Amazon's Twitch streaming platform has introduced a control allowing creators to prevent their channel content from being used to train generative AI models, but the setting is enabled by default, meaning content is automatically shared for AI training unless streamers manually opt out. According to The Register's August 13, 2026 report, Twitch chief product officer Mike Minton candidly explained the company's decision to require streamers to actively disable the feature rather than seek their permission first.
The new "Training for Generative AI" control applies to all channel material, including live broadcasts, archived videos, clips, highlights, text content, images, and chat conversations. When the setting remains enabled, Twitch may feed this content into Amazon's generative AI systems, with the trained models potentially deployed across Amazon's broader ecosystem beyond just the streaming platform. Stream audio, for instance, could refine speech-to-text capabilities that improve automated captions on Twitch and throughout Amazon's services. The control doesn't prevent chat messages posted in other creators' channels from being harvested—only the channel owner's setting determines whether their chat can be used for training purposes.
During a livestream addressing the policy changes, Minton offered an unusually blunt rationale for the default-on approach: "If it was opt-in, nobody would opt in. That's honestly the answer. So it's going to be on by default." The report notes that Minton had already confirmed at a 2024 event hosted by The Information that Amazon was using Twitch material to build AI models. The new control doesn't mark the start of this practice—it simply gives streamers a dedicated method to halt future use of their content. Twitch's language carefully specifies that opting out applies only to "future" generative AI model improvements, with the company offering no indication that disabling the setting removes material already incorporated into existing models.
The default-on configuration reflects a calculated trade-off: Amazon gains access to a vast trove of training data without needing to secure individual permissions, while streamers who want control must discover and toggle the buried setting themselves. The report indicates that The Register has asked Amazon which AI models have been trained on Twitch material, how long the practice has continued, whether the opt-out affects previously harvested content, and whether Twitch-trained models are sold to customers or third parties, though no responses were provided. Streamers now face a choice between passively contributing their creative work to Amazon's AI infrastructure or actively hunting down the switch the company would clearly prefer stayed untouched. The platform's candid admission that consent wouldn't work reveals how AI training economics increasingly depend on inertia rather than agreement, leaving creators to police their own boundaries in settings menus rather than being asked for permission upfront.