- Published
Amazon is facing a lawsuit, external over its move to train its AI models on videos people broadcast on the popular streaming platform Twitch.
The move faced backlash from users when it was announced in August - with people having the option to opt-out if they do not want their data used this way.
The class action lawsuit has been brought on behalf of millions of streamers who use Twitch, alleging its parent company Amazon used their videos to train AI without their permission or proper compensation.
Twitch and Amazon have not commented on the lawsuit.
Streamers on the platform reportedly produced, external more than 215 million hours of content in the first few months of 2026 alone.
The lawsuit was brought by Connecticut-based streamer Warren Pandiscia.
It claims Twitch breached its contract with users by using that content as AI training data for Amazon's models without obtaining their permission.
It adds streamers should not have their work commercially exploited to develop AI without consent or compensation.
The lawsuit seeks damages for streamers whose content was allegedly used to train AI, as well as an order preventing Amazon from continuing the alleged practice.
It is not clear when Amazon began collecting Twitch users' data to train its AI models.
Twitch's chief product officer Mike Minton previously said he did not know if users' data had been scraped for training before it gave people the opportunity to opt-out.
He said he was unsure what Amazon "has done in terms of model training and what they've used and not used".
In Twitch's own FAQs about AI training, it said a person's audio might be used to "refine models that create speech to text".
It said this would help improve automatic subtitles on Twitch streams as well as Amazon videos.
Those that want to disable the feature would need to navigate to the Settings tab in their Streamer Dashboard, click the Security and Privacy tab, then scroll down to near the bottom of the options and toggle off "training for Generative AI".
However, the opt-out only applies to individual streams, meaning creators' content could still be used for AI training when they appear in a stream where the feature remains enabled.
Amazon, which bought Twitch for nearly $1bn (£740m) in 2014, has made AI a major focus as it competes with Google, Meta and other tech giants.
Sign up for our Tech Decoded newsletter to follow the world's top tech stories and trends. Outside the UK? Sign up here.
Related topics
- Published5 December 2025
Facts Only
* Amazon is facing a lawsuit regarding the use of livestreams from Twitch to train AI models.
* The class action was brought by Warren Pandiscia, a Connecticut-based streamer.
* The claim alleges Twitch breached its contract by using user content for AI training without permission.
* The suit asserts streamers should not have their work commercially exploited for AI development without consent or compensation.
* The lawsuit seeks damages and an order preventing Amazon from continuing the alleged practice.
* Streamers reportedly produced over 215 million hours of content in the first few months of 2026.
* Twitch's chief product officer indicated uncertainty about data scraping prior to the opt-out mechanism being available.
* Twitch FAQs mention audio may be used to refine speech-to-text models for subtitles on Twitch streams and Amazon videos.
* The opt-out setting in Streamer Dashboard only applies to individual streams, not all content usage.
Executive Summary
Amazon is facing a class action lawsuit from streamers alleging that Amazon used video content from the Twitch platform to train its AI models without obtaining consent or providing proper compensation. The lawsuit was initiated by Connecticut-based streamer Warren Pandiscia, claiming Twitch breached its contract with users by using this content for AI training without permission. The suit further asserts that streamers should not have their work commercially exploited for AI development without consent or compensation, seeking damages and an injunction against Amazon's practice.
Twitch's Chief Product Officer stated they were unaware if user data had been scraped for training before the opt-out feature was implemented, acknowledging uncertainty regarding what Amazon has specifically used versus not used in model training. Twitch's own FAQs mention that audio may be used to refine models for speech-to-text, which benefits subtitle improvements across Twitch and Amazon videos. While users can opt-out of AI training via their Streamer Dashboard settings, this option only applies to individual streams; content could still be used if the feature remains enabled on a specific stream.
Full Take
The narrative presents a tension between platform terms of service, user consent mechanisms, and the commercial realities of large-scale data ingestion for AI development. The core conflict lies in the discrepancy between user-facing opt-out options and the potential for passive, aggregate data use across platform ecosystems. The claim that opting out on an individual stream does not prevent broader exploitation suggests a systemic ambiguity in how consent is managed when content flows through multiple corporate entities (Twitch to Amazon).
The structure relies on framing data usage as inherently exploitative against individual creators, setting up a moral demand for compensation and control rather than simply debating the technical mechanisms of model training. The pattern involves leveraging the public awareness surrounding AI advancement to assert a principle of digital labor rights over content. The implication is that the architecture of platform agreements often prioritizes corporate development interests over explicit creator autonomy when data is commodified, creating an asymmetry where transparency is intentionally obscured through layered notices.
What mechanisms govern contractual obligations across such vast, multi-party platforms when the consent is provided via granular, stream-level toggles? What does the lack of clarity regarding the scope of what "used and not used" imply about data ownership in evolving media environments? If consent is technically offered but practically circumvented by default settings or aggregation, where should the locus of legal accountability reside for systemic exploitation?