Twitch has introduced a control allowing streamers to prevent their content from being used to train Amazon's generative AI models. The setting is enabled by default, meaning all stream material—including live broadcasts, videos, clips, highlights, text, images, and chat—feeds into Amazon's AI development unless creators manually disable it.

Amazon's AI can use audio from streams to improve speech-to-text and captioning across Twitch and other Amazon services. The company confirmed it has been using Twitch data for AI training since at least 2024; the new control marks the first time creators have a direct opt-out mechanism.

Mike Minton, Twitch's chief product officer, justified the default-enabled approach with blunt candor: "If it was opt-in, nobody would opt in. That's honestly the answer. So it's going to be on by default."

Streamers should note two constraints. First, only the channel owner's setting controls whether chat messages are used for training—individual viewers cannot protect their own chat contributions on channels where the owner leaves training enabled. Second, opting out applies only to future model improvements; Amazon has not stated whether previously collected material will be removed from existing models or clarified which of its products rely on Twitch-trained AI.

The Register sought clarification on which Amazon models have incorporated Twitch content, the duration of data collection, and whether retroactive opt-out has any effect, but Amazon had not responded as of publication.