Voice Deepfake Detection and the Scarlett Johansson OpenAI Row

OpenAI launched GPT-4o with a voice called Sky that sounded uncannily like Scarlett Johansson. Johansson said she had declined to voice the product, and was shocked at the similarity. OpenAI paused the voice. Lawyers got involved.

How it worked

OpenAI said Sky was voiced by a different actress. Johansson's team was not convinced, and frankly, neither was anyone who heard it. Whether it was deliberate cloning or convergent casting, the result was the same: one of the world's most recognisable voices, seemingly replicated without permission.

The case became the defining voice-rights fight of the AI era. If a company as scrutinised as OpenAI can ship this, what is happening in the shadows? The case is now the reference point for every AI voice-rights argument, and it will be cited for years.

Detection angle

Voice deepfake detection can compare a suspect voice against reference samples and score the similarity. A deepfake voice detector does not care about corporate explanations: it measures acoustic fingerprints. Audio deepfake detection is how disputes like this get settled with data instead of PR statements.

Wondering how to detect deepfake voice calls and cloned voices? Run the audio through a detector, compare against known samples, and do not accept coincidence as an answer when the waveforms match. Voices are biometric data. Treat them that way. The law is catching up slowly. The tech is not waiting.

Sounding like someone is not a coincidence at scale. Especially when there is serious funds riding on the resemblance. Try the AI Detector on suspicious voice audio.