Twitch, the livestreaming platform owned by Amazon, introduced a setting allowing creators to prevent their channel content being used to train Amazon’s generative AI models. The setting was enabled by default, meaning creators had to opt out rather than explicitly agree. Twitch says material potentially covered includes livestreams, video-on-demand recordings, clips, highlights, stream chat, and text or images displayed on a channel. It says such data may help improve models that generate or synthesise text, audio, images or video; one example given is improving speech-to-text captioning across Twitch and Amazon.

The change prompted criticism from streamers and other users, centred on consent, transparency and the default choice. During an official Twitch broadcast, chief product officer Mike Minton acknowledged that an opt-in approach would probably attract few participants, saying that was the reason for the opt-out design. Twitch distinguishes this training from some operational or safety-focused AI features, such as moderation tools, which the new setting does not disable. The policy also states that opting out prevents future training on a channel’s covered content, rather than necessarily addressing any previous use, leaving some creators seeking greater clarity about how their material has been handled.