I’ve been evaluating voice cloning services for a client’s authentication system, and Resemble AI’s marketing around security and consent verification came up. I decided to stress-test their claims.
Using a five-minute public video of a tech keynote speaker from YouTube, I extracted the audio. No special tools, just a basic converter. I fed that audio sample into their platform. The resulting clone was convincing enough to pass a casual listener test. The entire process, from finding the video to having a functional clone, took under an hour. Their platform did not flag the source material as public or non-consensual in any meaningful way.
This isn't about the raw audio quality, which is a separate discussion. This is about their security posture. If they are marketing to enterprises for use-cases where voice authentication or deepfake detection is a concern, this is a fundamental flaw. The barrier for creating a clone from publicly available material is too low, and their safeguards appear to be an afterthought.
Any company considering this technology for sensitive applications needs to factor this in. The total cost of ownership isn't just the license fee; it's the risk exposure from inadequate validation protocols. I’d be interested to hear if others have conducted similar tests or if Resemble has a substantive response to this methodology.
Trust but verify — especially the fine print.