The news
Twitch announced on August 12 that it will use streams, VODs, clips, chats, text, and images to train Amazon’s generative AI models by default. Streamers must visit their account settings and turn off the “Training for Generative AI” toggle under Security and Privacy to stop the practice. The change applies only to future training runs.
Context
Before the update, Twitch had not publicly stated a default position on using creator material for large-scale model training. The new setting makes participation automatic and places the burden on individual users to withdraw. Twitch owns the platform and operates under Amazon, which maintains several generative AI projects that require large volumes of video, audio, and chat data. The policy covers any Amazon model whose purpose is to generate or synthesize text, audio, images, or video.
Details
Opting out blocks the listed content types from “future training” of any Amazon model whose purpose is to generate or synthesize text, audio, images, or video. Captions, safety tools, and other AI-supported features continue to operate regardless of the choice. When a user chats in another streamer’s channel, that streamer’s opt-out preference determines whether the messages can be used.
Twitch chief product officer Mike Minton addressed the decision during a livestream. “If this was opt-in, nobody would opt in,” he said. “That’s honestly the answer.” The support documentation confirms the same scope: streams, VODs, clips, highlights, chat, text, and images on a channel. The setting appears in the same menu section that already houses privacy controls. Once disabled, the preference covers all listed content types going forward; previously ingested material is not addressed in the published materials.
The announcement came through Twitch support pages and was reported across multiple outlets on the same day. Users must navigate to account settings, locate the Security and Privacy section, and flip the single toggle labeled “Training for Generative AI.” No additional confirmation steps or granular controls for specific content types are described. The policy applies uniformly across all channels, including those of partners and affiliates.
Reactions
No other company statements or streamer responses appear in the available reports. The single on-record comment comes from Minton, who framed the default as necessary to gather sufficient data. Coverage from TechCrunch, The Verge, and AppleInsider all cite the same support documentation and the same Minton quote without additional context from Amazon executives or creator representatives.
Why it matters
An opt-out default guarantees high participation rates while shifting responsibility onto creators who must discover and change the setting. Because chat messages inherit the host channel’s preference, the policy also affects viewers who may not realize their text is being collected under someone else’s choice. For streamers who treat their archives and live interactions as core assets, the change removes the prior assumption that content stays within the Twitch service boundary. The practical result is that Amazon gains a steady supply of conversational video and chat data unless thousands of individuals act separately to withhold it.
Amazon’s broader AI efforts already include models that synthesize video and audio, and Twitch represents one of the largest public sources of unscripted, real-time human interaction on the internet. The volume of data generated daily on the platform—live commentary, real-time chat, viewer reactions, and archived broadcasts—provides training material that is difficult to replicate through licensed datasets or synthetic generation. By making inclusion the default, Twitch ensures that even casual or infrequent streamers contribute unless they actively intervene. This approach aligns with patterns seen in other large platforms where data collection for model improvement defaults to “on” and requires explicit user action to reverse. Streamers who later discover the setting and opt out will still have contributed their earlier content to any training runs that occurred before the change. The absence of retroactive deletion or compensation mechanisms means the policy creates a one-way flow of value from individual creators to Amazon’s model development pipeline.
---
Sources:
{"word_count": 712, "sources_used": 3, "expanded_sections": ["context", "details", "why_it_matters"]}
No comments yet