Amazon Trains Generative AI Models on Twitch Streamer Content by Default

Twitch accounts are now automatically enrolled in a data-sharing program that feeds creator content directly into parent company Amazon’s artificial intelligence systems. The automated inclusion sweeps up thousands of hours of video and audio from livestreams, potentially capturing creators who remain unaware of the background change.

The update, rolled out without advance warning, targets a broad spectrum of channel material. According to platform documentation, the data harvested for model development includes streams, VODs, clips, stream chats, and pictures and text on a channel. The models being trained are designed to generate or synthesize text, audio, images, or video.

Mary Kish and Mike Minton Address User Backlash on Livestream

Platform executives faced immediate hostility during an official broadcast addressing the update. Twitch Head of Community Mary Kish and Chief Product Officer Mike Minton spoke to a live audience of nearly 3,000 users whose chat messages flooded the stream with anti-sentiments. When viewers questioned why the program was not designed with an opt-in framework, Minton offered a blunt explanation.

“If this was opt-in, nobody would opt in. That’s honestly the answer.”

Mike Minton, Chief Product Officer at Twitch, via TechCrunch

We don’t expect you to be happy or excited about this, Kish said, adding I don’t expect anyone to react to this favorably.

High-profile creators immediately took to social media to urge followers to disable the feature. Twitch user Legundo posted a warning on X urging the community: EVERYONE make sure to turn this off and tell Twitch we don’t want this crap!

How to Find and Disable the Training for Generative AI Toggle

For creators wishing to block their broadcasts from feeding Amazon’s models, the opt-out mechanism is located deep within account controls. Users must navigate to their general account settings rather than the creator dashboard. Selecting the security and privacy tab reveals a toggle labeled Training for Generative AI positioned near the bottom of the page.

Turning off the switch stops future data collection, but it leaves several distinct platform features active. Twitch confirmed that disabling generative AI training does not shut down all machine learning tools on the service. Automated safety systems like AutoMod, caption generation for clips, and recommendation algorithms continue to operate because they do not retain channel data to train standalone generative models.

Chat participation also carries specific rules regarding consent. According to platform documentation, the channel owner’s opt-out preference governs whether chat messages can be used for training. An individual who disables the setting on their own channel but chats on a broadcast that remains opted in has no separate protection for those messages.

Years of Silent Data Harvesting and Unanswered Questions

While the account toggle is new, Amazon’s utilization of live video for development purposes has a longer history. During an event hosted by The Information in 2024, Minton confirmed that Amazon was using streaming material for artificial intelligence prototyping. However, executives during the recent announcement could not clarify the exact scope of past collection efforts.

When questioned about whether broadcasts had already been fed to the parent company prior to the opt-out rollout, Minton admitted uncertainty. I don’t actually know the answer to that question because I don’t know what Amazon has done in terms of model training, and what they’ve used and not used, Minton said, adding that he was not responsible for Amazon’s training efforts.

The platform’s support documents emphasize that the new control applies strictly to future training, leaving open questions about the volume of material already harvested since Amazon acquired Twitch in 2014.