Twitch to Let Amazon Train AI on Streams Unless Creators Opt Out

Twitch streamers now have their broadcasts feeding Amazon's generative AI models by default, with an opt-out buried in a settings menu rather than a checkbox they had to actively agree to. The platform confirmed the policy this week, drawing immediate anger from a creator base that had no say before the switch was flipped.

To stop their content from being used, streamers must go into channel settings, open the security and privacy tab, and manually toggle off "training for generative AI." Anyone who doesn't take that step is presumed to have consented.

Twitch's Head of Community, Mary Kish, and Chief Product Officer, Mike Minton, tried to get ahead of the backlash on a livestream that drew close to 3,000 viewers. The format itself underscored how unusual the moment was: executives fielding live, public pushback over a data policy rather than a product feature.

Minton didn't dress up the reasoning. Asked why the setting wasn't opt-in from the start, he said plainly that under those terms almost nobody would agree to it.

"If it were opt-in, nobody would opt in."

That admission landed as confirmation of what many creators already suspected: the default was chosen because it maximizes the pool of training data, not because it reflects what streamers would actually choose. Minton also said he couldn't confirm whether Amazon had already trained on Twitch broadcasts recorded before the policy was made public, leaving open the question of how far back the practice may extend.

Not an isolated case

Twitch isn't alone in defaulting users into AI training. Meta already draws on public Facebook and Instagram content to train its models, and a genuine opt-out is mostly limited to users in the U.K., where regulatory pressure has forced the company's hand.

The pattern across both platforms suggests opt-out defaults have become the industry's preferred path of least resistance: technically offering choice while structuring it so most users never exercise it. For Twitch specifically, the stakes are personal in a way they aren't on more passive social feeds. Streamers' voices, faces, mannerisms and improvised commentary are exactly the kind of material generative video and voice models are hungry for, and many creators view that footage as their professional identity rather than incidental public data.

Whether the backlash forces Twitch to reverse course, as some U.K. Meta users effectively did through regulation, remains an open question the company has not addressed beyond the livestream's damage control.