Back to notes

February 20, 2026 · 5 min read

A Voice Cloning Use Case That Raised More Questions Than Answers

A client wanted to clone their CEO's voice for scaled video content. The technology worked fine. The questions it raised didn't have clean answers.

Hot TakeEthics

The request came from a client's marketing team: clone the CEO's voice, with his full consent and a recorded sample, to produce localized video content in languages he doesn't speak, at a volume that would be impossible to record in person. Technically trivial. The CEO was enthusiastic and fully consenting. And I still spent more time on the ethics memo than the implementation.

Consent from the subject isn't the whole question

The CEO consenting to his own voice being cloned answers one question and leaves several open. Once a usable voice model exists, who controls it — is it retained by the vendor, deletable on request, reusable for anything beyond the agreed scope? What happens if the company and the CEO part ways — does the clone get retired, or does it quietly keep existing in someone's model weights? And a harder one: once this is normal for the CEO, does it become an expectation for every executive the company hires, consent becoming a condition of employment rather than a free choice?

The CEO's consent answered whether we could build it. It didn't answer who controls it once it exists, or what happens to it after he's no longer the CEO.

We ended up building it, but with contractual terms that hadn't existed in the vendor's default agreement: an explicit data retention limit, a defined deletion process tied to the CEO's departure from the role, and a disclosure requirement that any AI-generated video using the clone carry a visible label. None of that was technically necessary to ship the feature. All of it was necessary to ship it responsibly.

  • Individual consent to a voice clone doesn't resolve questions of ongoing control, retention, and reuse — negotiate those explicitly, in writing.
  • A disclosure label on AI-generated executive content costs almost nothing and meaningfully reduces downstream reputational risk.
  • Think through departure and offboarding scenarios before the clone exists, not after someone leaves and the model is still sitting there.
  • The technology being consensual and well-intentioned in this one case doesn't mean the precedent it sets is automatically fine.

This is the kind of project where the interesting engineering work is finished in an afternoon and the interesting work that actually matters is a legal and ethical conversation that takes two weeks. Treat that ratio as normal for anything touching synthetic likeness, not as scope creep.