The news
Twitch confirmed it has used content from livestreams to train generative AI models owned by its parent company Amazon. The practice ran with an opt-out setting that was disabled by default and placed deep inside the platform’s security options. Streamers and viewers reacted with immediate anger once the toggle surfaced.
Chief product officer Mike Minton stated during a company stream that the models had incorporated streamed video. He also said he did not know the full extent of user data that may have been included. Head of community Mary Kish joined the same stream to address the policy, yet the discussion left many questions open.
Context
Prior to the announcement, Twitch had given no public indication that live broadcasts were being routed into Amazon’s AI systems. The sudden appearance of an opt-out control, set to allow scraping unless manually changed, altered the prior assumption that user content stayed within the platform’s direct viewing and archiving functions.
The change affects every streamer whose broadcasts have run since the training began, an interval whose start date remains undisclosed. Viewers who appeared on camera or in chat logs during those streams are also swept into the same data pool without separate notice.
No start date for the training program has been released. The company has not said whether the practice began months or years earlier, leaving creators without a way to determine which archived videos may already sit inside Amazon’s datasets. The absence of any public timeline means both current and former streamers must assume their entire history on the platform could be involved.
Details
Minton described the use of streamed video in direct terms during the explanatory broadcast. He did not provide a timeline for when the practice started or a list of data categories that had been processed. The admission that he lacked visibility into the scope of captured user information stood as the clearest statement on record.
The opt-out control itself sits inside the security section of account settings rather than in privacy, content, or creator tools. Its default state permits continued use of new and archived streams for model training. No public dashboard or export tool has been offered to let creators review which of their past broadcasts entered the datasets.
No figures on the volume of streams processed or the number of accounts affected have been released. The company has not stated whether audio, chat text, or metadata beyond video frames were also used. The single source document notes that Minton was upfront about video ingestion yet conceded he did not know the full range of user data that may have been swept up.
Reactions / counterpoints
Streamers and viewers expressed immediate outrage once the buried toggle became known. The placement of the control inside security settings rather than creator or privacy sections drew particular criticism for making the default scraping hard to discover. Mary Kish’s participation in the follow-up stream did not resolve the core concerns about transparency or the lack of an opt-in mechanism.
The source article records that the explanatory stream itself failed to reassure many participants, with the only clear beneficiaries appearing to be the Amazon systems ingesting the data. No additional company statements have clarified whether the default setting will change or whether affected creators will receive any form of notice or compensation.
Why it matters
Creators who built audiences on Twitch now face the possibility that their work has already contributed to commercial AI systems without affirmative consent or compensation. The buried default setting and the executive admission of incomplete knowledge about ingested data both point to weak internal controls over how user-generated content is reused.
For anyone who streams or watches on the platform, the episode shows that participation carries an open-ended risk of secondary use by Amazon’s AI efforts. Until Twitch publishes a clear inventory of what data was taken and when, streamers lack the information needed to assess or limit that exposure. The episode also illustrates how a parent company’s broader AI ambitions can override platform norms around creator ownership without prior notice.
---
Sources:
{"sources":[{"publisher":"Rock Paper Shotgun","title":"Twitch are scraping streams for AI training by default, and even their own execs don’t know what user data may have been swept up","url":"https://www.rockpapershotgun.com/twitch-are-scraping-streams-for-ai-training-by-default-and-even-their-own-execs-dont-know-what-user-data-may-have-been-swept-up","published_at":"2026-08-13T10:01:24.000Z","summary":"Livestreaming giants Twitch have admitted using streamed content to train the generative AI models of parent company Amazon. While it’s not clear when the platform began feeding its users’ work into Jeffy B’s Planetkilling Machines, yesterday’s announcement of an opt-out toggle – buried deep within Twitch’s security settings, and set to enable AI stream-gobbling by default – sparked outrage among streamers and viewers that such scraping was occurring at all. Twitch’s chief product officer Mike Minton and head of community Mary Kish were wheeled out to explain the policy in a stream of their own, though it’s possible the only ones reassured by the discussion were the Amazon bots themselves. Although Minton was upfront about the models’ use of streamed video, he conceded that he didn’t know "}]}
No comments yet