Tigerfish left police interview transcripts and body-cam footage open to anyone online, as first reported by TechCrunch on November 20, 2026. The San Francisco transcription firm, which serves police, prosecutors and tech giants, stored the files in an Amazon S3 bucket with no password.
What Happened?
- The exposed files included police interview transcripts, body-cam footage, raw customer audio and video, and client invoices.
- Researcher Anurag Sen, 30, found the files in October, but his warning email landed in Tigerfish’s spam filter.
- Tigerfish blocked public access soon after reporters contacted it and says all of its other S3 buckets also block public access.
- Tigerfish has not said whether it notified customers, a group that includes police departments and prosecutors’ offices.
Sen found the data on an Amazon-hosted S3 bucket that was openly listed. Anyone who guessed its simple web address could browse the files without a password. The contents covered finished transcripts and the source recordings behind them, from police interviews to body-cam video.
A researcher’s warning went to spam
Sen spotted the exposure in October and alerted Tigerfish, but he heard nothing back. He then shared the details with TechCrunch, which verified the data and its source before contacting the firm. Tigerfish secured the bucket soon after and said Sen’s email had been caught in its spam filter.
Tigerfish president Adam Goldberg said in an email that the company had “blocked all public access to the bucket.” The block covers directory listings and individual files, he said. Goldberg added that Tigerfish verified its other S3 buckets block public access and is reviewing its storage security more broadly.
Sen has hunted exposed datasets for the past decade from his island home in Indonesia, work he likens to unpaid air-traffic control. His earlier finds include U.S. military emails leaking from a Department of Defense cloud server. He also uncovered the inner workings of a national license plate surveillance system.
Sen says simple mistakes drive many leaks
Sen scans the web daily with automated scanners and specialist search engines. He said many exposures he finds come down to basic human error, despite the focus on AI speeding up vulnerability discovery. His example is an engineer who disables authentication on a Friday test database and forgets to restore it before the weekend.
Platforms have started adding guardrails against that kind of slip. Amazon has made it harder to make S3 buckets public without clear multi-step warnings. Vibe-coding platforms Lovable and Supabase have also changed their products to curb user mistakes. Those fixes arrive as AI-generated code pushes more unchecked software onto the web.
Exposed cloud storage is a cross-provider weakness, the same gap behind rising attacks on Azure Blob Storage. For Tigerfish’s clients, the problem sits in the vendor layer. Police and prosecutors who send recordings to an outside transcriber inherit that firm’s storage settings. The same third-party risk surfaced this week in JPMorgan Chase’s vendor hack warning.
Open Questions
The evidence so far shows the files were reachable. It does not show who else opened them. Four questions are still unanswered:
- How long was the bucket public before Sen found it?
- Did anyone besides Sen and the reporters download files?
- Which agencies and companies had material in the bucket?
- Were any other Tigerfish buckets public before the company checked them?
Agencies and companies that used Tigerfish should ask the firm whether their files sat in the exposed bucket. They should also ask whether it can see who accessed them. Teams running S3 storage can confirm that every bucket blocks public access, a check that helps reduce the risk of a similar leak.
A researcher’s warning landed in a spam filter, and the storage fix came only after a reporter got involved. That leaves the disclosure path as the harder repair for a firm holding police recordings. Customer notification is the next checkpoint, and Tigerfish has not said whether it has started.



















































