Twitch is harvesting streamer data for Amazon AI training — and it's opt-out, not opt-in

Twitch has quietly rolled out a generative AI training program that feeds streamer content to Amazon's AI models, and every account is enrolled by default. The feature was revealed not through a blog post or direct announcement, but via a Twitch Support post on X about a related opt-out setting — a low-visibility move that left most streamers in the dark.

Twitch chief product officer Mike Minton confirmed the system during a livestream and gave a remarkably candid reason for making it opt-out rather than opt-in: "If it was opt-in, nobody would opt in." The platform's own FAQ spells out what the training covers — streamer audio, for example, can be used to refine speech-to-text models that benefit not just Twitch but Amazon broadly. Opting out is straightforward: go to Settings, open the Security and Privacy tab, scroll to the bottom, and toggle off "Training for Generative AI." The catch is that most streamers won't know to look. A Twitch User Voice thread calling for all AI features to be opt-out by default — originally posted back in May — has ballooned to 228 pages and nearly 14,000 votes in favor.

Twitch is owned by Amazon, so the data pipeline here runs directly from streamers to one of the world's largest AI operations. A separate post from Twitch's own account muddied things further by claiming "Twitch is not training GenAI models on streamer content" — a statement that contradicts the FAQ, and which appears to be a technical dodge: Amazon trains the models, not Twitch itself, even though Twitch is Amazon. The distinction is meaningless in practice.

Community reaction has been overwhelmingly negative. Replies to the Twitch Support post on X are a mix of opt-out instructions and expletives, and streamer Zach Bussey, who flagged the program on Bluesky back in July, has continued pushing for transparency. The consensus from creators is blunt: Twitch knows they don't want this, and did it anyway.