TL;DR

A class action filed on behalf of millions of Twitch users accuses Amazon of training AI models on their broadcasts without asking or paying. The claim, brought by Connecticut streamer Warren Pandiscia, argues Twitch broke its user contract by feeding that footage to Amazon’s models. It seeks damages plus an order stopping the practice; neither company has commented.

Twitch does offer an opt-out, and where it sits tells you something. A creator must open the Streamer Dashboard settings, find Security and Privacy, scroll near the bottom of the list and switch off generative AI training. That is four steps deep for a setting that governs whether your work becomes training data.

The more serious flaw is scope. The toggle applies per stream, so a creator who has disabled it can still be captured whenever they appear on someone else’s broadcast with the setting left on. Consent that a third party can override on your behalf is not really consent, and that is the gap the litigation is aimed at.

Twitch’s own explanation of the feature is narrower than the claim: audio may help refine speech-to-text models, improving automatic subtitles across Twitch and Amazon video. When the collection began is unclear. Chief product officer Mike Minton has previously said he did not know whether data had been scraped before opt-outs existed, or what Amazon had used.

Why UK creators should care

The volume explains the stakes — Twitch streamers put out over 215 million hours in the opening months of 2026 alone. Amazon paid close to £740m ($1bn) for the platform in 2014 and has since made AI central to competing with Google and Meta.

Training-data disputes have so far concentrated on text and images. Extending them to live video matters for UK creators because livestreaming carries incidental third parties by default: collaborators, guests, chat participants, people who never agreed to anything. UK copyright and data protection law approaches consent differently from US contract law, and a British claim would likely turn on the UK GDPR question of lawful basis rather than breach of terms.

Looking forward

Watch whether platforms respond by making opt-outs account-level instead of per-stream, which would settle the appearing-on-someone-else’s-stream problem without a court doing it for them. For UK businesses using livestream or video platforms commercially, the practical action is unglamorous: read what the terms say about training rights, and check whether the opt-out you assume is protecting you actually covers the footage you appear in.

Share this article