Copyright Takedown Notices Don’t Require Services to Find and Remove Other Identical Copies–Athos v. YouTube

I previously summarized this case:

Athos owns the copyrights to many classic Mexican films. Users regularly upload its film clips to YouTube. These uploads have irritated Athos since 2014. However, Athos rejected YouTube’s fast-lane options for copyright owners (such as the Copyright Match Tool, Content Verification Program, and Content ID), allegedly because it did not want to agree to release YouTube from liability. Instead, it chose the standard DMCA takedown notification route, which functioned as intended. YouTube expeditiously responded to all DMCA takedown notices, and “Youtube has terminated 1,240 accounts against whose videos Plaintiff submitted takedown requests.” However, after years of tediously submitting standard DMCA takedown notices (allegedly over 3,300 repeated takedown notices in 3 years) and still seeing clips uploaded, Athos had enough and exercised the option it had retained to sue YouTube. Its complaint covers about 7,000 clips from 570 films. YouTube invoked its DMCA safe harbor.

YouTube won at the district court. The 11th Circuit affirms, but the opinion made my head hurt.

First, the opinion spends a lot of time rejecting Athos’ arguments: 7 pages explaining why the plaintiff made “unsustainable” interpretations of the record evidence and another 5 pages on why the Nimmer treatise is wrong about red flags of infringement.

Second, the court was flummoxed by the plaintiff’s wacky advocacy. Here’s an example of the out-there positions Athos took:

According to Athos, once it notifies YouTube of its ownership of a given copyright and identifies a single location (e.g., a URL) where the copyrighted material appears on the YouTube website, YouTube then has a video hash sufficient for it to locate additional infringing material on the site. Athos argues that this confluence of technologies generates actual or red flag knowledge of infringement that should result in YouTube being required to remove any match generated by the hash-matching technologies or otherwise risk losing its safe-harbor protection under § 512(c) of the DMCA. Athos is in effect demanding that YouTube be required to administer the operation of its copyright management tools in a way that it never has before and in a manner that it did not design them for, on behalf of and without any input from a user other than a single DMCA takedown request.

Collectively, the plaintiff’s heterodox arguments led to yet another disjointed, messy, and difficult-to-blog 512(c) safe harbor opinion. This post recaps some of the standout passages, but I couldn’t sanewash the plaintiff’s arguments or the opinion enough to produce a coherent narrative. This may be a collorary to “hard cases make bad law“–wacky advocacy makes wacky opinions.

About the Technology

Athos seemed to envision YouTube’s anti-infringement tools as sentient/near-omniscient autonomous agents that could magically eliminate all instances of infringement once YouTube received a takedown notice. The court essentially tells Athos to stop watching so much 1970s sci-fi:

• “with respect to YouTube’s hash-matching technology, it is possible that a video-hash match identified by the hash-matching process is not actually a violation of the underlying copyright”
• The plaintiff “has offered no evidence that any of YouTube’s video-hash-matching or other tools have any functionality to perform a legal analysis of whether an identified video-hash match constitutes copyright infringement. In fact, YouTube has presented unrebutted evidence that all of its copyright management tools were designed with user input and management in mind because it would be unfeasible for YouTube to operate the tools by itself without such input.”

Scienter

Athos appeared to argue that takedown notices identify not only the specific URL listed but also put the recipient on notice to find other identical copies that the recipient must eradicate. And if a service got too many takedown notices, that indicated they should know their service is a cesspool of piracy and should rearchitect their business. These arguments did not persuade the appeals court:

• “Athos has not offered any evidence that YouTube ever had actual knowledge of infringement that was not being removed from the site because the parties agree that YouTube expeditiously removes specific videos identified by URL in a DMCA takedown request.”
• “the video-hash matches generated by YouTube’s technologies, as we understand them from this record, cannot give YouTube actual knowledge of infringing material because the tools do not perform any analysis to determine whether the hash matches they generate are legally infringing or not.” In other words, it’s not enough that other files exactly match an identified item. That doesn’t answer the legal question of whether the matching file is itself infringing.
• “the tools at most produce lists of possible infringement. To require YouTube to not only run its copyright management tools in the manner Athos suggests, but to then analyze the fruits of those tools to locate other instances of infringement, would amount to requiring YouTube to “affirmatively [seek] facts indicating infringing activity,” in direct contravention of the balance established by Congress in the DMCA. See § 512(m)(1).” In other words, any effort to find identical copies (whether already on the site or uploaded in the future) would impose an affirmative filtering duty that runs afoul of 512(m). The court says YouTube’s tools don’t give rise to “willful blindness” for the same reason. Note the comparison to the Take It Down Act, which will require services to seek out matching items. The Take It Down Act thus imposes affirmative filtering obligations and ensures overremovals.

The Nimmer Disgression

The court spends an unusually long time responding to Nimmer’s treatise. It appears that the treatise remixed some unfortunate dicta from UMG v. Shelter Capital that has been generally disregarded by subsequent courts:

On Nimmer’s reading of § 512(c), a service provider’s CEO indicating awareness of a news article reporting on its website being a general den of infringing activity may be sufficient to establish a triable issue of fact with respect to red flag knowledge.

This invites the court to muse about possible ways that Nimmer might be right, even if it has nothing to do with the case at hand:

Nimmer’s view may be correct in the context of web directories, where a website exclusively dedicated to content piracy might openly advertise itself as such so conspicuously as to necessitate expeditious removal from the directory without requiring knowledge of a specific example of confirmed infringement on the website.

Say what now? This passage is so confused. It’s dicta. It distinguishes between hosting and promotion, but both are publishing functions. 512(c) doesn’t expressly distinguish between hosting and promotion–once a file is stored at the user’s direction, 512(c) applies equally to each. It hypothesizes a “website exclusively dedicated to content piracy” even though we have relatively few examples of this and they don’t last long. The Fung case dealt with a piracy-focused site and disqualified the 512 safe harbor several ways. The court had no reason for musing about unrelated circumstances, and unsurprisingly it found tons of trouble doing so.

After the court’s stroll down hypothetical lane, the court says the Nimmer treatise does not help the plaintiff:

For a video hosting service like YouTube, if general knowledge that the service is being used to occasionally achieve infringement were sufficient to require expeditious removal of all infringing activity, the service provider would then be left to scour its website for all instances of potential infringement in the hopes of retaining its safe-harbor protection. Even after such scouring, the service provider would still be left not knowing whether it had removed enough material (or the right material) to secure safe-harbor protection through subsection (iii). This is the sort of self-monitoring process that the DMCA emphatically does not require of service providers. See § 512(m)(1)….

the simple accumulation of adequately-addressed DMCA takedown requests alone cannot amount to red flag knowledge of other infringing activity which precludes safe-harbor protection.

Note that the inverse proposition would blow up the entire DMCA notice-and-takedown scheme because it would treat notices as identifying more than the specific items they target.

Right and Ability to Control

No one has any idea what the DMCA “right and ability to control” language means…something something about “substantially influencing” users or directing what they do. In any case, YouTube doesn’t have it:

YouTube’s moderation and content management features do not constitute a right and ability to control for purposes of the DMCA….YouTube can remove material and users from its website, can promote and auto-play select videos algorithmically, and can set policies for content moderation on the site. The record contains no evidence that YouTube exercises control over any more substantial a portion of the YouTube website than did Vimeo in Vimeo II.

Implication

Like yesterday’s blog post, this case sought to blow up the DMCA’s notice-and-takedown scheme and impose liability on items that had never been subject to takedown notices. It’s amazing how, a quarter-century-plus later, plaintiffs ares still willing to throw resources at that goal. Fortunately, those legal efforts have generally failed (so far).

Case Citation: Athos Overseas Ltd. v. YouTube, Inc., 2026 WL 44924 (11th Cir. Jan. 7, 2026)