I knew there was a thing called Twitch, but I didn’t know what it did. I learned from the Complaint in Pandiscia v. Twitch Interactive, Inc. that Twitch, which was acquired by Amazon in 2014, is the world’s leading video platform and community for gamers. Because my daughter sometimes relaxes by watching her housemate play video games, I can sort of imagine what that means, but IYKYK, I suppose. The complaint adds details from back in 2014: 55 million unique users, 15 billion minutes of content from 1 million broadcasters. Okay, so huge. In 2026, its even bigger, with 240 million monthly active users and up to 6.9 million creators who go live on the platform each month.
Earlier this month, Twitch confirmed that Amazon uses the platform’s content for AI training. If users don’t want their content used for that purpose, they can opt out (or so Twitch claims — users are skeptical). One might ask, “Why not make it opt in?” Twitch’s Chief Product Officer Mike Say-the-Quiet-Part-Loud Minton explained that nobody would opt in. Plaintiff Warren Plandiscia, on behalf of a purported class of content creators, alleges that Twitch has been mining content for the purposes of AI-training since 2024. Mr. Minton claims not to know how Amazon makes use of its access to user-created content.
In rough outline, the Complaint alleges that Amazon controls a family of generative AI models, including “Nova Reel,” a commercialized text-to-video model. Plaintiffs believe that their content was used to train Nova Reel. Class members have the ability to limit access to their content to paid subscribers. They thus allege that they did not expect their work product would be copied at scale without consent. Amazon was thus enriched by content that it used to train its generative AI model without the consent of the content creators. Those content creators have been harmed in that their intellectual property has been used to enrich Amazon, and they were never compensated for their contributions to this commercial venture.
Plaintiffs allege that Twitch’s Terms of Service (ToS) in effect during the class period, at least October 27, 2023 through August 11, 2026, expressly prohibited the policy announced on August 12th. According to the Complaint, Section 7 of the ToS prohibits “use of any data mining, robots, or similar data gathering or extraction methods” and “downloading (except page caching) of any portion of the Twitch Services, the Materials, or any information contained in them.” Section 10 prohibits accessing any Twitch resource “by means of any robot, spider, scraper, crawler, or other automated means for any purpose.” Twitch’s Privacy Notice, like its ToS, were amended on August 12th to permit notice that Twitch would use content provided by users, among other purposes, to train generative AI.
But it pays to read complaints in full, because only in ¶87 of this Complaint do we learn that if you tried as of August 20, 2026 to access Amazon’s “Generative AI Development Disclosure,” you got this image:
This is all very confusing. I mean, how can you bring a class-action lawsuit against such a pro-doggo business entity? On the other hand, at this point in the complaint, how do we know whether Brandi is a real dog of Amazon? It may be AI generated. It may be an image that Amazon swiped from some poor, unsuspecting user. In any case, that’s a very good dog.
Despite a lot of rhetoric in the complaint about intellectual property, this case is really a straightforward breach of contract claim. Plaintiff alleges breach of an implied contract and breach of the implied covenant of good faith and fair dealing, unjust enrichment, breach of an express contract, and statutory claims under California law. The Complaint seeks injunctive and declaratory relief, as well as money damages.
The theft of intellectual property allegations might be relevant to the damages claim and the unjust enrichment claim. I’m not sure how one would quantify damages, but Paragraph 132 of the Complaint provides a model.
A commercial market exists in which developers of AI systems pay content platforms and rights holders substantial sums to license content for model training: Google, for example, is reported to have negotiated a $60 million per year fee to Reddit for the right to train its AI models on Reddit users’ posts, and Reddit has reported more than $200 million in AI data-licensing revenue, including a comparable arrangement with OpenAI. By taking Plaintiff’s and Class Members’ audiovisual works and communications without any license or negotiation, Defendants obtained for free what participants in that market pay tens of millions of dollars per year or more to acquire, and avoided the acquisition costs that their competitors bear . . . .
What plaintiffs really want, I suspect, is the injunction. They want an order prohibiting Amazon from using their content to train generative AI. But what if it really is true that Amazon has been doing that since 2024? How can a court unbake that cake? A prospective injunction might be possible, but at some point, users have notice of the purposes to which their content will be put, and there’s an opt out, so that will be tricky as well.
Thanks to ContractsProf Matthew Bruckner, who shared the case with me, and to his student Hakeem Osman, who shared the case with him.
Recent Comments