Adobe Firefly BIPA voice-training class action
Summary
On May 14, 2026, seven Illinois journalists, podcasters, and audiobook narrators filed a proposed class-action lawsuit against Adobe in the US District Court for the Northern District of Illinois, alleging that Adobe trained its Firefly generative AI models on the plaintiffs' voiceprints in violation of the Illinois Biometric Information Privacy Act. The suit is one of nine parallel BIPA complaints the same plaintiffs filed against major AI vendors that week. Adobe has not issued a public statement on the complaint.
What happened
- The named plaintiffs are Carol Marin, Phil Rogers, Robin Amer, Alison Flowers, Lindsay Dorcus, Yohance Lacour, and Victoria Nassif, including Pulitzer and Peabody recipients and professional audiobook narrators.
- The complaint alleges Adobe captured the plaintiffs' voiceprints from publicly available recordings, stored them, and ingested them into the training pipeline behind Firefly without written consent, notice, or a published retention policy.
- The complaint cites BIPA's requirements for written consent and disclosure of biometric retention practices, and seeks damages on behalf of an Illinois class.
- The same legal team at Loevy + Loevy filed parallel complaints against Amazon, Apple, Google, Meta, Microsoft, NVIDIA, ElevenLabs, and Samsung on the same day.
- Loevy + Loevy's filing alleges Adobe "treated the human voices that built Firefly as ownerless."
Timeline
- 2026-05-14 -- Loevy + Loevy files nine BIPA class-action complaints in the Northern District of Illinois, including the suit naming Adobe and targeting Firefly.
What remains unclear
- Adobe has not issued a public statement on the complaint.
- The complaint does not enumerate which specific recordings were ingested or how Adobe sourced them.
- The complaint does not identify which Firefly speech model, partner-supplied model, or training corpus is alleged to contain the plaintiffs' voiceprints.
- The court has not yet ruled on class certification or on whether BIPA applies to publicly broadcast voice recordings used for model training.
Broader context
Illinois' Biometric Information Privacy Act has become the most-tested US statute for training-data disputes because it imposes per-violation statutory damages and predates the current generation of voice-cloning models. Whether voiceprints derived from publicly broadcast speech qualify as protected biometric identifiers under BIPA is unresolved, and the outcome of these coordinated filings will set early precedent for what consent looks like when a generative model is trained on the recorded human voice.
Sources
