Twitch has allowed Amazon to use streamers' content — including audio, video, chat and images — to train its generative AI models, with the data-sharing setting turned on by default. Users can opt out, but Twitch's chief product officer Mike Minton admitted on a livestream that the setting is default-on because "if it's opt-in, nobody would opt-in."
Some users report the setting reverting to on after they've turned it off. Twitch says the collected data won't be resold to other companies, and gave examples like using streamers' audio to improve automatic subtitles, but did not clarify when the collection began or how much data has already been used for training.
Twitch's head of community, Mary Kish, told viewers during the livestream that she didn't expect users to be happy about the policy, acknowledging that data collection for AI training has become industry standard. The announcement drew heavy criticism in the stream's chat and from users concerned about implications for game developers whose titles are played on the platform.
The full dispatch is available from the source below.