dv should immediately audit their usage and consider suspending the service until the vulnerability is confirmed patched.
Sentiment
Concerned and critical, with emphasis on vendor inaction and risks of AI SaaS tools
“hacker pulled data from 181,874 online meetings... 23 governments were hit... all because of 1 missing Firestore rule... this vulnerability was open from January 28 to August 4... suggests using local model quill”
“tl;dv (Too Lazy; Didn't Validate): 181,874 Meetings Left Wide Open - I reported this on January 28th, 2026. It is now July 2026. Six months later. The Firestore database is still wide open. The CTO never responded”
notably referencing their own prior responsible disclosure
“AI meeting assistant tl;dv left 181,874 meeting records exposed, including live conference IDs. The vulnerability was reported 6 months ago but ignored. A strong case for running your own open-source meeting tools.”
“A Google Firebase misconfiguration in tl;dv, an AI meeting tool, let attackers query any user's meeting information and potentially join calls. Government agencies and corporations were exposed. AI-powered SaaS tools create new data exposure vectors that traditional vendor risk assessments do not cover.”
Split
Strong criticism of tl;dv's six-month inaction on responsible disclosure (~80% of reactions); secondary push toward self-hosted/open-source alternatives vs. trusting AI SaaS (~20%).