Twitch will utilize your streams to train Amazon's AI unless you locate the setting.

      Twitch has introduced a setting that prevents Amazon from using streamers' content, including streams, clips, and chat interactions, for training generative AI. This setting is off by default, implying that the training continues unless users actively opt out. Chief Product Officer Mike Minton explained this default setup, stating, “If it was opt-in, nobody would opt in.”

      During a recent livestream with Community Head Mary Kish, Minton responded to streamer Zach Bussey's discovery of this toggle, which gained attention after a screenshot circulated. He expressed his comfort with AI training and indicated that he would not opt out if he were a streamer.

      The content covered under this setting includes almost everything: streams, previous broadcasts, clips, chat messages, and text and images on channel pages. Minton provided an example where a streamer's voice could be utilized to enhance speech-to-text models, benefiting captions both on Twitch and in Amazon's AI offerings.

      However, the setting has limitations when it comes to chat—while it protects content on a user's channel, messages sent in other streamers' chats can still be collected if those streamers have not opted out. Such gaps have previously led creators to explore other platforms designed specifically for human artists.

      It's worth noting that while the toggle is new, the practice of using content for AI training is not. When asked if their videos had already been utilized, Minton stated he was unsure about Amazon's prior training activities, although reports suggest the platform has contributed content to AI training for a few years.

      Additionally, opting out only halts the training aspect; AI continues to operate on Twitch for purposes like chat moderation, ad selection, and recommendations, and this change does not impact that functionality.

      This setting can be found at twitch.tv/settings/security. When questioned during the stream about why Twitch hadn’t merely informed users via email, Kish remarked that not everyone checks their emails.

      Choosing a default setting that users would reject aligns with the textbook definition of a dark pattern. What is noteworthy in this case is that an executive candidly explained the rationale behind it without embellishment.

      There is also no information in the coverage regarding how this default setting operates in Europe. This is significant, as opt-out consent for AI training has faced challenges in that region before, and Meta had to pause a similar approach in 2024 until regulators permitted its continuation.