Twitch’s new AI setting looks like a control. In the lawsuit now facing Twitch and Amazon, it also works as a timestamp. The platform added a switch on August 12 that lets channel owners opt out of having their content used for future generative AI training by Amazon. Eight days later, longtime streamer Warren Pandiscia filed a proposed class action arguing that the harder question is what happened before anyone was given that choice.
Twitch now tells creators what the switch covers. The lawsuit asks whether earlier use of Twitch content was covered by the deal creators thought they had accepted in the first place.
The case was filed on August 20 in the U.S. District Court for the Northern District of California. Its claims are allegations, not findings of fact, and Twitch and Amazon have not been found liable.
The opt-out only controls future Amazon training
Twitch’s current account settings say channel content can be used to train generative AI content models at Amazon. Streams, past broadcasts, clips, stream chat, images and text can all fall within the setting.
The key word is future. Turning the option off stops future Amazon generative AI training tied to that channel. Twitch says the switch does not stop other uses described in its Privacy Notice, including AI-supported Twitch features such as recommendations and moderation.
Amazon’s earlier use of Twitch content does not rest only on the lawsuit. In 2024, Mike Minton told The Information that Amazon was using Twitch content for AI in a “prototyping, not in any kind of production scale, capacity.” Asked on August 12, 2026 what Amazon had already used for model training, he said he did not know.
That confirms some pre-2026 AI use, but not the scope alleged in the complaint. The case still has to establish whether Pandiscia’s own material was used, which Amazon systems received Twitch data, and whether the terms in force at the time authorized that use. The existence of an opt-out today is not proof that Twitch lacked the necessary rights yesterday.
Twitch changed its terms on the same day
August 12 was not only the day the AI setting appeared. Twitch also changed its Terms of Service and Privacy Notice.
The complaint compares an older content-license clause, which allowed Twitch to exercise rights over user content in connection with “monetizing the Twitch Services,” with the August 12 wording, which says those rights may be exercised in connection with “Twitch and/or its affiliates’ business.” The new text expressly refers to affiliate businesses rather than only monetizing the Twitch service, which is why the change matters to the plaintiff’s argument about Amazon.
Pandiscia’s lawyers argue that the amendment shows the previous terms did not authorize Amazon’s separate generative AI business. The timing alone does not settle that. The new wording may have expanded Twitch’s rights or simply made them more explicit; the case would have to resolve what the older license covered. What matters here is that the contract changed on the same day creators were given an explicit Amazon AI training control.
The complaint makes a similar argument about the Privacy Notice. It says the January 2026 version did not list AI training among Twitch’s stated uses of personal information, while the August 12 version added language covering the development or deployment of generative AI models and services.
Your AI choice does not follow you into someone else’s chat
The setting has another wrinkle: the choice belongs to the channel, not to every person whose content appears there.
Twitch says a channel owner can turn off future Amazon generative AI training for that channel. But if a user writes in another streamer’s chat, the other channel’s setting applies. Opting out on your own channel does not carry that preference into every chat you join.
Twitch’s own list shows how broad “channel content” is here: the stream and its chat, VODs, Clips, Highlights, text and images can all be eligible. The opt-out is simple at channel level, but it is not a personal preference that follows a user everywhere on Twitch.
The lawsuit now has to prove what Amazon actually used
Pandiscia is not a celebrity streamer. The complaint says he has streamed on Twitch for close to ten years, has more than 900 followers and has produced over 1,000 hours of content. That makes the dispute easier to see as a creator-platform issue rather than a fight centered on one famous streamer.
The case is not built as a straightforward copyright-infringement claim. Its four causes of action are breach of implied contract, unjust enrichment, breach of express contract and California unfair competition. That puts Twitch’s own contracts, privacy promises and August policy changes at the center of the dispute.
The biggest unanswered question is still basic: what content did Amazon actually use? The complaint alleges that Twitch creator material was used to train Amazon generative AI products, but the filing does not show that Pandiscia’s own 1,000-plus hours were ingested or identify which Amazon models, if any, used his material.
That makes August 12 more useful as a dividing line than as the end of the story. Twitch has now given channel owners a clear choice for future Amazon generative AI training. The lawsuit is about whether creators should have had that choice earlier, what the old terms allowed, and what Twitch and Amazon can show about the content used before the switch appeared.
Sources: Twitch account settings; Pandiscia v. Twitch Interactive, Inc. et al. complaint; The Information’s 2024 reporting on Mike Minton’s comments.