Google AI Accused of Voice Theft
Radio host David Greene alleges that Google’s AI podcast tool, NotebookLM, cloned his voice without his authorization. The claim underscores the growing risks of large-scale voice and likeness theft by generative AI models, posing new threats to privacy and authenticity for journalists and public figures.
- Google's defense, articulated by spokesperson José Castañeda, is that the allegations are "baseless" and that the voice used in NotebookLM was created from a paid professional actor. - The lawsuit, filed in Santa Clara County, California, cites an AI forensic firm's analysis which concluded with a 53% to 60% confidence score that Greene's voice was used to train the AI model. - The legal precedent for such cases in the U.S. is rooted in "right of publicity" lawsuits, such as those successfully brought by Bette Midler and Tom Waits, which established that hiring "soundalike" performers for commercials can be a form of appropriation. - Under the European Union's AI Act, companies deploying such systems face strict transparency obligations; Article 50 of the regulation mandates that users must be informed when they are interacting with an AI system and that AI-generated synthetic content must be labeled as such. - Voice data can be classified as biometric data under the GDPR, meaning its use for training an AI model would require explicit consent from the individual, a higher standard than for other types of personal data. - The dispute follows a similar high-profile incident in 2024 where actress Scarlett Johansson alleged that an OpenAI voice called "Sky" closely resembled her own, despite her having declined an offer to voice the assistant. - Greene's legal team from Boies Schiller Flexner is also representing book authors in a separate high-profile AI copyright lawsuit against Meta, indicating a broader legal strategy against tech firms' use of training data. - A key legal question is whether an AI imitating a voice constitutes a new, independent creation or an infringing derivative work; past U.S. court rulings have suggested copyright law only protects original fixed recordings, not the abstract qualities of a voice.
