Twitch confirms default AI training opt-in, sparking creator backlash
Chief Product Officer Mike Minton defends the policy, noting competitors like Google and Meta offer no such choice, while warning that external scraping remains a reality for public content.
Twitch, the Amazon-owned live streaming platform, has confirmed it will utilise streamers’ content to train generative artificial intelligence models, automatically enrolling all users in the programme by default. The announcement has triggered significant dissent within the creator community, with thousands of users demanding the feature be implemented on an opt-in basis rather than an opt-out one.
In response to the outcry, the platform has introduced a toggle in account settings that allows users to manually exclude their data from generative AI training. The setting covers streams, including on-demand versions, clips, stream chats, and images. Users can access this preference via Settings > Security and Privacy on the web interface, or under Profile > Settings > Security and Privacy in the mobile application.
Twitch’s Chief Product Officer Mike Minton stated that the default opt-in configuration was chosen because the company anticipated minimal voluntary participation. “If it was opt in, nobody would opt in. That's honestly the answer,” Minton said during a recent Patch Notes stream. He emphasised that the primary distinction of Twitch’s approach is the respect for user wishes to opt out, a feature he noted is absent from many other major services.
The exclusion applies strictly to generative AI training and does not prevent the use of data for other machine learning functions essential to the platform’s operation. Minton explained that systems such as AutoMod and automated captions rely on AI models to evaluate text and transcribe audio, noting that these features are necessary for the service to function. Data utilised for these purposes is not retained for training generative models, and Minton assured creators that opting out will not impact their discoverability or reach on the platform.
Minton also drew comparisons to industry peers, highlighting that Google and Meta utilise user content from YouTube, Facebook, and Instagram for AI training without offering opt-out mechanisms. While Google uses only a subset of its catalog, neither platform provides a direct exclusion tool for creators. Minton further cautioned that external AI laboratories may already be scraping publicly available Twitch content for training purposes, suggesting that platform-level controls offer limited protection against broader data harvesting.
