Voiceprints under BIPA

Nine class actions allege Alphabet, Amazon, Apple, Meta, Microsoft, NVIDIA, Samsung, Adobe and ElevenLabs took voice recordings from Illinois broadcasters, podcasters and audiobook narrators to train

Nine class actions commenced in the Northern District of Illinois between 11 and 13 May 2026 allege that Alphabet, Amazon, Apple, Meta, Microsoft, NVIDIA, Samsung, Adobe and ElevenLabs extracted voiceprints from recordings of Illinois broadcasters, podcasters and audiobook narrators in the course of training AI foundation voice models. The plaintiffs say none of this was preceded by notice, written consent or a publicly available retention policy. The statute they sue under is the Illinois Biometric Information Privacy Act 2008.

The plaintiffs and the defendants

The named plaintiffs include Carol Marin (CBS News, 60 Minutes, Chicago Tonight), Phil Rogers (NBC Chicago, WBBM), Robin Amer (The Washington Post and The City podcast), Yohance Lacour (You Didn’t See Nothin’), Lindsey Dorcus (audiobook narrator for Penguin Random House, Hachette and Disney), Victoria Nassif (Chicago PD; audiobooks for Penguin, Hachette and Audible) and Alison Flowers (Invisible Institute; Somebody podcast). Flowers is named in every action except those against Amazon and Apple. The plaintiffs are represented by Loevy + Loevy with Ross Kimbarovsky as lead counsel.

The statute

Section 10 of BIPA defines biometric identifier as a retina or iris scan, fingerprint, voiceprint, or scan of hand or face geometry. Voiceprint has sat in the statutory definition since the Act was passed in 2008. Section 15 sets the obligations that BIPA actions are typically built on. Section 15(a) requires a publicly available retention and destruction schedule. Section 15(b) requires written notice and written consent before collecting a biometric identifier. Section 15(c) prohibits selling, leasing or otherwise profiting from biometric data. Section 15(d) restricts disclosure without consent or statutory exception, and Section 15(e) requires reasonable care in storage and protection. Section 20 fixes statutory damages at $1,000 for negligent violations and $5,000 for intentional or reckless violations, plus reasonable attorneys’ fees, costs and injunctive relief.

Damages after Clay

After Cothron v. White Castle, Section 15(b) and Section 15(d) claims could accrue with each scan or transmission. That was the source of the per-scan exposure problem in workplace fingerprint litigation. A biometric timeclock used repeatedly over several years could produce thousands of alleged statutory violations for the same employee.

The Illinois General Assembly passed Public Act 103-0769 in August 2024 to narrow that exposure. For repeated Section 15(b) collection and Section 15(d) disclosure, the statute now limits recovery to one award per person, per method, for the same biometric identifier or information. On 1 April 2026, the Seventh Circuit in Clay v. Union Pacific Railroad Company, No. 25-2185, held that the amendment applies retroactively to cases pending at the time of enactment.

The result is a narrower damages framework. The question is no longer whether repeated ingestion events can multiply exposure in the same way as repeated fingerprint scans. It is whether the alleged extraction of a voiceprint from training recordings falls within BIPA at all.

The Section 10 question

The decisive question across the nine complaints is whether training a foundation voice model on a recording extracts a voiceprint within the meaning of Section 10. The plaintiffs say it does. A foundation voice model is trained on pitch, cadence, tone and vocal-tract characteristics. On that case, the mathematical representation those features produce inside the model is a voiceprint by another name. The defendants are likely to say it does not. On the defence view, a recording is audio data and a voiceprint is a stored mathematical identifier used to authenticate a speaker. If the data is generalised across many speakers and is not retained as a speaker-authentication template, the defence will say it is neither. Microsoft has already advanced that position in separate Teams litigation: Basich v. Microsoft Corp., W.D. Wash.

What to watch

Two thresholds will shape the litigation.

First, whether the Northern District of Illinois accepts the plaintiffs’ theory that training a foundation voice model on a recording is collection of a voiceprint within Section 15(b).

Second, whether the post-Clay damages framework leaves enough exposure and settlement leverage to support coordinated nine-defendant litigation. The Section 10 voiceprint question will decide the first point. Early consolidation decisions, dismissal motions and any first settlement structure will show the second.

Sources

Loevy + Loevy press release, 14 May 2026.

Chicago Sun-Times, 19 May 2026.

Capitol News Illinois, 15 May 2026.

740 ILCS 14/1 et seq. (Illinois Biometric Information Privacy Act).

Public Act 103-0769 (eff. 2 August 2024).

Cothron v. White Castle System, Inc., 2023 IL 128004.

Clay v. Union Pacific Railroad Company, No. 25-2185 (7th Cir. 1 April 2026).

Basich v. Microsoft Corp., W.D. Wash. (pending).

CourtListener, Basich v. Microsoft Corporation, No. 2:26-cv-00422, W.D. Wash.

UC Today, Microsoft Teams Lawsuit: BIPA Class Action Targets AI Voice Data, 17 February 2026.

Law360, Microsoft Says Teams Info Not ‘Voiceprint’ Under BIPA.