The Voices That Trained AI Are Fighting Back Under Illinois Law

A professional microphone in a recording studio with pop filter

TL;DR:

What: Seven Illinois-based journalists, voice actors, and narrators filed class-action lawsuits on May 14 against ten major tech companies for scraping their voices to train AI without consent
Who's sued: Amazon, Apple, Google, Meta, Microsoft, NVIDIA, ElevenLabs, Adobe, and Samsung
The legal weapon: Illinois BIPA, the same biometric privacy law that forced Meta to pay $650 million and Google to pay $100 million
The damages: $1,000 per negligent violation, $5,000 per intentional violation, per person, uncapped, applied to a class of potentially millions
Products named: Google Text-to-Speech, NotebookLM Audio Overviews, Meta Voicebox, Adobe Firefly, ElevenLabs voice cloning
What to watch: Defendant responses, court scheduling, and whether this triggers a settlement wave like the facial recognition cases did

Ten Lawsuits in One Day

On May 14, 2026, a group of seven plaintiffs walked into federal court in the Northern District of Illinois and filed class-action lawsuits against ten of the biggest names in tech. The claim: these companies scraped hundreds of thousands of hours of human speech to train AI voice models without asking, without paying, and without telling anyone.

The plaintiffs aren't anonymous consumers. They're some of the most recognizable voices in Chicago journalism and audio production:

Carol Marin: CBS News, 60 Minutes, Chicago Tonight. Multiple Emmy and duPont-Columbia awards
Phil Rogers: NBC Chicago, WBBM. Emmy Award winner
Robin Amer: Peabody Award-winning journalist and podcaster
Yohance Lacour: journalist and audio producer
Alison Flowers: Pulitzer Prize finalist, investigative journalist
Lindsay Dorcus: professional audiobook narrator
Victoria Nassif: professional audiobook narrator and voice actor

Their attorney, Ross Kimbarovsky of the civil rights firm Loevy + Loevy, didn't mince words: "What we are seeing is an illegal and unethical exploitation of talent on a massive scale, and one of the largest violations of biometric privacy ever committed."

Why BIPA Changes Everything

Most AI training lawsuits are filed as copyright claims. They're expensive, slow, and the legal framework wasn't built for AI. Courts are still figuring out whether training a model on copyrighted data is fair use. Nobody knows how those cases end.

These lawsuits skip that fight entirely. They're filed under the Illinois Biometric Information Privacy Act (BIPA) and that changes the math completely.

BIPA has been law since 2008. It's the strongest biometric privacy statute in the country. It covers fingerprints, facial geometry, retina scans, and voiceprints. And here's what makes it lethal for defendants:

Private right of action: You don't need a regulator to sue on your behalf. Any individual whose biometrics were collected without consent can file a claim.

Statutory damages: $1,000 per negligent violation, $5,000 per intentional or reckless violation. Per person. These aren't theoretical. They're written into the statute.

No actual harm required: You don't have to prove your voiceprint was sold, leaked, or misused. The collection itself, without consent, is the violation.

Proven track record: Meta paid $650 million over facial recognition. Google paid $100 million. TikTok paid $92 million. BIPA isn't theoretical litigation. It's the law that already extracted billions from Big Tech.

The complaints allege the defendants made "deliberate institutional decisions" to violate BIPA, skipping the consent process entirely to train their voice models faster. If courts agree that was intentional, the $5,000 per-violation rate kicks in. Multiply that across a class of potentially millions of Illinois residents whose voices appear in training datasets, and the numbers get astronomical.

What They Built With Stolen Voices

The lawsuits name specific products powered by voice data the defendants allegedly scraped without consent:

Google: Gemini Live, NotebookLM Audio Overviews, YouTube auto-dubbing, Google Cloud Text-to-Speech, Google Assistant
Meta: Voicebox, Meta AI voice assistant
Adobe: Firefly (the generative AI model family)
Microsoft: Azure AI Speech, Copilot voice features
Apple: Siri voice synthesis, Personal Voice
Amazon: Alexa voice models, Amazon Polly
NVIDIA: NeMo speech AI toolkit, Riva conversational AI
ElevenLabs: Voice cloning and text-to-speech platform
Samsung: Bixby voice assistant, Galaxy AI features

The complaints describe how these companies "ingested hundreds of thousands of hours of human speech" from publicly available recordings (broadcast journalism, podcasts, audiobooks) and extracted voiceprints without consent. A voiceprint, the complaints explain, is a "digital fingerprint of the human voice," a mathematical representation as unique and identifiable as a fingerprint or facial scan.

The Adobe complaint is particularly blunt: the company "treated the human voices that built Firefly as ownerless: ignoring the speakers' rights, taking their voiceprints without asking, paying them nothing, and giving them no notice that their voices were being used at all."

Why This Isn't a Copyright Case

Copyright lawsuits against AI companies have been grinding through courts for years. The New York Times sued OpenAI. Getty Images sued Stability AI. Authors, musicians, and visual artists have all filed claims. Most of them are stuck in discovery or arguing about fair use.

The BIPA angle sidesteps the copyright debate entirely. These aren't claims about who owns a recording. They're claims about who owns a biometric identifier extracted from that recording. And BIPA has already answered that question: the person it belongs to.

Under BIPA, a company must do three things before collecting biometric data:

Inform the subject in writing that their biometric data is being collected
Disclose the specific purpose and duration of collection
Obtain a written release from the subject

None of the ten defendants did any of this, according to the complaints. They didn't notify. They didn't disclose. They didn't get written consent. For every Illinois resident whose voice appears in their training data, that's a standalone BIPA violation.

The other legal angle: Illinois's Right of Publicity Act, which gives residents control over the commercial use of their identity, including their voice. The defendants didn't just collect voiceprints. They used them to build commercial products worth billions. That's a separate violation with its own damages.

The Precedent That Makes This Dangerous

BIPA isn't untested. It's the most battle-hardened biometric privacy law in the country. Here's the settlement history:

Meta (Facebook): $650 million (2021), facial recognition in photo tagging
Google: $100 million (2023), facial recognition in Google Photos
TikTok: $92 million (2022), facial and voice data collection
Snapchat: $35 million (2023), facial recognition in filters
Clearview AI: $50 million (2024), facial recognition database

Every one of those cases settled. None went to a full trial. Companies look at the per-violation math, look at the class size, and write a check. That pattern is exactly what makes the voice AI cases so threatening: the defendants know how this plays out. They've already been through it.

The difference this time is scale. Facial recognition cases involved photos uploaded to a single platform. Voice training data comes from everywhere: broadcast archives, podcast feeds, YouTube, audiobook platforms, public recordings. The potential class is enormous.

What This Means for the AI Industry

If these lawsuits succeed (or more likely, settle for large sums) the ripple effects go way beyond ten companies.

Every AI company that trained voice models on publicly available audio without consent has the same BIPA exposure. That includes startups, research labs, and companies that licensed training data from aggregators who themselves didn't obtain consent. The liability chain could extend far beyond the named defendants.

ElevenLabs is a particularly interesting case. The company built its entire business on voice cloning technology. It's already faced prior litigation from voice actors. A BIPA judgment or settlement could threaten its core business model, and every competitor that works the same way.

The broader message: the AI industry's "scrape first, negotiate later" approach to training data just hit a wall. Copyright law is ambiguous enough to argue about. BIPA isn't. You either got written consent or you didn't. And they didn't.

What You Can Do

If you're an Illinois resident: You may be part of the class. If your voice appears in any publicly available recording (podcasts, YouTube videos, broadcasts, audiobooks) and any of these companies used that content for AI training, you have potential BIPA claims. Watch for class certification notices. Loevy + Loevy's BIPA AI class action page has details.

If you're a voice professional: Document everything. Save copies of your published work, contracts, and any communications with platforms that host your content. If your voice was used to train AI without consent, that documentation strengthens a claim.

If you care about biometric privacy: Support BIPA-style legislation in your state. Only a handful of states have biometric privacy laws, and none are as strong as Illinois. The reason these lawsuits have teeth is because of one state law passed in 2008. Imagine if every state had one.

Check your AI tool permissions: If you use AI voice features (Siri, Google Assistant, Alexa, or voice cloning tools) review what voice data they're collecting and storing. Opt out of voice data sharing where possible. Your voiceprint is a biometric identifier, whether the company collecting it treats it that way or not.

The Bottom Line

The AI industry spent years scraping the internet's voices to build billion-dollar products. They treated human voices as raw material: free for the taking, no permission needed, no compensation offered. The people whose voices were taken are now using the only law strong enough to fight back.

BIPA has already cost Big Tech billions in the facial recognition era. The voice AI era could cost them even more. The training datasets are bigger, the products are more profitable, and the violations (if courts agree they happened) are more numerous.

Ten companies. Seven plaintiffs. One law. The defendants know how BIPA cases end. The only question is how much the check will be.

Sources

Published: May 16, 2026