ByteDance must face YouTubers' DMCA suit over AI training scrape

A federal judge will let the YouTubers' DMCA case against ByteDance proceed, holding that the creators plausibly alleged the TikTok parent bypassed YouTube's technical locks to harvest their videos for AI training. Judge Jacqueline Scott Corley of the U.S. District Court for the Northern District of California denied ByteDance's motion to dismiss the first amended complaint in Ted Entertainment, Inc. et al. v. ByteDance Inc., Case No. 3:25-cv-10933, in an order dated October 2 recorded on the court docket.

The plaintiffs are Ted Entertainment Inc., which runs YouTube channels including h3h3 Productions, and Matt Fisher, who posts golf videos under MrShortGame Golf, per MediaPost. They allege ByteDance circumvented YouTube's technological protection measures, or TPMs, to mass-download videos used to train its generative video systems, including MagicVideo and Seedance. A hearing on the motion at the end of September went badly for ByteDance.

The amended complaint names five measures it says ByteDance defeated: an obfuscated signature system called a rolling cipher; IP-based blocking and rate limiting; short-lived, session-bound streaming URLs; CAPTCHA human-verification challenges; and proof-of-origin tokens that confirm a request comes from an authorized player, as Trademark Lawyer Magazine reported. The filing alleges ByteDance employees disclosed using large YouTube-derived datasets, HD-VILA-100M, Panda-70M and HowTo100M. Those datasets index references to millions of videos rather than holding the files, so the videos still had to be pulled straight from YouTube before training. The order quotes the complaint's claim that employees said they would "download videos with ASR transcripts and audio from YouTube." The filing names the descrambling tool yt-dlp among the methods used.

ByteDance made two arguments. It said the measures govern downloading rather than access to the works, because YouTube videos are free to watch, and it said the creators lacked standing. Bloomberg Law reported that Corley rejected the access argument.

Corley held the plaintiffs plausibly alleged an access control under section 1201(a) of the Digital Millennium Copyright Act. She pointed to the proof-of-origin token, which the complaint says is generated during an active playback session and refused when missing, so even ordinary viewing runs through a process the copyright owner authorized. She relied on the Ninth Circuit's 2017 Disney v. VidAngel decision, which held that one measure can work as both an access control and a copy control, and that a defendant can still circumvent an access control when authorized ways to reach the work exist. That displaced Hattler v. Ashton, a 2017 district ruling ByteDance pressed at the hearing.

On standing, Courthouse News reported that Corley accepted a pocketbook injury. In the order she wrote that the creators plausibly lost per-view advertising revenue and lost YouTube Premium revenue, and lost views and watch hours that feed the site's recommendation and Partner Program systems. She found that injury traceable to the alleged circumvention, and found statutory standing under section 1203, holding that copyright owners whose works were protected by a measure sit inside the class Congress authorized to sue.

Two limits matter. Corley did not decide whether training on the videos infringes copyright, and she did not decide that ByteDance circumvented anything. The fight moves to discovery on whether ByteDance actually bypassed the measures and what it took.

There is precedent. Judge Andre Birotte Jr. of the Central District of California let Ted Entertainment and Fisher bring nearly identical claims against Snap in August, and Corley quoted that ruling's line that YouTube's barriers "may be modest, perhaps by design given YouTube's business model, but they are barriers nonetheless." MediaPost reported that Ted Entertainment has filed similar suits against Amazon, Apple, Meta and OpenAI, all pending in California and Washington state. Our September report on the hearing covered the argument; this order is the ruling. The decision also sits beside the fair-use line, including our item that an appeals court held training an AI on Westlaw notes is not fair use, a separate question this case leaves open.

For anyone who trains, licenses or fine-tunes video models, the order keeps the DMCA in play as a route that tracks how data was taken, not just what a model produced. Publicly viewable is not the same as freely copyable: at least at the pleading stage, a platform's anti-scraping machinery counts as an access control, so a dataset built by defeating it carries legal risk the training labels do not show. Builders who source video through official APIs and licensed corpora stay clear of the theory the court just let advance.

The order resolves ByteDance's motion to dismiss, so the case now moves past the pleading stage.