Nine Tech Companies Face Chicago Suits Over AI Voice Training Data

Nine technology companies are fighting proposed class actions in Chicago that accuse them of using recorded human voices to train AI systems without permission. The cases could test whether Illinois biometric-privacy law can govern a training-data pipeline built from recordings available online.

The defendants are Apple, Amazon, Meta, Microsoft, Nvidia, Samsung, Alphabet, Adobe and ElevenLabs. The plaintiffs are journalists, podcasters, voice actors and audiobook narrators who filed nine lawsuits in the Northern District of Illinois between May 13 and May 19.

Consent, not copyright, is the core dispute

The complaints allege that the companies collected recorded voices and used them to train AI voice models without disclosure or consent. They invoke Illinois’s Biometric Information Privacy Act, or BIPA, alongside state privacy, publicity, consumer-fraud and deceptive-practices laws.

BIPA requires companies collecting biometric data to give notice and obtain consent. It also requires a written policy covering data destruction and bars the sale of that data. The suits do not allege copyright infringement, separating this fight from other litigation over AI training material.

$1,000Negligent BIPA violation

Illinois narrowed BIPA penalties in 2024, but the law still provides a $1,000 penalty for a negligent violation, applied per victim and statutory subsection.

$5,000Intentional or reckless BIPA violation

An intentional or reckless violation can carry a $5,000 penalty under the same per-victim and per-subsection framework.

That structure explains why the cases carry stakes beyond a simple claim for lost licensing revenue. Plaintiffs’ lawyer Ross Kimbarovsky said each defendant faces potential damages of hundreds of millions of dollars. That is an advocacy-side estimate, not a court determination of liability or damages.

The defense rests on proof and place

The companies deny the allegations and have filed motions to dismiss. Their central position is that the claims are speculative: the plaintiffs have not established that their recordings were used to train the companies’ systems, and the defendants argue BIPA may not apply if the alleged misuse did not occur in Illinois.

The competing legal theories

• Plaintiffs’ position: collecting and systematically harvesting voice data without the required process is itself the harm BIPA addresses, even if a speaker cannot point to a commercial product using a cloned version of their voice.
• Defendants’ position: the complaints do not adequately show that the recordings were used for AI training, and alleged online availability alone does not establish that use.
• Meta’s position: the company says it gives people information about how their data is used and that the plaintiffs’ claims are incorrect.

Apple’s filing makes the evidentiary divide unusually concrete. Its lawyers argued that the plaintiffs allege only that their recordings are online and offer no facts supporting an inference that Apple used those recordings to train its technology. The argument frames the first threshold question: what must a plaintiff show about a model developer’s training-data inputs before the case can proceed?

A single judge could shape the next phase

The nine cases are currently assigned to seven federal judges in Chicago. Apple has asked to consolidate them before one judge, a procedural move that could bring the overlapping questions of consent, training-data proof and Illinois nexus into one track.

For now, the lawsuits remain allegations, and the companies’ dismissal arguments have not been decided. But the cases put a sharper question before the court than a broad dispute over internet data: whether voice recordings used in AI development trigger a consent-and-retention regime designed for biometric information.