Twitch and Amazon are being subjected to a class action lawsuit regarding the training of AI using livestream content.

      On August 20, a Twitch streamer filed a lawsuit against Twitch and Amazon regarding the use of livestreams for training Amazon's generative AI. The complaint is 37 pages long and was submitted to the Northern District of California, San Francisco Division, with the case number 3:26-cv-08721. The primary claim in the document revolves around a contract dispute. On August 12, Twitch introduced an opt-out option and revised the section of its Terms of Service that dictates how it can use a creator’s content.

      The section that was modified on August 12 is 8(a)(i), which previously granted Twitch a license for a specific purpose. The complaint cites the original wording, stating: “Twitch can exercise these rights in connection with monetizing the Twitch Services.” The updated phrasing now reads “in connection with Twitch and/or its affiliates’ business,” with Amazon identified as the affiliate. Additionally, the amendment included a new statement indicating that by streaming, users are “directing Twitch to also cache and/or store your User Content and every individual piece of it.” The complaint asserts that such language did not exist in earlier versions, claiming that these changes demonstrate the prior terms “did not convey the rights Defendants exercised.”

      Concerning content deletion, the complaint references section 8(a), which specifies that the license “terminate[s] once you delete such User Content from the Twitch Services, or generally by closing your account.” AI training is not listed among the uses that persist after deletion, according to the complaint. It also highlights two other Twitch rules: section 7 prohibits “use of any data mining, robots, or similar data gathering or extraction methods,” and section 10 forbids access “by means of any robot, spider, scraper, crawler, or other automated means.”

      Regarding the prior privacy notice, last updated on January 15, 2026, it outlined the ways Twitch utilized user information, with no mention of AI training. On August 12, Twitch added new text that states it uses information “to test and improve technology, including machine learning and artificial intelligence,” and another clause refers to “using data to develop or deploy our generative AI models and services.” Two commitments from both versions remain: Twitch pledged not to make its policies “less protective of personal information collected in the past” without consent and promised users prior notice before sharing their data with third parties.

      The plaintiff, Warren Pandiscia, resides in Fairfield County, Connecticut, and has been streaming on Twitch for nearly a decade, with his channel having over 900 followers and more than 1,000 hours of content. He has invested tens of thousands of dollars in equipment, props, and subscriptions, focusing on building LEGO sets and playing video games. After discovering the existence of the opt-out setting, he chose to opt out. He is represented by Almeida Law Group and Sterlington.

      The complaint quotes Mike Minton, Twitch’s chief product officer, who, during a live session on August 12, defended the default inclusion of users in the setting by stating, “If it was opt-in, nobody would opt in. That’s honestly the answer.” In response to users’ inquiries about whether their content had already been used for training, he expressed uncertainty, stating: “I don’t actually know the answer to that question because I don’t know what Amazon has done in terms of model training and what they’ve used and not used.”

      The complaint also references a previous statement made by Minton in 2024, where, as chief monetization officer, he confirmed at an event that Amazon uses Twitch to train AI models but mentioned it was primarily for prototyping rather than production scale.

      The settings discussed pertain to channels rather than individual users. According to Twitch’s help pages, if someone chats on another person's stream, the opt-out preferences of that channel dictate whether that chat can be utilized for training. The complaint states that Twitch provided no advance notice about the changes, lacking any email alerts or pop-ups. A reporter, Zach Bussey, discovered the setting, and users have reported that their opt-out status sometimes reverted to enabled after being disabled. The opt-out option was covered in a report on August 13.

      Pandiscia makes four claims: breach of implied contract, unjust enrichment, breach of express contract, and violation of California's Unfair Competition Law at Business and Professions Code section 17200. The implied contract claim names only Twitch, while the other three include both companies. The filing does not include a copyright claim, stating that Twitch streams and videos “are not registered with the US Copyright Office.” It clarifies that lack of registration does not imply they are unprotected or valueless. The California Invasion of Privacy Act is mentioned only as a basis for the unfair competition claim and is not a separate count.

      The class is defined as “All content creators who used the Twitch platform at twitch.tv and whose content was used to train Amazon’s AI