Introduction

Tl;dv meetings exposed a hard lesson in cloud security: the AI meeting recorder let any free user query 181,874 meeting records from its Firestore database, because the meetings collection had no tenant isolation. The researcher who found the flaw disclosed it on January 28, 2026. It is still unpatched, and the story finally reached a wide audience on August 10 when it hit the front page of Hacker News.

Tl;dv drops a bot into Google Meet, Zoom, or Teams calls, records them, transcribes them, and writes AI summaries. It claims over two million users. Those recordings include sales negotiations, job interviews, performance reviews, and internal strategy sessions.

What Happened

Security researcher Bob Da Hacker found the issue in late January 2026. Signing up for the platform is free, and the token exchange that every user performs gives access to a Firestore database where all meetings, from every tenant, live in one collection. He reported it through the company's own contact and emailed the CTO directly. Six months of follow-ups produced vague reassurances and then silence. As of late July 2026 the database was still wide open.

The writeup went up on the researcher's blog on August 4. On August 10 it gathered more than 590 upvotes on Hacker News, pushing the exposure into the spotlight.

How the Exposure Works

The flow is simple. When you sign up, tl;dv authenticates you with a JWT and exchanges it for a Firebase token through gw.tldv.io/v1/users/firebase/token. That token lets you query Firestore at projects/lmi-store/databases/(default).

The meetings collection has no tenant isolation. Any authenticated user, including a free-tier account, can list every meeting across every organization on the platform. Each record includes the creator's email address, the conference ID, which is a joinable Google Meet or Teams room, the provider, the recording status, and timestamps.

Firestore security rules exist precisely for this. A minimal tenant-scoped rule set locks documents to their owners:

match /databases/{database}/documents { match /meetings/{meeting} { allow read, write: if request.auth != null && request.auth.uid == resource.data.ownerId; } }

The rest of tl;dv's collections (users, chats, transcripts, recordings, teams) already return 403 for cross-tenant queries. Only meetings was left open.

How Bad It Is

The researcher enumerated 181,874 meeting records belonging to 84,312 unique users across 35,003 email domains. The impact goes far beyond metadata, because the metadata includes live call access:

• Around 1,000 meetings are in recording status at any given time, and their conference IDs are joinable rooms. An attacker with a bot could walk into live calls uninvited.
• Over 1,000 meetings were public, meaning video and transcripts were watchable, exposing 715 invitee emails across 228 domains.
• Government meetings from 23 countries appear in the data, including .gov domains from Brazil, Ukraine, the United States, Japan, Israel, and Belize.
• University meetings show up from Berkeley, the University of Tokyo, and other .edu and .ac domains.
• Corporate customers include Mitsui-Soko, with 484 meetings across four regional offices, plus HubSpot, Confluent, and others.

The researcher demonstrated the live call risk by joining a Google Meet run by the Malaysian Ministry of Education, where a presenter was speaking to 157 participants, and a call where university students were screen-sharing a startup project.

The Disclosure That Went Nowhere

Tl;dv's security page lists SOC 2, GDPR, and EU AI Act compliance badges and promises a security team response within 24 hours. In practice, the CTO never replied to any of the emails sent over six months. The researcher also found a separate problem: an internal World Cup prediction app at worldcup.tldv.io exposes a Player API with no authentication, leaking the names and corporate emails of 43 players, including 19 tl;dv employees.

The disclosure timeline:

Date	Event
Late January 2026	Tenant isolation bug found and reported
January 28	First contact with the company
February to March	Follow-ups, no fix
July 2026	Still unfixed, still no response

What This Means for Developers

Three practical takeaways apply beyond this one vendor.

First, treat Firestore security rules as code. Every collection needs default-deny rules reviewed in the same PR that ships the schema. One forgotten collection is all it takes, and "we fixed it for everything except meetings" is a real failure mode, not a hypothetical.

Second, vendor risk includes data your company records. If a sales or product team uses an AI meeting recorder, that vendor holds client names, deal terms, and interview feedback. Check whether the platform does tenant isolation and whether it has a history of responding to researchers.

Third, the exposure window matters. This data was queryable for at least seven months. Assume that anything recorded on such a platform is potentially visible to outside parties until the vendor proves otherwise.

What Users Should Do

If your organization uses tl;dv:

• Stop recording sensitive calls until the vendor confirms the collection is locked down.
• Ask the vendor, in writing, for a status update and for confirmation that your meeting records were not accessed.
• Watch for targeted phishing, since leaked emails and call metadata make convincing pretexts easy.
• Review who inside your company can access recordings, and rotate any links or shared artifacts that came from public meetings.

Frequently Asked Questions

Is my tl;dv meeting data exposed?

If you used tl;dv since at least January 2026, your meeting metadata was queryable by any authenticated user. A subset of meetings were fully public, including video and transcripts.

Can attackers join live calls?

Yes. The leaked conference IDs are real joinable rooms, and roughly 1,000 meetings were actively recording at any moment.

Has tl;dv fixed the problem?

Not as of late July 2026. The researcher's August writeup says the Firestore collection remained open and the company had not responded.

What is tenant isolation?

Tenant isolation means one customer's data cannot be read or modified by another customer. It is enforced with per-document security rules, row-level policies, or separate databases.

Should I delete my tl;dv account?

That is a risk decision for each team. At minimum, stop feeding it sensitive conversations and ask for data deletion if you no longer need the recordings.

Key Takeaways

• A missing Firestore rule set exposed 181,874 meeting records across 84,312 users, with no tenant isolation.
• Live conference IDs were visible, letting anyone join roughly 1,000 active recordings at any time.
• The issue was reported in January 2026 and remained unfixed for over six months with no vendor response.
• Securing Firestore means default-deny rules on every collection, reviewed like application code.
• Vendor risk reviews should cover AI services that record and store sensitive audio and video.

Conclusion

The tl;dv case is not an exotic exploit. It is a configurable cloud database with one missing rule, in a product that records people's most sensitive conversations. The fix is straightforward, which makes the six-month silence the real story. If your team uses an AI meeting recorder, audit its security posture now, before a researcher like this one finds your calls in someone else's query.

---

Sources: Bob Da Hacker disclosure, Hacker News discussion, tl;dv