Connecticut streamer Warren Pandiscia filed a proposed class action lawsuit on August 20 in the Northern District of California against Twitch and parent company Amazon, alleging the platforms harvested his streams, clips, chat logs, and other content to train Amazon's generative AI models without consent or compensation. The 37-page complaint claims the scraping began as early as 2024—well before any disclosure—and that Twitch defaulted every account into the program because, as Chief Product Officer Mike Minton admitted, “If it was opt-in, nobody would opt in.” Pandiscia, who has streamed over 1,000 hours and has roughly 900–1,000 followers, opted out the moment the buried setting appeared but argues the damage is permanent because AI training data cannot be clawed back.

The suit alleges breaches of implied and express contracts, unjust enrichment, and violations of California’s unfair competition law, while highlighting a critical loophole: the opt-out applies per channel, not per account, meaning a creator’s voice or image can still be harvested if they appear in someone else’s opted-in stream. Twitch reportedly sent no email or notification when flipping the switch on August 12; the change was discovered by a reporter. The complaint seeks damages, disgorgement of profits, and injunctive relief to stop the practice and remove materials from training datasets.

This isn’t an isolated grievance. It follows a pattern of AI companies quietly ingesting vast troves of creator work under vague terms of service, betting that most users won’t notice or fight back. Twitch’s own admissions and the lawsuit’s timeline expose a deliberate design choice to prioritize Amazon’s commercial AI ambitions over creator control. Whether the class gets certified or the case survives motions to dismiss, the filings lay bare the extractive reality behind the “setting” that was never really optional.