Twitch has started allowing content from creator channels to be used to train Amazon’s generative AI models. The setting is turned on by default, so creators who do not want their material included must disable the option in their account security settings.
What the new setting permits
The platform’s new configuration authorizes the use of channel content to “train models of generative AI at Amazon.” That scope can include live broadcasts, saved videos (VODs), clips, chat messages, images and written texts tied to a channel.
Twitch cited examples such as using stream audio to improve speech-to-text models — which could enhance the platform’s automatic captions — but noted the training could also apply to Amazon services beyond Twitch. The company said the allowed data types are not limited to audio.
Opt-out model and Twitch’s explanation
The change uses an opt-out approach: creators are automatically included unless they manually turn the setting off. During a broadcast addressing the update, Twitch’s head of product, Mike Minton, said the company chose this route because he believed that requiring explicit opt-in would lead to very low participation.
“If it were opt-in, nobody would opt in,” Minton said.
As a result, the option remains enabled by default.
How chat messages are treated
Twitch noted a particular detail about chat: even if a creator disables training for their own account, messages they send in another channel may still be included in training data when that other channel has the option enabled. In other words, participation depends on the channel where content is posted.
How to opt out
Creators who do not want their channel content used for Amazon’s generative AI training can disable the option in their account security settings. Twitch’s stated path to the control is:
Twitch > Settings > Security > Generative AI training
Turning off this toggle prevents the channel’s content from being used for that training. Twitch warned, however, that disabling the setting does not remove or stop the platform’s other AI-based features; tools for ad selection, chat moderation and other automated systems will continue to operate.
Company comments and internal choices
During the discussion stream, Mary Kish, Twitch’s head of community, said that some forms of AI are already embedded in Twitch’s operations and that she does not believe the platform can function without certain AI applications. Kish also disclosed that she personally turned off the option to allow her content to be used for training.
Minton said that, if he were a streamer, he would leave the option enabled.
Criticism over rollout and transparency
Users criticized how the change was implemented, noting the feature appeared in settings without broad prior notice to creators. Executives demonstrated during the broadcast where the setting could be found after complaints that it was hidden within account options.
Kish acknowledged the criticism about the lack of an email announcement, arguing that creators do not always read email and that direct messages and community dissemination were other ways to share the update. Twitch also said it currently does not plan to publicly identify which channels have allowed or refused AI training.
Broader debate for platforms and creators
The decision places Twitch amid a larger discussion about whether platforms can use user-generated content to develop AI models. For Twitch specifically, the debate is heightened by the variety and volume of data a single live stream can contain: hours of video and audio, chat interactions, images, text and community-produced materials.
For now, channel owners retain the option to prevent their own content from being used directly for Amazon’s generative AI training, but the default opt-out design has drawn criticism from creators and viewers alike.
