Twitch has introduced a default-on setting that allows channel content to train generative-AI content models at Amazon. Eligible material includes livestreams, videos on demand, clips, highlights, channel metadata, text and images—and stream chat.
Creators can turn the setting off under Security and Privacy. But the control exposes a deeper consent problem: when someone posts in another creator's chat, the channel owner's preference determines whether that message can be used for training.
Why it matters
AI training consent is becoming a product-interface decision. A default setting, its placement and the identity of the person allowed to change it can determine how years of creator work—and audience participation—are reused.
For platforms, contractual permission is not necessarily the same as informed consent. A setting can be technically available while remaining practically invisible to many of the people whose contributions it affects.
The audience-chat issue makes the design challenge especially clear. Platforms built around collaborative participation cannot always reduce consent to the preference of a single account owner. More granular controls, visible channel status and clearer notices could become meaningful trust signals for creator communities.
As platforms compete for creators and viewers, transparent AI-data controls are likely to become part of the product itself—not merely a paragraph in a privacy notice.