Published
August 27, 2026
A Connecticut streamer has sued Twitch and its parent Amazon, alleging they fed years of streams, VODs, clips and chat logs into Amazon's generative AI without permission or payment. It is a freshly filed complaint, so there is no settlement and nothing to claim — but if you stream on Twitch, the opt-out Twitch switched on by default in August is the part worth checking today.
Allegations Only · No Settlement Yet
This article describes a class action complaint. The statements below are unproven allegations. Twitch Interactive, Inc. and Amazon.com, Inc. have not been found liable, have not yet answered the complaint, there is no certified class, and there is nothing to claim at this time. This page is informational and is not legal advice.
What Is This About?
A proposed class action accuses Twitch Interactive, Inc. and its parent Amazon.com, Inc. of using creators' streams, saved videos, clips, chat logs and channel artwork as training material for Amazon's generative AI products without asking permission and without paying for it. The complaint, Warren Pandiscia v. Twitch Interactive, Inc. and Amazon.com, Inc. (Case No. 3:26-cv-08721), was filed on August 20, 2026 in the U.S. District Court for the Northern District of California, San Francisco Division.The timing is the heart of it. On August 12, 2026, Twitch added an account setting called "Training for Generative AI" that lets a creator block their channel content from being used to train Amazon's models. The setting arrived switched on for everyone, with no email and no pop-up warning first. The complaint alleges that Twitch content had already been going into Amazon AI development since no later than 2024 — years before any creator had a way to say no. Neither company has responded to the allegations in court.
Status
Complaint Filed · August 20, 2026
Pandiscia v. Twitch Interactive, Inc. · U.S. District Court, N.D. California · no answer or responsive motion from either defendant on the docket as of August 27, 2026
Core Allegation
Creator content used to train Amazon AI without consent or payment
Streams, VODs, clips, stream chats and channel imagery, allegedly starting no later than 2024 — before any opt-out existed
Claims Pleaded
Contract and unfair competition — not copyright
Breach of implied contract and the implied covenant · unjust enrichment · breach of express contract · California UCL, Cal. Bus. & Prof. Code § 17200
Can I Claim?
No — nothing to claim yet
No settlement, no fund, no claim form; no class has been certified. What you can do today is check your own opt-out setting
What Twitch Turned On, and When
On August 12, 2026, Twitch's support account announced that it had added a setting letting creators opt out of having their channel content used to train generative AI models across Amazon. According to Twitch's own help documentation, the content covered by the program includes a channel's streams, VODs, clips, stream chats, and the pictures and text on the channel page — and the resulting models are not limited to Twitch. Twitch's FAQ gives the example of a creator's audio helping refine speech-to-text models that would improve captions at Twitch and across Amazon generally.Two design decisions turned an ordinary product announcement into a complaint. The first is that the setting shipped enabled for every account, so every creator was enrolled before they knew the program existed. Twitch sent no email and displayed no notification; the setting was noticed by a reporter and spread from there. The second is what Twitch's chief product officer, Mike Minton, said when asked about it during a live Q&A: "If it was opt-in, nobody would opt in. That's honestly the answer."
That candor is now quoted back at the company in the complaint, which uses it to argue that Twitch knew its creators would refuse and structured the program to avoid ever having to ask. Asked in the same session whether creator content had already been used, Minton answered: "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 Two-Year Gap the Case Is Really About
If the program had started on August 12, this would be a disagreement about defaults. The complaint's more serious allegation is that it did not.It points to remarks Minton made in 2024, when he was Twitch's chief monetization officer, at an event hosted by The Information. Asked whether Amazon uses Twitch to train AI models, he is reported to have answered, "Yeah, for sure," adding that the work was "in a prototyping, not in any kind of production scale, capacity" and that it was happening "within the bounds of user trust" and "within the bounds of privacy regulations." The complaint alleges that creators were never told, that the Terms of Service and Privacy Notice were not amended to reflect it, and that no opt-out of any kind existed at the time.
The complaint draws its sharpest inference from that pairing: a company that in 2024 said it was operating "within the bounds of user trust" is, on the plaintiff's account, the same company that in 2026 explained the default-on design by saying nobody would have agreed. Whether those two statements can be reconciled is the kind of question that gets litigated rather than assumed, and Minton's 2024 characterization of the work as prototyping rather than production is a fact the defendants may well press.
The Contract Rewrite the Complaint Zeroes In On
The most technical part of the filing is also the part most likely to decide it. Both defendants will argue the Terms of Service already gave them what they took. The complaint's answer is that Twitch quietly rewrote those terms on the same day it announced the program — which, it argues, is evidence the old terms did not cover it.Three changes are singled out:
• The purpose of the content license. Section 8(a)(i) of the operative Terms of Service granted Twitch and its sub-licensees the right to use, reproduce, modify and adapt user content, and stated that "Twitch can exercise these rights in connection with monetizing the Twitch Services." On August 12, 2026, the complaint alleges, that sentence became "in connection with Twitch and/or its affiliates' business" — broad enough to reach Amazon's separate AI business.
• A new recital of user "direction." The same amendment allegedly added language providing that by streaming, a creator is "directing Twitch to also cache and/or store your User Content and every individual piece of it." The complaint says no comparable language appeared in any prior version.
• The Privacy Notice. The version in effect through August 11, 2026 enumerated how Twitch uses user information — operating the service, security, recommendations, advertising, fraud prevention — and, per the complaint, said nothing about training artificial intelligence. The August 12 revision allegedly added language about using data "to develop or deploy our generative AI models and services," linking to an Amazon disclosure page that the plaintiff's counsel says returned an error when checked on August 20.
From there the complaint makes its cleanest argument: if the old license already permitted AI training, none of these edits would have been necessary — and neither would an opt-out toggle, since an irrevocable license that already conveyed the right would make opting out meaningless.
It also turns Twitch's own rules around. Sections 7 and 10 of the Terms of Service prohibit users from data mining, bulk downloading, or using any robot, spider, scraper or crawler against the Twitch Services. Creators, the complaint argues, reasonably read those prohibitions as describing what nobody was allowed to do to platform content. Separately, it points to the Privacy Notice's promise that Twitch "will not materially change our policies and practices to make them less protective of personal information collected in the past under a prior version of the Privacy Notice without the consent of affected individuals," and its promise of notice and a chance to decline before information is shared. The complaint alleges both promises were broken.
Why This Is a Contract Case and Not a Copyright Case
Most of the AI training litigation of the past two years has run on copyright — the $1.5 billion Anthropic authors settlement, the consolidated OpenAI copyright cases, and the publishers' suit over Gemini all sit there. This one deliberately does not, and the complaint says why: Twitch streams and videos are not registered with the U.S. Copyright Office.That matters procedurally. Registration is generally required before a copyright owner can sue for infringement of a U.S. work, and a live stream produced on a Tuesday is not something most creators register. So the case is built on the agreement instead — what Twitch promised, what the license actually covered, and whether taking the content outside that scope enriched the companies at creators' expense. The complaint is careful to add that a lack of registration does not make the work valueless or unprotected.
The practical consequence is that this case will likely be decided on contract interpretation rather than on fair use. That is a different battlefield, with different odds, and it is why the August 12 amendments carry so much weight in the filing.
The Chat Problem: Your Opt-Out Doesn't Follow You
One detail deserves attention from anyone who types in Twitch chat, not just people who broadcast. The setting is scoped to a channel, not to a person.Twitch's help documentation states that if you chat on someone else's stream, that channel owner's opt-out preference governs whether the chat can be used for training. So a viewer who turns the setting off on their own account has not stopped their messages in other channels from being collected, and a channel owner who leaves the default in place is, in the complaint's framing, purporting to authorize the capture of messages written by people who were never asked.
The complaint builds a wiretapping-style factual record around this — describing how streams and chat travel across Twitch's network in real time and arguing that Amazon is a separate entity from Twitch and not a party to those communications. But it is worth being precise about what it does with that record: the California Invasion of Privacy Act appears as a predicate for the Unfair Competition Law's "unlawful" prong, not as a standalone cause of action. There is no separate CIPA count in the complaint. Whether one gets added in an amended filing is something to watch.
Users have also reported that the opt-out toggle reverted to the enabled position after being switched off. That is a user-reported issue repeated in the complaint, not something Twitch has confirmed — but it is a reason to re-check the setting rather than assume it stuck.
What the Complaint Says the Content Is Worth
To show a measurable benefit, the complaint reaches for the licensing market that already exists. It cites reporting that Google negotiated a fee in the range of $60 million a year for the right to train on Reddit users' posts, and that Reddit has reported more than $200 million in AI data-licensing revenue including a comparable arrangement with OpenAI. The argument is that Twitch and Amazon obtained for free the same category of asset that other AI developers pay tens of millions a year to license.On the demand side, the complaint points at Amazon Nova Reel, the text-to-video and image-to-video model launched in December 2024 and sold through Amazon Bedrock, where enterprise customers are charged per second of video generated. It quotes Amazon's own service-card language describing pre-training on "licensed and proprietary data, open-source datasets, and publicly available data where appropriate," and alleges on information and belief that Twitch content sits inside that "proprietary data."
It also cites a March 2024 Amazon research paper, VidLA: Video-Language Alignment at Scale, in which Amazon researchers described using 20 million videos from a public YouTube-derived dataset to build a training corpus of roughly 800 million video clips — offered as evidence of Amazon's appetite for video training data at scale. And it references a separate pending case, Ted Entertainment, Inc. v. Amazon.com, Inc. (W.D. Wash.), whose amended complaint alleges Amazon obtained YouTube videos for Nova Reel using tools designed to defeat YouTube's access controls. Those are allegations in another lawsuit, not findings, and Amazon has not been found liable there either.
The contrast the complaint draws from all of this is simple: at YouTube, Amazon allegedly had to get around someone else's platform; at Twitch, it owned the platform and held the content already.
Who the Proposed Class Covers
The complaint proposes a single nationwide class: all content creators who used the Twitch platform at twitch.tv and whose content was used to train Amazon's AI products without consent. Judges assigned to the case and their immediate families, the defendants and their affiliates, officers, directors and employees, and counsel for both sides are excluded.The filing estimates the class likely runs to millions of people and says members can be identified from the defendants' own records. It invokes federal jurisdiction under the Class Action Fairness Act, alleging more than $5 million in controversy and more than 100 class members.
No class has been certified. Certification is a contested motion that comes much later, and this definition has an obvious pressure point built into it: membership turns on whether a given creator's content actually went into Amazon's training data, which is a fact currently known only to the defendants. Expect that to be argued about.
The named plaintiff is a Connecticut resident in Fairfield County who has streamed on Twitch for close to ten years, has more than 900 followers, and says he has streamed over 1,000 hours — largely LEGO builds and video games — while spending tens of thousands of dollars on equipment, props and subscriptions. He says he turned the AI training setting off as soon as he learned it existed.
What the Lawsuit Asks For
The complaint pleads four causes of action. Breach of implied contract and of the implied covenant of good faith and fair dealing runs against Twitch alone; unjust enrichment (pleaded in the alternative), breach of express contract, and violation of California's Unfair Competition Law run against both defendants.On relief, it seeks damages, restitution and disgorgement of the profits and avoided costs attributable to the alleged use. The injunctive requests are the more striking half: an order requiring express, informed, person-scoped consent before creator content and communications are used for AI training — including consent from chat participants — a durable and effective opt-out, corrective disclosure, and an order to identify, segregate and delete unlawfully obtained content and to stop using training corpora and models to the extent they are derived from that material.
That last request is the one to watch, because the complaint itself explains why it is hard: once content is ingested, it argues, the material is absorbed into the model's weights and is not capable of clean deletion or retraction. Courts have not settled what a remedy looks like when the allegedly improper input cannot simply be handed back.
Where the Case Stands Now
The case is at its earliest stage. The complaint was filed August 20, 2026 with a jury demand, and the docket shows summonses issued and the case assigned on August 21, 2026, along with the district's standard initial case management scheduling order setting alternative dispute resolution deadlines. It is currently before Magistrate Judge Lisa J. Cisneros; in the Northern District of California a case assigned to a magistrate judge proceeds that way only if all parties consent, so reassignment to a district judge is a routine possibility. An administrative motion under Civil Local Rule 7-11 was filed on August 26, 2026.Neither Twitch nor Amazon has filed an answer or a responsive motion as of publication, and neither company has publicly commented on the lawsuit. Their first substantive filing — most often a motion to dismiss in a case like this one — will be the first real read on how the terms-of-service argument lands. Because the complaint leans so heavily on the scope of the Section 8 license, an early fight over what those terms permitted is close to inevitable.
Is There a Settlement or Claim Form?
No. This is a newly filed lawsuit, not a settlement.There is no settlement fund, no claim form, no payout and no deadline to act. For money to reach any creator, the case would have to survive the defendants' early motions, win class certification, and then either settle or prevail — a path measured in years that frequently ends before it gets there. Treat any site offering to let you "file a claim" against Twitch or Amazon over AI training today as a warning sign.
What Twitch Creators Can Do Today
Whatever happens to the lawsuit, the setting it describes is live right now, and it defaults to on.• Check the toggle. In Twitch account settings, open the Security and Privacy section and find the option labeled "Training for Generative AI." Turning it off applies to your channel's content going forward.
• Check it again later. Some users have reported the setting reverting to enabled after being switched off. Confirming it a second time costs nothing.
• Understand what it does not cover. Per Twitch's documentation, your messages in other people's chats are governed by those channel owners' settings, not yours.
• Keep your own records. If you would want to show later what you streamed and when, your VOD and clip history, channel analytics and export data are the evidence that would matter. Screenshots of the setting's state when you found it are worth keeping too.
None of that is a claim, and none of it requires a lawyer. It is the ordinary housekeeping that makes a difference if the case ever reaches a stage where class membership has to be shown.
Frequently Asked Questions
Is there a Twitch AI settlement or claim form?
No. This is a class action complaint filed on August 20, 2026, not a settlement. There is no settlement fund, no claim form, no payout and no deadline. Neither Twitch nor Amazon has been found liable, no class has been certified, and there is nothing to claim at this time.
What does the Twitch AI training lawsuit allege?
That Twitch and Amazon used creators' streams, videos on demand, clips, stream chats and channel imagery to train Amazon's generative AI models without consent or compensation, that the practice began no later than 2024, and that the opt-out setting Twitch added on August 12, 2026 arrived after the fact and was switched on by default. The complaint pleads breach of implied contract and the implied covenant of good faith and fair dealing, unjust enrichment, breach of express contract, and violation of California's Unfair Competition Law. These are unproven allegations, and neither company has responded in court.
How do I stop Twitch from using my channel to train AI?
Twitch's setting is called "Training for Generative AI" and lives in account settings under Security and Privacy. Turning it off applies to your own channel. Twitch's help documentation states that if you chat on someone else's stream, that channel owner's preference governs whether the chat can be used for training, so your own setting does not follow you into other channels. Some users have reported the toggle reverting to the enabled position, so it is worth checking it again after you change it.
Does opting out remove content Twitch already used?
Twitch has described the setting as governing future training. The complaint alleges that content already ingested cannot practically be pulled back out of a trained model, which is why it asks the court to order the companies to identify, segregate and delete unlawfully obtained material and to stop using models derived from it. No court has ruled on that request.
Why is this a contract lawsuit instead of a copyright lawsuit?
The complaint states that Twitch streams and videos are not registered with the U.S. Copyright Office. Registration is generally a prerequisite to filing a federal copyright infringement suit over a U.S. work, so the case is built on contract and California's Unfair Competition Law instead, arguing that Twitch's own Terms of Service and Privacy Notice did not permit the use.
Who would the proposed class cover?
The complaint proposes a nationwide class of all content creators who used twitch.tv and whose content was used to train Amazon's AI products without consent. Judges, the defendants and their affiliates, and counsel for both sides are excluded. No class has been certified, and the definition could be narrowed, expanded or rejected as the case proceeds.
Is this related to other Amazon or Twitch lawsuits?
It is separate from Twitch's earlier disputes with its own users, from the Amazon Fire TV viewing-data case, and from the copyright-based AI training cases against other developers. The complaint does cite a pending suit in Washington, Ted Entertainment, Inc. v. Amazon.com, Inc., whose allegations about how Amazon obtained YouTube videos for Nova Reel are unproven in that case as well.
Sources
• Court records — Warren Pandiscia v. Twitch Interactive, Inc. and Amazon.com, Inc., No. 3:26-cv-08721, Class Action Complaint and docket (N.D. Cal., filed August 20, 2026).• Twitch — Terms of Service (Sections 7, 8, 10 and 19)
• Twitch — Privacy Notice
• Twitch Help — Account Settings: Training for Generative AI
• Amazon Web Services — Amazon Nova Reel AI Service Card
• VidLA: Video-Language Alignment at Scale — Amazon research paper (March 2024)
• Ars Technica — Twitch content has trained Amazon AI for years, but users can opt out now
• The Register — Twitch feeds your streams to Amazon's AI unless you tell it to stop
• Engadget — Twitch and Amazon hit with lawsuit over training AI with streamers' content
• GeekWire — Amazon's Twitch is using livestreams to train its generative AI
Free settlement alerts
Get notified when new class actions open to claims
Join thousands of readers who get the latest class action settlements you may qualify for — delivered straight to your inbox.
For more class actions keep scrolling below.
Status
Complaint Filed — Allegations Only
Case Title
Warren Pandiscia v. Twitch Interactive, Inc. and Amazon.com, Inc.
Case Number
3:26-cv-08721
Court
U.S. District Court, N.D. California (San Francisco Division)
Judge
Hon. Lisa J. Cisneros (Magistrate Judge)
Date Filed
August 20, 2026
Defendants
Twitch Interactive, Inc. · Amazon.com, Inc.
Claims
Breach of implied contract & implied covenant · Unjust enrichment · Breach of express contract · Cal. Bus. & Prof. Code § 17200
Jurisdiction
Class Action Fairness Act, 28 U.S.C. § 1332(d)
Jury Demand
Yes — demanded by plaintiff
Official Website
Twitch Terms of Service
