U.S. District Judge André Birotte Jr. in the Central District of California denied Snap’s motion to dismiss the Digital Millennium Copyright Act claim brought by Ted Entertainment, golf-video creator Matt Fisher, Golfholics, and Nicole Chmura. MediaPost first reported the ruling, and Law360 separately reported that the court found the creators had plausibly alleged that users encounter measures intended to protect videos. The litigation is still at the pleading stage: the court has not found that Snap scraped the videos, trained a model on them, or violated the DMCA.
For AI developers and organizations buying training-data services, the immediate significance is narrower and more practical than the broad copyright fight around generative AI. A court has allowed plaintiffs to test whether technical friction on a public video platform—such as anti-bot detection and IP-rate controls—can qualify as an access control under the DMCA’s anti-circumvention provision. Being able to watch a YouTube video in a browser may not, by itself, settle whether automated acquisition of that same video was legally unrestricted.
The case turns on access controls, not ownership alone
The consolidated lawsuit alleges that Snap obtained YouTube content without authorization and used it to train generative AI technology capable of producing video from text or image prompts. The creators’ theory includes ordinary copyright claims, but the dismissal ruling focused on a separate legal route: Section 1201 of the DMCA.
That provision prohibits circumventing a technological measure that effectively controls access to a copyrighted work. It is often associated with password gates, encryption, DRM, and other obvious locks. Snap’s defense, as described by MediaPost, was that the plaintiffs had not alleged such a barrier because YouTube users can generally view the videos without paying, authenticating, or entering a password.
Judge Birotte rejected the idea that free public viewing ends the analysis. The order, as reported by MediaPost, says public accessibility does not establish that the works are “technologically unrestricted.” That sentence is the important part of the ruling, but its meaning should not be overstated. It does not convert every platform term of service, robots file, or rate limit into a DMCA access control. It means the plaintiffs’ allegations about YouTube’s technical measures were sufficiently specific to proceed beyond an early dismissal request.
The complaint identifies YouTube systems that allegedly detect and block unauthorized downloading, including IP-address blocks triggered by excessive download attempts. Those measures are fundamentally different from a creator merely stating “do not copy” in a video description. The plaintiffs are attempting to show that Snap, or contractors acting for it, defeated technical enforcement rather than simply violated a platform rule.
Why public data is no longer a complete answer
AI teams have often treated public availability as a useful dividing line in dataset collection: if content can be reached without a user account, it may be technically easy to acquire at scale. The Snap ruling is a warning that technically reachable and legally unprotected from automated extraction are different propositions.
The distinction matters most where a platform provides content to human viewers but actively detects mass downloading, abnormal request patterns, bot signatures, or automated attempts to retrieve underlying media files. A developer may see a page, player, or public API response; an automated collector may instead interact with delivery infrastructure in ways the platform is designed to identify and restrict. The lawsuit contends that YouTube had exactly this kind of enforcement machinery.
For software organizations, that shifts diligence away from an overly simple “public versus private” assessment. Dataset provenance reviews should identify how material was acquired, whether collection software had to evade throttles or blocks, whether credentials, proxies, rotating IP addresses, or download workarounds were used, and whether the source platform had technical barriers distinct from contractual restrictions.
The court did not establish a general prohibition on web scraping. Nor did it decide whether Snap’s alleged conduct bypassed a qualifying measure. But dismissals are often where novel legal theories die. This one survived, giving plaintiffs the chance to seek evidence about the actual collection workflow, the platform defenses in place, and what Snap knew about both.
A surviving claim can be costly before a verdict
Snap’s loss is procedural, but it carries real operational consequences. Discovery in a case built around AI training data can reach far beyond a single disputed copy of a video. Plaintiffs may seek internal communications about dataset sourcing, vendor contracts, crawler configuration, model-development records, data-retention policies, and technical documentation showing how files were obtained and filtered.
That possibility is why companies building AI features should not dismiss this as a dispute confined to Snapchat. A model developer may acquire training material directly, license a corpus, contract with a data broker, use open-source datasets, or inherit data through an acquisition. Each path produces a different evidentiary trail, but none eliminates the need to know whether the source material was collected through methods that encountered and defeated a platform’s protections.
The case also illustrates a problem for enterprise customers deploying generative AI rather than training foundation models themselves. Contractual assurances that a supplier’s data was “publicly available” or “lawfully sourced” can be too vague to allocate risk. Procurement and legal teams should press for concrete representations about collection methods, third-party data suppliers, known disputes, indemnity limits, and the handling of claims arising from alleged anti-circumvention or copyright violations.
A company does not need to be the entity that wrote the crawler to be drawn into costly questions about what it trained on and how that material entered its development pipeline. Documentation that is boring in a project tracker can become decisive evidence in court: source inventory, acquisition date, collector identity, applicable platform rules, technical controls observed, and approval records.
The ruling leaves the central facts unresolved
The record currently supports several firm conclusions and several that remain allegations. The case was filed in January 2026 and later consolidated with a related action. Court docket information confirms Judge Birotte presides over the matter, while Law360 and MediaPost independently reported the denial of Snap’s motion to dismiss the creators’ DMCA theory.
What is unresolved is equally important. There has been no finding that Snap itself operated a scraper, that any alleged tool circumvented YouTube protections, that the videos entered a Snap training corpus, or that any resulting model output infringed a creator’s rights. Snap may still prevail on the facts, on the legal definition of an effective access control, or on other claims and defenses later in the case.
The supplied reporting also says Ted Entertainment has brought similar litigation against Amazon, Apple, Meta, ByteDance, and OpenAI. Those cases should not be treated as interchangeable. Each turns on its own alleged collection practices, platform protections, technical evidence, and jurisdiction. A ruling about claimed YouTube controls in California does not decide whether another company used the same tools, collected the same data, or faces the same legal exposure.
Still, the fact pattern is likely to recur. Generative-AI litigation has often centered on whether training copies are infringing and whether model outputs are substantially similar to protected works. This case points attention one step earlier: how the training files were obtained.
Dataset governance is becoming a technical control
The practical lesson is not to abandon public-web research, crawling, or machine-learning development. It is to treat dataset acquisition as an engineering, compliance, and security function rather than an informal preliminary task.
Teams that collect public-web material should maintain controls proportionate to the scale and sensitivity of the project:
- They should document the source, collection mechanism, authorization basis, and technical restrictions observed for each dataset.
- They should stop and escalate when collection requires defeating rate limits, bot challenges, IP blocks, authentication requirements, or download restrictions.
- They should vet data vendors for the provenance of their corpora instead of accepting broad claims that material was scraped from public sources.
- They should preserve crawler logs and configuration records long enough to explain how a dataset was built if a rights holder challenges it.
Those steps will not decide the legal merits of the Snap case, and they cannot turn unauthorized material into authorized material. They do reduce the chance that an organization discovers too late that its AI supply chain depends on a collection process no one can explain.
Snap now faces discovery unless the case is resolved or an appeal changes the procedural posture. The concrete consequence for the rest of the AI industry is already visible: public video is no longer a sufficient shorthand for low-risk training data when automated collection must push through a platform’s technical defenses.