Twitch and Amazon face legal action over using livestreams to train AI
Developed over time first seen 2 months ago
Amazon and its subsidiary Twitch are facing a class action lawsuit over the use of streamers' broadcasts to train Amazon's AI models without explicit permission or compensation. The case, filed by Connecticut-based streamer Warren Pandiscia on behalf of millions of Twitch users, alleges breach of contract and unauthorised commercial exploitation of their work, and seeks both damages and a court order to stop the practice. It follows a backlash in August, when Twitch introduced an opt-out option for AI training but it emerged this only applies stream-by-stream, meaning creators can still have their content used when appearing on others' streams that have not disabled the feature.
Twitch reportedly hosted over 215 million hours of streamed content in the first few months of 2026 alone, underlining the scale of material potentially drawn on. Twitch's chief product officer, Mike Minton, has said he cannot confirm whether user data was scraped for training before the opt-out was introduced, and the company's own FAQ acknowledges that streamers' audio may be used to refine speech-to-text models that improve subtitles on Twitch and Amazon videos. Neither Twitch nor Amazon, which bought the platform for close to $1bn in 2014 and has made AI central to its competition with Google and Meta, has commented publicly on the lawsuit.
- Twitch streamers suing Amazon over unauthorised AI training on their content
- Opt-out feature only works per-stream, not across all appearances
- Amazon and Twitch have not commented on the lawsuit
New here? Start with this
Twitch is a live-streaming platform, mostly used for video games and other real-time broadcasts, that Amazon has owned since 2014. In August 2026, Amazon revealed plans to use content from Twitch livestreams to train its artificial intelligence models, prompting a strong reaction from streamers who had not agreed to have their broadcasts used this way. A class action lawsuit has now been filed against both companies on behalf of millions of Twitch users, arguing this breaches Twitch's agreements with its creators and uses their work commercially without permission or payment.
Central to the dispute is how Twitch responded to the backlash: it added an opt-out option in streamers' account settings, but this only covers a person's own broadcasts, not appearances on other people's streams, so people can still end up in AI training data without directly agreeing to it. It is also unclear exactly when Amazon began using this footage, with even a senior Twitch executive saying he does not know what data has already been used.
The case is significant beyond Twitch itself, as it touches on a wider debate over how technology companies use content created by ordinary users to build AI systems, and whether after-the-fact opt-outs offer adequate protection compared with obtaining consent in advance. Given the huge volume of material streamed on Twitch, the outcome could influence how other platforms handle similar AI training practices involving user-generated content.
Both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Advocates for the streamers argue that livestreaming is creative labour, often a primary income source, and that using it to train commercial AI models without clear upfront consent or payment is a form of exploitation regardless of what buried terms-of-service clauses may technically permit. They contend that opt-out systems introduced only after public backlash, and which fail to protect creators who simply appear on someone else's stream, show that consent was never meaningfully sought, and that people should have control over and be compensated for how their image, voice and performances are commercialised by a powerful corporate parent.
The case against
Those sympathetic to Twitch and Amazon would argue that streamers agreed to broad platform terms of service when they signed up, that hosting hundreds of millions of hours of content inevitably requires flexible rights to operate and improve the service, and that using publicly broadcast material to train AI is a reasonable extension of innovation that platforms have long relied on. They would also note that Twitch did introduce an opt-out mechanism in response to concerns, reflecting a genuine, if imperfect, attempt to balance creator interests with the practical realities of developing new technology at scale, and that legal ambiguity over what data was actually used does not necessarily equate to deliberate wrongdoing.