Twitch has begun using creator content to help train generative AI models belonging to its parent company, Amazon, with creators automatically included in this use unless they manually opt out. The decision prompted a swift and widespread reaction from the Twitch community, particularly because the platform chose an opt-out model rather than requesting prior consent.
Live-stream recordings are an important source for training AI models, providing thousands of hours of audio and video content. However, Twitch users fear that their content could be used to train Amazon’s models without their knowledge, especially since creators record lengthy streams of themselves and their voices for several hours each week.
Twitch Responds to User Objections
During a livestream on the platform’s official channel, Twitch Head of Community Mary Kish and Executive Product Director Mike Minton spoke before an audience of nearly 3,000 angry users, while the chat filled with comments opposing generative AI.
When Minton was asked why participation had not been made optional from the outset, he said the candid answer was that if use were based on an opt-in system, no one would agree to it. He added that this was why the platform adopted an opt-out model.
Twitch understands that a significant part of its community opposes the use of generative AI, particularly because popular AI products are trained on books, images, videos, and other materials collected from the internet without their owners’ consent.
Uncertainty About Previously Used Content
Twitch presented the change as the addition of a setting allowing users to opt out of using their channel content to train generative AI content models through Amazon, rather than directly announcing that it had begun training models on user content. This left some creators confused about whether their recordings had already been sent to Amazon.
When Minton was asked whether user content had previously been used for training, he said he did not know the answer because he did not know what Amazon had done regarding model training, or which content it had and had not used.
Kish noted that Twitch is not the only platform that uses public content to train AI models. She said Meta uses public content from its platforms to train its models, meaning that public account data on Facebook and Instagram may already have been used for this purpose. She added that users in the United Kingdom can opt out of Meta’s training, while elsewhere this option is available only by making accounts private, an impractical step for creators who generate income from their accounts.
How to Opt Out
To opt out of AI training on Twitch, users should go to Channel Settings, not the Creator Dashboard, then open the Safety and Privacy tab, scroll to the “Generative AI Training” option, and disable it.
Kish said that providing an opt-out option reflects Twitch’s response to the voice of its community, which does not want generative AI models trained on its content.