Amazon, Meta, Microsoft, NVIDIA, Samsung and other major technology companies are facing nine proposed class-action lawsuits in Chicago over allegations that they used people’s voices to develop AI systems without obtaining the required consent.
The lawsuits were filed in federal court in Illinois by journalists, podcasters, audiobook narrators and voice actors who say recordings of their voices were used to create voiceprints and train commercial AI voice technologies.
The cases name Amazon, Apple, Google and its parent company Alphabet, Meta, Microsoft, NVIDIA, Samsung, Adobe and AI voice company ElevenLabs. The plaintiffs are seeking damages and other remedies under Illinois’ Biometric Information Privacy Act, commonly known as BIPA.
The lawsuits put a new focus on how AI companies obtain and use human voices, particularly recordings that are already publicly available online.
What the plaintiffs are alleging
The plaintiffs claim the companies extracted unique characteristics from their voices in recordings available through platforms such as podcasts, YouTube and audiobooks.
Those characteristics can be used to create a voiceprint, a biometric identifier that can distinguish one person’s voice from another.
According to the complaints, the companies then used those voiceprints in AI training and development without first informing the individuals, explaining how long the biometric information would be retained or obtaining the written consent required under Illinois law.
The plaintiffs include well-known journalists and audio professionals such as Carol Marin, Phil Rogers, Robin Amer, Lindsay Dorcus, Victoria Nassif, Yohance Lacour and Alison Flowers.
Several of the plaintiffs have built careers around their voices through broadcast journalism, podcasting, audiobook narration and voice acting.
Their argument is that making a recording publicly available does not automatically give an AI company permission to extract a person’s biometric information from it and use that information commercially.
Why Illinois law is central to the cases
The lawsuits rely heavily on Illinois’ Biometric Information Privacy Act.
BIPA requires companies that collect certain biometric identifiers to provide notice, explain the purpose and duration of collection and obtain written consent before collecting or using the information in covered circumstances.
The law specifically covers biometric identifiers and information, including voiceprints.
Illinois’ law is also notable because it allows individuals to bring private lawsuits over alleged violations.
That has made BIPA one of the most important US laws governing the collection and use of biometric information.
The law has previously been used in major cases involving facial recognition and other biometric technologies. Meta, for example, agreed to a $650 million settlement in a previous Illinois biometric privacy case involving facial recognition.
The new cases test whether the same legal protections can apply to the way AI companies process people’s voices for machine learning.
The technology companies have denied wrongdoing and are seeking to have the cases dismissed.
One of the major legal questions is whether BIPA can apply when the alleged collection or use of biometric information happened outside Illinois.
The defendants are also expected to challenge other aspects of the plaintiffs’ claims as the cases progress. Reuters reported on August 20 that lawyers for the companies and the plaintiffs were already arguing over the scope of the cases.
The lawsuits do not primarily focus on copyright infringement.
Instead, the central issue is whether the companies collected and used biometric information without following Illinois’ consent and disclosure requirements. That makes the cases different from the copyright lawsuits that have dominated much of the legal debate around AI training data.
Why voice data is different from ordinary training data
The dispute also raises a difficult question about what makes a person’s voice different from ordinary content published online.
A journalist may publish hundreds of hours of interviews, broadcasts or podcasts. A voice actor may have thousands of hours of recorded performances available through audiobook platforms and other services.
Those recordings can be useful to AI companies developing speech-generation systems because they contain examples of how real people speak, pronounce words and express themselves.
But a person’s voice is also a biometric characteristic.
Unlike a written article, photograph or other piece of content, a voice can be used to identify or imitate the individual who produced it.
That is why the plaintiffs argue that simply finding a recording online should not be treated as permission to extract a person’s voiceprint and use it to build commercial AI systems.
The case could affect AI voice development
The lawsuits come as AI companies increasingly develop systems that can generate realistic human speech from short prompts.
Modern voice models can reproduce accents, speaking patterns and other characteristics of human speech, making it possible to create synthetic narration, dubbing, voice assistants and other applications.
That creates new commercial opportunities but also raises questions about consent and compensation for the people whose recordings help make those systems possible.
If the plaintiffs succeed, the cases could create additional legal pressure on AI companies to document where voice data comes from and obtain appropriate consent before using it to develop commercial systems.
A ruling in favour of the companies, on the other hand, could narrow the circumstances in which Illinois’ biometric privacy law can be used against AI developers.
For now, the cases remain allegations. No court has determined that the companies unlawfully used the plaintiffs’ voices.