Twitch has offered creators a way to keep their work out of Amazon’s generative-AI training—but only after placing them in by default. That choice has turned a new privacy control into a fresh fight over consent.

Amazon bought Twitch in 2014, and the question of whether the platform’s vast archive of video, audio and chat was feeding AI models had been hanging over creators for years. At a 2024 event, then-chief monetization officer Mike Minton answered, “Yeah, for sure,” when asked whether Amazon used Twitch material for AI training, while qualifying it with user trust and privacy rules.¹

This week, Twitch made the policy explicit: streams, VODs, clips, chats and channel images or text may be used in future training of Amazon models unless a creator switches off the setting in Security and Privacy.¹ Twitch says training could improve speech-to-text captions on its service and elsewhere across Amazon.

The announcement landed badly. Twitch community chief Mary Kish told viewers, “We don’t expect you to be happy or excited about this,” as the livestream chat filled with anti-AI messages.² For critics, the issue is not merely that Amazon wants valuable hours of creators’ voices and video; it is that many may never notice the toggle, while the resulting systems could compete with the people whose work trained them.

Minton’s explanation of the default sharpened that anger: “If this was opt-in, nobody would opt in. That’s honestly the answer.”³ Twitch argues the control itself reflects community pressure and is more protective than the norms governing much publicly available online material.²

But the company has not resolved the central transparency question. Asked whether Amazon had already trained on Twitch livestreams, Minton said he did not know what Amazon had used and was not responsible for its training efforts.² Creators now have a switch; they still lack a clear accounting of what happened before it existed.