BIPA Voiceprint Lawsuits 2026: What the Cases Mean for AI Training

What happened In May 2026, a coordinated wave of class-action lawsuits under the Illinois Biometric Information Privacy Act (BIPA) was filed against Google, Amazon, Apple, Meta, Microsoft, NVIDIA, ElevenLabs, Adobe, and Samsung. The allegation is consistent across all suits: these companies extracted voiceprints from publicly available audio — podcasts, audiobooks, news broadcasts, YouTube videos — and used them to train voice AI models (Gemini Live, NotebookLM, Alexa, Siri, and others) without obtaining written consent as required by BIPA. Why BIPA matters here BIPA is the strongest biometric privacy law in the United States. Enacted in 2008, it requires companies to: Inform individuals in writing that their biometric data is being collected Obtain written consent before collecting it Publish a retention schedule specifying how long the data will be kept and when it will be destroyed Protect the data with reasonable security standards The law applies to "biometric identifiers" including fingerprints, retina scans, and voiceprints. Each violation carries damages of $1,000 for negligent violations or $5,000 for reckless or intentional violations — per person. The key legal question The defense from the tech companies is expected to be that publicly available audio does not require consent. The plaintiffs argue that BIPA has no "publicly available" exception — consent is required regardless of where the audio was found. A separate but critical issue: these voiceprints are now encoded into the parameters of AI models. Unlike a password or a credit card number, you cannot change your voice. If your voiceprint is embedded in an AI model, there is no way to delete it without retraining the entire model from scratch. Who is affected If you have ever: Recorded a podcast episode Been interviewed on a news broadcast Posted a video on YouTube with your voice Narrated an audiobook Recorded a voice memo that ended up in a training dataset ...your voiceprint may already be in one or more of these models. The plaintiffs are seeking class certification for all Illinois residents whose voices were used without consent. Depending on how the court rules, the class could expand to include anyone whose voice was used, regardless of state — though BIPA's Illinois nexus requirement limits the statutory claim to state residents. What this means for AI training The lawsuits highlight a structural problem in AI development: the dominant training paradigm requires massive datasets, and the easiest way to build them is to scrape the internet without asking permission. The audio industry is particularly exposed because: Voice is inherently biometric — it identifies you uniquely Voice is irreplaceable — you get exactly one, and its biometric features do not change Voice AI is booming — ElevenLabs, NotebookLM, and Gemini Live all rely on voice models trained on scraped data If the plaintiffs win, it could establish that scraping biometric data for AI training requires affirmative consent, which would fundamentally change how voice AI models are built. What you can do Remove your audio from public platforms if you can. Old podcast episodes, interviews, and YouTube videos can be taken down or made unlisted. Check ElevenLabs, OpenAI, and other voice AI companies for opt-out forms. Several now offer voice data removal requests in response to these lawsuits. Support BIPA-like legislation in your state. Only Illinois, Texas, and Washington have comprehensive biometric privacy laws. Use voice cloaking tools for any new public recordings. Researchers have developed adversarial audio perturbations that prevent voiceprint extraction without audibly distorting the recording. Your voice is a biometric identifier, just like your fingerprint. The law is only beginning to catch up to that fact.