Twitch has handed creators a switch — but not the choice they wanted. Amazon’s streaming platform is allowing users to opt out of AI training on their broadcasts, confirming a practice that many streamers say should have required permission before it began.

Amazon bought Twitch in 2014, and as the company races to expand its AI infrastructure, the livestreaming service has become part of the wider argument over what counts as usable public data.

On Wednesday, Twitch introduced a privacy-setting control that lets streamers block Amazon’s AI models from training on their content. The key friction is the default: creators are automatically included unless they find the setting and turn it off.¹

The announcement was also the first clear confirmation that Amazon’s models were being trained on Twitch livestreams. Mary Kish, Twitch’s head of community, anticipated the backlash in a company livestream: “We don't expect you to be happy or excited about this.”² The chat quickly filled with anti-AI messages, while prominent streamers urged followers to disable the option.

Twitch executives presented the control as an unusual safeguard in an internet increasingly treated as AI training material. Chief product officer Mike Minton argued that “almost any publicly available content is being used for model training in one form or another, whether permitted or not.”² But that defense did little to answer creators’ central complaint: a privacy toggle is not the same as informed consent.

Minton also said he could not say whether Amazon had previously used Twitch broadcasts for model training, stressing that he was not responsible for Amazon’s efforts. Amazon did not respond to a request for comment. Twitch says it already uses AI for captions, clips and content classification, though not generative features. For streamers, the new setting has sharpened rather than settled the question of who controls the value in a live broadcast.