Twitch has added an account setting allowing users to block Amazon from using their streams, clips, VODs, and chat logs to train generative artificial intelligence models. The toggle is active by default across accounts, prompting user backlash and raising transparency questions about data collection practices across the streaming platform.
Twitch has introduced a new account setting that allows users to prevent Amazon from using their channel content to train generative AI content models. The toggle, titled Training for Generative AI,
is located within the Security and Privacy tab of account settings. According to platform documentation, switching the toggle off excludes a user’s streams, VODs, clips, stream chats, and channel pictures and text from future model training.
The change brings formal visibility to a data pipeline that has operated automatically for years. While the company confirmed the opt-out option, the setting is enabled by default for all accounts, meaning creators who never visit their security settings are automatically contributing their broadcast data to Amazon’s technological development.
Default Opt-In Settings Draw User Backlash
The decision to make AI training active by default sparked immediate criticism from creators and viewers when streaming reporter Zach Bussey published screenshots of the hidden menu option. During a livestream addressing the backlash, Twitch chief product officer Mike Minton defended the default setting with candid remarks about participation rates.
Mike Minton, chief product officer at Twitch, stated that if the feature required opt-in, nobody would choose to participate, noting that was the honest answer.
Minton added that the company did not expect users to react favourably to the data collection policy. Alongside head of community Mary Kish, Minton walked viewers through the account menu to demonstrate how to disable the feature. Kish acknowledged during the broadcast that the collection of data for AI training has become an industry standard, noting that users are unlikely to feel excited about the practice.
The livestream chat filled with hundreds of critical comments. One viewer argued that nobody wants to feed AI systems with personal creativity and content, while another suggested that Twitch missed an opportunity to establish a protective industry standard against automated scraping.
What the Opt-Out Covers and What Remains Active
Twitch’s updated documentation draws distinct boundaries around what the opt-out toggle governs. Turning off the setting halts the use of channel material in future training of Amazon models designed to generate or synthesize text, audio, images, or video. However, the setting is not retroactive; content already harvested before a user flips the switch remains part of previously developed models.
Furthermore, the exclusion applies exclusively to generative training models. Other platform systems continue to process user content regardless of the toggle’s position. According to the platform’s FAQ documentation, automated features that operate the service itself remain fully active.
- Streamer growth and monetization tools, such as real-time sponsorship campaign assistance
- Viewer discovery systems and content recommendations
- Community safety mechanisms, including automated chat moderation via AutoMod and automated captioning
Platform representatives explained that features like AutoMod and caption generators process data to deliver immediate services rather than retaining content to build new generative systems. Minton and Kish noted that disabling safety features for individual accounts would weaken protections across the entire platform.
Chat Rules and Unresolved Questions on Historical Scraping
The opt-out control also introduces specific rules for community participation. Under the platform guidelines, chat messages follow the preferences of the channel owner rather than the individual chatter. An individual who disables AI training on their own account but participates in the chat of a streamer who has left the default setting enabled has no separate protection for those messages, permitting them to be harvested through the host channel.
Uncertainty remains regarding the historical extent of Amazon’s data harvesting. Minton admitted during the platform stream that he did not know whether user data had already been scraped for training prior to the update, stating he was unsure precisely what Amazon had used or omitted in past model development phases.
As tech giants compete aggressively in generative technology development, it remains unclear how long Amazon will maintain an opt-out mechanism for creators before platform policies evolve further.
