Amazon Is Being Sued by YouTubers Who Say It Scraped Their Videos to Train an AI Tool Without Permission

What Amazon is accused of is a certain kind of audacity; it is not the crime of desperation or the corner-cutting of a scrappy startup struggling to survive, but rather the methodical, infrastructure-scale operation of a company that just decided it wanted something and went ahead and took it.

In a lawsuit filed in a federal court in Seattle, a group of YouTube creators alleges that Amazon stole millions of their videos without their consent in order to create a product that it currently sells to business clients.

Company Sued	Amazon.com, Inc.
AI Product at Issue	Nova Reel — a text-to-video generative AI model launched December 2024, available via AWS Bedrock
Lead Plaintiff	Ted Entertainment, Inc. (TEI) — the California-based media company owned by Ethan and Hila Klein, operators of h3h3 Productions and H3 Podcast Highlights
Other Plaintiffs	Matt Fisher (MrShortGame Golf, 500,000+ subscribers) and Golfholics (130,000+ subscribers)
Combined Reach	2.6 million+ subscribers, ~4 billion combined views, 5,800+ original videos across all plaintiffs
Court Filed	U.S. District Court for the Western District of Washington, Seattle
Legal Basis	Digital Millennium Copyright Act (DMCA), Section 1201 — anti-circumvention provisions; copyright infringement
Alleged Method	Automated download tools using virtual machines with rotating IP addresses to bypass YouTube’s access protections
Datasets Cited	HD-VILA-100M (Microsoft Research Asia, 2021) and HD-VG-130M (Peking University/Microsoft) — both published for academic use only
Relief Sought	Monetary damages, restitution, and injunctive relief to stop Amazon from distributing a model trained on their content
Amazon’s Response	Declined to comment, citing ongoing litigation
Similar Cases	Same plaintiff group has also sued Nvidia, Meta, ByteDance, Snap, OpenAI, and Apple

The names of the plaintiffs are not obscure. The primary plaintiff is h3h3 Productions, which is run by Ethan and Hila Klein, and Ted Entertainment Inc., the company behind the hugely successful H3 Podcast. Beside them are Golfholics, a channel with over 130,000 subscribers, and Matt Fisher, the creator of MrShortGame Golf.

These three accounts together account for about four billion views and 2.6 million subscribers. They claim that Amazon fed everything into a machine without a single phone call, email, or dollar. These are real creators with real audiences who spent real years creating something.

The lawsuit targets Amazon’s text-to-video generative AI system, Nova Reel, which was introduced via AWS Bedrock in December 2024. In order to compete with OpenAI’s Sora and Google’s Veo for business clients, the model takes text prompts and images and outputs brief video clips.

It’s possible that the majority of Nova Reel users are unaware of the source of the training materials. It’s also possible that Amazon’s legal team made a calculated wager that fair use would cover it because they knew exactly what they were doing.

The description of how the alleged scraping was carried out, rather than just the accusation itself, is what makes the complaint noteworthy. The filing claims that Amazon did not merely license any content or use YouTube’s API to request videos. In order to avoid being detected and blocked by YouTube’s detection systems, it instead used automated downloading tools in conjunction with virtual machines that rotated through IP addresses.

That’s not a coincidence. Engineering is that. That person wrote code specifically designed to make the extraction undetectable. It’s difficult to ignore the discrepancy between what this lawsuit claims actually occurred and what Amazon’s marketing copy says about honoring creators.

The complaint focuses on two academic datasets that were published solely for academic purposes: HD-VILA-100M, which was created by Microsoft Research Asia in 2021, and HD-VG-130M, which was created by researchers at Peking University and Microsoft. Importantly, only URLs pointing to YouTube content were present in both datasets rather than the actual video files.

A business would need to download the videos in order to train an AI video model on that content. The plaintiffs claim that Amazon used the previously mentioned circumvention infrastructure to do precisely that, on a large scale. They point out that a different lawsuit against Nvidia alleged the same pattern.

Section 1201 of the Digital Millennium Copyright Act, which forbids evading technological protection measures, serves as the foundation for the legal theory. The claim is that Amazon purposefully and repeatedly circumvented YouTube’s systems for limiting bulk access to its video library, which are considered to be such protections.

The ramifications extend far beyond Amazon if courts adopt this framing. It would imply that downloading YouTube videos in large quantities for AI training violates the DMCA regardless of whether the content is accessible to the general public due to the method of acquisition.

One particularly noteworthy passage in the complaint is “Once AI ingests content, that content is stored in its neural network and not capable of deletion or retraction.” The plaintiffs are seeking more than just damages for past injuries. They are dealing with an issue for which there is currently no legal remedy.

There might be no practical way to reverse what has already been done if a business trains a model on your work without your consent and ships that model to thousands of business clients. In contrast to a typical copyright claim regarding a book or article, that is an existential argument rather than a legal one.

The case comes at a time when courts are genuinely unsure about whether AI training on copyrighted content counts as fair use. The system is currently processing dozens of similar cases, including those involving authors against Meta, the New York Times against OpenAI and Microsoft, and musicians against Google. A few have made up their minds. A few have been fired. The main issue that everyone is debating—where fair use ends and infringement begins—has not been fully addressed by any of them.

Regarding the lawsuit, Amazon declined to comment. It doesn’t reveal anything, and that’s standard procedure. However, as this develops, there’s a sense that the generative AI sector built a large portion of its infrastructure on the premise that training data was essentially free, or at least publicly accessible, and that courts might not share. The Seattle YouTubers may not prevail. However, their question is precisely the right one.