Streaming platform Twitch has added a setting giving users the option to allow their streams to be used to train AI models, but it is turned on by default, Ars Technica has reported.

The Amazon-owned company has added a “Training for Generative AI” control to channel settings, allowing streamers to prevent their content being used to train Amazon’s future models.

If the setting is enabled, channel content including livestreams, videos on demand, images, and chat messages can be used to improve the company’s models, Twitch said.

For example, Twitch says, audio from streams could be used to enhance text-to-speech models to improve captions “across Amazon”.

In a livestream discussing the changes, Twitch’s chief product officer Mike Minton reportedly explained that the firm had set the feature on by default because “[i]f it was opt-in, nobody would opt in”.

“That's honestly the answer. So it's going to be on by default,” he added.

Amazon has been using Twitch data to train models prior to this feature. Ars Technica reported that Minton had confirmed at a 2024 event held by news publication The Information that data from the streaming site was sometimes used for this purpose. He added: “I mean, I think obviously within the bounds of user trust within the bounds of privacy regulations, which vary all around the world, but of course we have a role to play in that.”

Ars Technica reached out to Amazon for comment on how long it has used Twitch content for teaching but not yet received a response.

---