A creator who refuses training still supplies the material behind Twitch's AI recommendations, sponsorship help and AutoMod. Training use had been on by default, a practice that surfaced only when the toggle appeared.
Twitch streamers can now block what they publish from training the AI models of Amazon, which owns the platform. The control is a toggle marked Generative AI Training, filed under security and privacy in account settings on Twitch's website and on its phone app, and it covers model training and nothing else.
The company presents the setting as an acknowledgment that creators should be the ones choosing what happens to the material they make. The wording printed beside the toggle is narrower: switching the option off still leaves Twitch and Amazon free to draw on a channel's content for the other uses laid out in Twitch's privacy notice. Those cover platform features powered by AI — live help with sponsorship campaigns, recommendations that steer viewers to channels they might watch, and community safety tools such as AutoMod.
It is the default that drew the anger. More than 16,000 creators registered objections on a forum set up for the subject, and the protest flared after a livestream in which Twitch's head of community, Mary Kish, laid out the thinking behind the changes and granted that creators would react badly to them. Mike Minton, who runs product at Twitch, described his own answer as a frank one: the AI-training setting had to be switched on from the start, he said, because nobody would take part in it otherwise.
Minton said data arrangements of this kind exist well beyond Twitch, and that in his view it is fair to assume nearly all content posted publicly ends up feeding AI models one way or another, whether or not permission was given. He added that he cannot be sure of this, and that a great deal of it lies outside Twitch's own control.
Since March 2024, the service terms users accept on Twitch have obliged them to give the company, and anyone it licenses in turn, permission to use, copy, alter, adapt and circulate their content and to make derivative works from it. Before the new toggle appeared, nothing in the platform's rules said outright that material from creators might feed the training of generative AI models.
What none of this settles is the past. It has not been established when Twitch began drawing on what streamers publish, broadcast live and upload as video to train Amazon's systems, or whether Amazon's business partners are drawing on the same data.
Cover image: “PAX South 2016 - Twitch sign (24354667699)” by Daniel Benavides from Austin, TX, Wikimedia Commons, CC BY 2.0, cropped, resized, re-encoded.
