When a creator starts a stream on Twitch, their hours of recordings now help train Amazon's generative AI by default. Anyone who disagrees has to find the switch in the settings that Twitch added on 12 August themselves.

What changed in the settings

Twitch added a “Training for Generative AI” switch to the Security and Privacy section. If a creator leaves it on, Amazon can use their streams, recordings, clips, chat, text and images to train future models. The switch is on by default, so content goes into training until the creator turns it off manually.

Live streams and uncertainty over what has already been used

“If this was opt-in, nobody would opt in. That's honestly the answer,” said Twitch's head of product Mike Minton, when the company addressed the new settings on its official channel before nearly 3,000 viewers. Many of them posted rejecting messages in the chat. Alongside Minton, Mary Kish, head of community, also spoke.

Minton also admitted he is not sure what Amazon used for training from the content before the announcement. Parts of the community also object that the switch is found in account settings, not in the creator dashboard.

The setting shifts default consent to the platform. The creator supplies the performance, Amazon, as Twitch's owner, holds the recording and the data on which it builds its models. The switch exists, but the creator has to use it themselves.

For creators, this changes what it means to upload a stream. Alongside viewers and possible monetisation, it is now also input data for the models of the company that owns the platform. The default answer is yes.

By Tereza Tiller