The AI voice heist

For decades, the voices of Carol Marin and Phil Rogers were the trusted sound of Chicago broadcast journalism. Across America’s living rooms, the smooth cadence of Robin Amer and the deep, resonant delivery of Yohance Lacour guided millions through complex investigative podcasts. In thousands of homes, Lindsay Dorcus’s voice brought over 200 audiobooks to life.

These voices are distinct, highly trained, and heavily decorated with Pulitzers, Emmys, and Peabodys. They are also, according to a series of explosive new federal lawsuits, currently being duplicated inside the commercial servers of Silicon Valley without their owners' permission.

In a sweeping legal offensive, a coalition of the nation’s most celebrated audio journalists and voice actors has sued nine of the world’s largest tech entities: Alphabet (Google), Amazon, Apple, Meta, Microsoft, NVIDIA, Adobe, Samsung, and ElevenLabs.

The nine class-action lawsuits, filed in Chicago’s federal court by the civil rights law firm Loevy + Loevy, represent a critical battleground in the AI age. The plaintiffs allege that these tech titans built a multi-billion-dollar generative AI voice industry by scraping thousands of hours of proprietary audio from the internet, extracting unique "voiceprints" to train their foundational AI models.

The legal weapon being wielded against Silicon Valley is Illinois’ Biometric Information Privacy Act (BIPA)—widely considered the most stringent data privacy law in the United States.

Enacted in 2008, BIPA was designed to protect citizens from having their biological identities commodified. Under the law, any company collecting biometric identifiers—such as fingerprints, facial geometry, retina scans, or voiceprints—must first provide written notice, publish a clear data retention policy, and obtain an explicit, written release from the individual.

The tech giants, the lawsuits allege, bypassed every single one of these steps.

"The legislators who wrote and passed BIPA had the foresight to realize that biometric privacy was going to be a major civil rights issue in the 21st century. Social Security numbers can be changed, passwords can be reset, and credit cards can be cancelled, but once your biometric data is compromised, there’s nothing you can do about it", says Ross Kimbarovsky, Attorney, Loevy + Loevy.

A voiceprint is far more than just an audio recording. As characterized across the legal filings and reported by CBS Chicago, a voiceprint operates as "a digital fingerprint of the human voice"—a mathematical representation of someone's pitch, timbre, and resonance determined by a speaker's unique physical physiology, alongside speech patterns like accent, cadence, and articulation developed over a lifetime.

"Like a fingerprint, a voiceprint identifies the individual and cannot be changed," the lawsuits state, as detailed by CBS Chicago. "A person whose voiceprint has been taken cannot recover it by altering their voice—the biological and behavioral patterns that produced the voiceprint are the same ones used to speak every day."

For authors, literary estates, and book publishers, the specifics detailed in these newly unearthed complaints outline an existential disruption to the commercial book industry. The filings expose how tech giants are leveraging vertical integration to directly undercut human workers using their own creative output.

The complaint filed against Amazon highlights a uniquely predatory "closed loop" operational model that targets the audiobook landscape. The litigation points out that Amazon owns the full, end-to-end infrastructure of modern digital publishing:

• The sourcing pipeline: Amazon possesses and controls thousands of hours of master recordings directly through its subsidiary, Audible—including the extensive multi-book catalogs of human narrators like Lindsey Dorcus and Victoria Nassif.

• The training loop: Instead of sourcing data externally, Amazon is accused of using its position as a dominant distributor to ingest these human recordings internally, feeding them directly into the "Foundational Voice Models" powering its corporate text-to-speech engines.

• The commercial replacement: Amazon then sells these automated voices back to the market via features like Audible AI narration, ACX narrator voice replicas, and Kindle Direct Publishing (KDP) Virtual Voice.

"The product Amazon competes against the narrators with is built from the narrators' own voices," the complaint states.

This structural disruption hits publishers and authors at a staggering price disparity. According to the court documents, traditional audiobook production historically requires human narrators commanding standard industry rates of $250 to $400 per finished hour, running the total cost of a standard ten-hour novel to anywhere between $3,000 and $4,000.

By contrast, the lawsuit against ElevenLabs highlights that the company offers its automated synthetic platform for a flat fee of $99 per month via its "Pro Plan," allowing authors or independent publishers to bypass human narration entirely and generate a complete audiobook in just a few hours. ElevenLabs further entices self-publishing authors and independent houses by offering an automated distribution pipeline directly to Spotify, InAudio, and ElevenReader, packaged with an aggressive 60% royalty on direct sales.

The financial fallout isn't limited to standard narration. The complaints argue that generative text-to-speech models are actively targeting highly specific, culturally authentic vocal niches that bilingual authors and diverse publishers rely on for narrative depth.

Plaintiff Victoria Nassif, a Lebanese-Palestinian American narrator, built her distinct market value on her specialized ability to provide authentic Levantine Arabic-accented English narrations for major houses like Penguin Random House, Hachette, and Simon & Schuster.

However, Meta's Massively Multilingual Speech model (trained on data spanning 1,100 languages) and ElevenLabs' Multilingual v3 engine are engineered to automatically synthesize cultural vocal traits, distinct accents, and real-time cross-lingual "speech-to-speech" translations while attempting to mimic the original speaker's emotional tone. The lawsuits argue that this directly cannibalizes the unique professional assets of marginalized and specialized human performers.

The legal filings contend that the tech companies cannot claim ignorance of the law or technical limitations. The paperwork highlights a glaring double standard: Big Tech knows exactly how to respect voice rights when the financial or legal stakes are high enough for them.

The complaint against Meta reveals that in September 2024, the social media giant executed formal, multi-million-dollar voice-licensing contracts with Hollywood elite—including Awkwafina, Judi Dench, John Cena, Keegan-Michael Key, and Kristen Bell—specifically to power its Meta AI assistant. Similarly, in November 2025, ElevenLabs launched its curated "Iconic Marketplace," a stringent consent-and-licensing platform ensuring that living actors like Michael Caine and Matthew McConaughey, alongside the historic estates of Maya Angelou and Alan Turing, must affirmatively approve any commercial use of their voices.

The lawsuits emphasize that while these custom safeguards protect high-profile celebrities, the general pool of working voice talent and authors whose master tracks were scraped from the web are left completely uncompensated and unprotected.

Historically, BIPA lawsuits have centered on workplace infractions—such as a grocery chain requiring warehouse employees to use voice-verification or fingerprint timeclocks without proper paperwork (such as Amazon-owned Whole Foods' $300,000 settlement in early 2023).

But this litigation pushes into uncharted legal and technical territory because of where the data now lives. In the lawsuit against Google, obtained by Biometric Update, the plaintiffs argue that their voices cannot simply be deleted from a standard database.

“The voiceprints Google extracted from Plaintiffs are not stored in a database that can be deleted on request. They are encoded in the parameters of Google’s commercial voice models and reproduced in the audio that those models generate. At this point, the biometric data and the product are the same thing.”

This presents a terrifying prospect for AI developers. If federal judges rule in favor of the plaintiffs, the demanded remedy includes not only billions in statutory damages, but strict injunctive relief. The court could force these tech companies to cease unauthorized collection and destroy all unlawfully obtained voiceprints.

Because the biometrics are permanently woven into the algorithmic weights of the software, complying with the order could force tech giants to completely dismantle, delete, and retrain their flagship AI models from scratch—potentially wiping out systems like Amazon Polly, Meta Audiobox, and ElevenLabs v3 entirely.

As the legal world watches Chicago, the tech giants face a reckoning: the very human traits that make their AI sound so lifelike may ultimately be the flaw that forces them to be shut down.

At the time of publication, none of the tech companies named in the lawsuits have responded to requests for comment.