A security researcher says a missing access-control rule in the database behind tl;dv, an AI notetaker used with Zoom, Google Meet and Microsoft Teams, left more than 181,000 meeting records open to any signed-in user of the service for months — and in many cases let strangers slip into meetings while they were still underway.
The researcher, who publishes under the handle bobdahacker, disclosed the flaw publicly this month after describing repeated attempts to warn the company beginning in late January. The exposed dataset covered 181,874 meeting records tied to 84,312 unique users across 35,003 email domains, including government agencies in 23 countries and university accounts at institutions such as Berkeley and the University of Tokyo, according to the write-up. Dark Reading and other outlets have since covered the disclosure.
How the Flaw Worked
tl;dv stores meeting metadata in Google's Firestore database and issues authenticated users a token to query it. According to the researcher, nearly every collection in that database — users, chats, transcripts, clips, recordings, videos, notes, teams and organizations — carried a security rule restricting each account to its own data. The meetings collection did not. Any authenticated user could list meeting documents belonging to any other tenant on the platform, each one showing the creator's email address, the meeting provider, its recording status and, critically, the video-conference ID.
For meetings still in a recording state, that ID pointed to a live, joinable Google Meet or Teams room. The researcher reported that simply requesting admission while presenting as tl;dv's own notetaker bot succeeded in roughly 80 percent of tested cases — hosts see a generic automated participant, assume a colleague invited it, and let it in without checking which account was behind it.
The write-up describes joining a livestreamed session hosted by an institute tied to Malaysia's Ministry of Education with 157 participants present, and a university-linked meeting of about 21 people where attendees were discussing product development work. The researcher said roughly 1,000 recording sessions were vulnerable to this kind of live entry at any given moment, based on data pulled from the exposed collection.
What tl;dv Says
The researcher's account states the issue was first reported to tl;dv on January 28, with follow-up messages continuing through July, and that a company representative acknowledged the report and indicated it would be addressed, but that the flaw remained live and unfixed when the researcher went public roughly six months later. Neither the bobdahacker write-up nor the press coverage reviewed for this article cites a substantive on-the-record statement from tl;dv addressing the findings, and ubstandard.com could not independently confirm whether the underlying Firestore rule has since been corrected. Readers should treat the current status of the fix as unconfirmed pending a statement from the company.
Other collections all had proper security rules. The meetings collection was the exception.
bobdahacker, security researcher
The incident lands amid a broader run of scrutiny for AI meeting-notetaker tools, which by design record and store audio, video and transcripts of sensitive conversations on behalf of millions of users. Coverage of the disclosure, including a writeup from the Netizen Blog, has highlighted the risk that a single unlocked database collection can expose customers who never knew their calls were being indexed alongside everyone else's.
For organizations that used tl;dv during the exposure window described in the disclosure, the practical next step is auditing which meetings were recorded through the service and whether any sensitive sessions coincide with the dates cited in the researcher's data. Given the presence of government and university domains in the dataset, the disclosure may also draw attention from data-protection regulators in the jurisdictions involved, though no regulatory inquiry has been publicly confirmed as of this writing.
