Technology

Transcription Firm Used by Police and Tech Giants Left Customer Recordings Open Online

Martin HollowayPublished 2m ago3 min readBased on 1 source
Reading level
Transcription Firm Used by Police and Tech Giants Left Customer Recordings Open Online
Photo by Brett Sayles on Pexels

Tigerfish, a San Francisco transcription company whose customers include law enforcement agencies, government departments and Silicon Valley technology firms, left a large volume of internal data exposed to the public web, according to TechCrunch. The exposed material included customer transcriptions and source materials, among them transcripts of police interviews and body-cam footage.

The reporting also describes customers' raw audio and video footage, and invoices issued to police departments and prosecutors' offices. The files were stored in an Amazon-hosted S3 bucket, a container for stored files in Amazon's cloud storage service. Anyone who could guess the bucket's web address could reach the contents, and no password was required.

Security researcher Anurag Sen found the data online in October and alerted the company. He said he did not receive a response. Sen later provided details of the exposure to TechCrunch so that the incident could be flagged for resolution.

Tigerfish secured the exposed bucket soon after TechCrunch contacted the company. Company president Adam Goldberg said Tigerfish blocked all public access to the identified bucket and verified that all of its other S3 buckets were blocked to public access.

The mechanics will be familiar to anyone who has audited cloud storage. The bucket name was predictable and had no authentication layer, so exposure depended only on finding the address. S3 offers account-level and bucket-level Block Public Access settings, which can override permissive policies. A bucket holding sensitive recordings should never rely on an obscure name, in the same way a spare key should never be left under the doormat. Whether the misconfiguration was a policy override, a legacy setting or a deployment script error has not been established. The fix Goldberg described is specific: close the identified bucket, then check the account's other buckets against the same control.

The customer mix is what makes the exposure significant for the wider market. Transcription vendors sit in the middle of evidentiary workflows. A body-cam recording or police interview sent to a third-party service typically passes through a contractor's storage, staff and tooling before it reaches a prosecutor's file. Each handoff adds systems that must be secured. The same vendor also served private technology clients, so one storage misconfiguration reached both public-sector and commercial data pipelines.

In my view, the lesson for procurement teams is less about Tigerfish specifically than about how little visibility a buyer usually has into a vendor's storage. Contract clauses and security questionnaires rarely test whether a bucket answers an unauthenticated request. External attack-surface scanning, S3 access analysis and periodic review of public-access settings are routine controls, and they cost far less than disclosing evidentiary recordings.

The broader context here is that transcription is one of the more practical applications of speech recognition. Automated transcription and summarization of interviews and footage can shorten case preparation and reduce clerical load for agencies and legal teams. Those gains depend on a vendor's data handling being as disciplined as its models. The Tigerfish case involves a basic access-control failure, and the fix is well understood. The architecture that makes automated transcription useful to public institutions can be built with the same controls that protect the data it processes, and the tooling for doing so has matured considerably.