Amazon will train on Twitch streamers' content by default, unless they opt out | TechCrunch

Amazon has introduced a new policy that permits the company to use video and audio recordings from Twitch streamers to train its generative artificial intelligence models.

This training is enabled by default for all content creators on the platform, triggering widespread criticism from the streaming community. Twitch executives defended the opt-out structure during a live broadcast, acknowledging that a voluntary opt-in system would yield insufficient data because most creators would choose not to participate. Streamers have expressed deep concern over the automated collection of their likenesses and voices, especially given the extensive hours they spend broadcasting live each week.

The policy change was presented by Twitch as an update granting users a new control setting to opt out of AI training, rather than an announcement of the training program itself. This framing led to confusion among creators regarding whether their past broadcasts had already been utilized to train Amazon's models, a detail that Twitch representatives could not definitively clarify. While Twitch leadership pointed out that other major technology companies, such as Meta, employ similar default data-harvesting practices on public accounts, they framed the introduction of an opt-out toggle as a concession to community feedback. Creators wishing to protect their content must manually locate and disable the training permission within their security and privacy account settings.