Twitch announced that content from its platform — including livestreams, VODs, clips, stream chats, and channel images and text — will be used to train Amazon's generative AI models, with creators opted in by default2,3,7. Users who do not want their content included must manually opt out through settings at www.twitch.tv/settings/security.
The policy change, announced on August 12, drew immediate and concentrated backlash from the Twitch community. In an updated support page, Twitch confirmed that opting out prevents a user's streams, VODs, clips, stream chats, and channel pictures and text from being used in "future training of a model developed by Amazon whose purpose is to generate or synthesize text, audio, images, or video". Other AI-supported features such as captions and safety tools will continue to function for users who opt out of generative AI training.
A notable limitation applies to chat: if a user participates in chat on another creator's stream, the stream owner's opt-out preferences govern whether that chat content can be used for training.
Twitch Head of Community Mary Kish and Chief Product Officer Mike Minton addressed the backlash in a livestream on the official Twitch channel, facing a live audience of nearly 3,000 users, many of whom posted anti-AI sentiments in chat. Minton was direct about the opt-out-by-default design: "Why is it not opt-in? That's what everybody is spamming in chat. I get it. Let me opt in versus making me opt out. Well, there's an honest answer — If this was opt-in, nobody would opt in. That's honestly the answer," Minton said.
The announcement formalized a practice with longer roots. A company executive confirmed more than two years ago that Amazon was already using Twitch content for AI training. The new policy adds a user-facing opt-out mechanism where none previously existed.
For Amazon, the stream recordings represent a large reservoir of audio and video content for training AI models. Twitch users have raised concerns that many creators, who often record themselves live for many hours per week, may surrender their content to Amazon's AI training pipeline without realizing it. The Twitch community has been broadly opposed to generative AI use, given that prevalent generative AI products are trained on creative materials scraped from the internet.
ANALYSIS The opt-out-by-default structure maximizes the volume of training data Amazon can collect from Twitch, as Minton's own remarks make explicit. The chat-level limitation — where a stream owner's preferences override those of individual chat participants — means that even users who opt out may still have their chat contributions used for training if the channel they are chatting in has not opted out.
The addition of an opt-out toggle formalizes user control that did not exist during the prior period of AI training on Twitch content, but the default-on design ensures the practical barrier to data collection remains low.