ByteDance must face YouTubers’ AI training suit over alleged YouTube scraping

A California federal judge has refused to dismiss a lawsuit accusing ByteDance Inc. of circumventing YouTube’s technological protections to obtain videos allegedly used to train its generative AI systems.

Judge Jacqueline Scott Corley of the US District Court for the Northern District of California ruled on October 2 that Ted Entertainment Inc. and video creator Matt Fisher had plausibly alleged that ByteDance circumvented technological measures controlling access to their YouTube videos, allowing their claim under the Digital Millennium Copyright Act (DMCA) to proceed.

The ruling does not determine that ByteDance actually circumvented YouTube’s protections or that its use of YouTube content to train AI systems infringed copyright. Rather, the court held that the plaintiffs had pleaded sufficient facts to survive ByteDance’s motion to dismiss.

Alleged YouTube scraping for AI training

The case, Ted Entertainment, Inc. et al. v. ByteDance Inc., No. 3:25-cv-10933, centers on allegations that ByteDance obtained YouTube videos for use in developing AI products, including MagicVideo and Seedance.

The plaintiffs allege that ByteDance and its agents, contractors, and affiliates scraped large volumes of YouTube videos using tools and workflows that bypassed or evaded YouTube’s technological protections.

The amended complaint identifies five alleged technological protection measures used by YouTube: an obfuscated signature system known as a “rolling cipher”; IP-based blocking and rate limiting; short-lived, session-bound streaming URLs; CAPTCHA challenges triggered by automated activity; and proof-of-origin tokens verifying requests from authorized client environments.

The plaintiffs further allege that ByteDance employees disclosed the use of large-scale YouTube-derived datasets, including HD-VILA-100M, Panda-70M, and HowTo100M, to train the company’s AI models.

Those datasets contain references to millions of YouTube videos or clips, rather than the underlying audiovisual files. According to the amended complaint, the videos must be accessed and downloaded directly from YouTube before the datasets can be used for AI training.

The plaintiffs allege that ByteDance employees acknowledged downloading videos from YouTube and that ByteDance used additional YouTube-sourced content obtained through similar means.

They also allege that ByteDance used tools including yt-dlp to circumvent YouTube’s normal access controls and that the company’s acquisition of the videos violated YouTube’s terms of service. These remain allegations in the case.

Court rejects ByteDance’s access-control argument

ByteDance sought dismissal of the case on several grounds, including that the plaintiffs lacked standing and had failed to state a claim under Section 1201(a) of the DMCA.

The company argued that YouTube videos are publicly available for viewing and that the technological measures identified by the plaintiffs regulate downloading or copying rather than access to the copyrighted works themselves. Judge Corley rejected that argument at the pleading stage.

Section 1201(a)(1)(A) prohibits circumventing a technological measure that effectively controls access to a copyrighted work. The court explained that such a measure is one that, in the ordinary course of its operation, requires the application of information, a process or treatment, with the authority of the copyright owner to gain access to the work.

The plaintiffs alleged that even viewing a YouTube video requires users to satisfy technological conditions imposed by YouTube’s delivery systems.

In particular, they alleged that proof-of-origin tokens are dynamically generated by YouTube’s official player software during an active playback session and that requests without the required token are refused.

Judge Corley found that, drawing reasonable inferences in the plaintiffs’ favor, those allegations were sufficient to plausibly establish that the measures effectively controlled access to the works. The court therefore rejected ByteDance’s argument that the safeguards merely controlled copying or downloading.

Alleged circumvention was sufficiently pleaded

The court also found that the plaintiffs had plausibly alleged that ByteDance actually circumvented at least some of the technological measures.

The plaintiffs alleged that the relevant datasets contained references to their YouTube videos and that using those datasets required the videos to be downloaded directly from YouTube. They further alleged that ByteDance had extracted, replicated, or reused proof-of-origin token parameters outside the authorized playback environment in order to retrieve video data directly.

Judge Corley concluded that, given the allegations that ByteDance employees had stated they downloaded videos from YouTube and the absence at this stage of a plausible alternative explanation for how the downloads could have been obtained without circumventing the technological measures, the plaintiffs had adequately pleaded circumvention.

The court also relied on the Ninth Circuit’s 2017 decision in Disney Entertainment, Inc. v. VidAngel, Inc., which held that a technological protection measure can operate as both an access control and a copy control, and that authorized methods of accessing a work do not necessarily prevent a technological measure from qualifying as an access control under Section 1201(a).

Creators can pursue the DMCA claim

ByteDance also challenged the plaintiffs’ standing to bring a Section 1201 claim, arguing that the private right of action should extend to the owner of the technological measure rather than to individual copyright owners whose works were hosted on a third-party platform. The court rejected that argument.

Judge Corley held that copyright owners whose works were allegedly protected by the technological measures fall within the class of persons Congress authorized to sue under Section 1203 of the DMCA.

The court also found that the plaintiffs had plausibly alleged an injury traceable to the alleged circumvention, including lost advertising revenue resulting from video retrievals that generated no advertising impressions. The plaintiffs also alleged lost YouTube Premium revenue, lost views and watch hours, reduced algorithmic amplification, harm to the market for their content, and loss of control over their works. Accordingly, ByteDance’s motion to dismiss was denied.

No ruling yet on whether AI training infringed copyright

The ruling is significant, but it is important to distinguish the issue decided by the court from the broader question of whether AI companies can lawfully train models using copyrighted material obtained from online platforms.

The plaintiffs’ operative complaint asserts a DMCA Section 1201(a) anti-circumvention claim. Judge Corley was not asked at this stage to decide whether training an AI system on the plaintiffs’ videos constitutes copyright infringement.

Instead, the key issue was whether the plaintiffs had plausibly alleged that ByteDance circumvented technological measures controlling access to their copyrighted works. That distinction could prove important as copyright owners and AI developers continue to litigate disputes over the acquisition and use of training data.

The decision nevertheless provides a potentially significant avenue for copyright owners where AI companies are accused not simply of using copyrighted content, but of obtaining that content by bypassing technological restrictions.

A potentially important development for AI scraping disputes

The ruling adds to a growing body of US litigation examining the relationship between web scraping, technological protection measures, and AI training. Its particular significance is Judge Corley’s conclusion that, at the pleading stage, technological measures governing access to publicly viewable YouTube content can potentially qualify as access controls under Section 1201(a).

That could make the DMCA an important part of future disputes involving the large-scale automated collection of copyrighted material for AI training, particularly where content platforms deploy technical measures intended to restrict automated extraction. The case will now proceed beyond the motion-to-dismiss stage, where the parties can litigate the factual question of whether ByteDance actually circumvented YouTube’s technological measures and what content was obtained and used.

For now, however, the court has allowed the creators’ DMCA claim to proceed based on allegations that ByteDance bypassed technological protections while obtaining YouTube content for AI-related purposes.

← Back to News