Amazon gave Twitch streamers an AI training opt-out, but buried it under default-on

Twitch data feeds video and text generation models

Twitch has quietly added a new "Training for Generative AI" toggle to its settings. Disabling it prevents Amazon from scraping user content-videos, clips, chat histories, images, and profile texts-into training datasets for future generative AI models dealing with text, audio, image, or video synthesis.

The opt-out applies only to future models

This toggle does not erase history. Years of past audio and text have likely already been digested by Amazon's models, and flipping the switch won't undo that. Twitch also explicitly notes that even after opting out, content will still be used to run existing AI-supported features on the platform, such as AutoMod or captioning. The block strictly applies to future generative training.

Chat participants are bound by the host channel's rules

The opt-out mechanism comes with a notable technical loophole regarding user interactions. The training block only covers a streamer's own channel. When a user participates in another streamer's chat, their own opt-out preference is overridden by the host's settings. If the host leaves training enabled, comments from all visiting users can still flow into the dataset regardless of their individual preferences.

Default opt-in guarantees the data pipeline remains full

The most critical detail isn't the toggle itself, but its initial state. The feature is enabled by default for all existing and new accounts. Because it is buried within the Security and Privacy tab and deployed without a mandatory notification, the vast majority of users will never know it exists, ensuring Amazon's data acquisition continues uninterrupted.

Lilith's verdict

Burying an opt-out toggle in the privacy settings and leaving it on by default is a calculated bet that 95% of people are too lazy to ever look for it. A corporate opt-out is really just an attention test in disguise.