Photo of John Gray

The court in Planner 5D v. Facebook, one of the first cases about trade secrets in artificial intelligence datasets, has granted the motion to dismiss on the copyright claims and denied the motion on the trade secret claims.

Trade secrets. The court held that the amended complaint sufficiently pleaded that the Planner 5D data files were trade secrets and that the defendants used improper means to acquire those trade secrets.

Defendants first argued that the Planner 5D files were not trade secrets. Princeton contended that Planner 5D designed its website to transmit object or scene data files to users’ browsers, where they were automatically stored in the browser’s cache, which enabled any user to access and reproduce those files. While the terms of use prohibited downloading material generally and prohibited using scraping tools, the terms of use expressly excluded “page caching” from the list of prohibited activities. Princeton relied on multiple exhibits and a declaration purportedly showing these technical details, but the court rejected much of that evidence because courts typically do not consider extrinsic evidence (e.g., exhibits not attached to the complaint) to resolve a motion to dismiss.

Taking the complaint’s factual allegations as true, the court held that Planner 5D sufficiently elaborated on the structural and legal measures used to protect its data files so that they could be classified as trade secrets. In other words, Planner 5D used a combination of legal and technical measures as reasonable efforts to protect its information, even though the information was technically accessible online. Specifically, Planner 5D alleged that each object and scene file is located at a “unique” and “secret” internet address that is “never shown” to Planner 5D’s users. Identification of the internet address, while technically possible, is impossible without using software developer tools that violate the terms of use. Both defendants have cited to Arkeyo, LLC v. Cummins Allison Corp., 342 F. Supp. 3d 622 (E.D. Pa. 2017), in which the court denied a preliminary injunction hearing where the plaintiff published a full set of its software on the internet without employing industry standard protections for source code, such as encryption or other technological measures. Arkeyo noted that a trade secret plaintiff cannot commit the “cyber equivalent of leaving its software on a park bench.” But the court distinguished Arkeyo because Planner 5D made its URLs inaccessible without violating terms of use, whereas in Arkeyo the URLs did not have random names.

The court also held that the complaint alleged that Princeton used improper means to acquire the trade secrets. The court held that Princeton knew or should have known that Planner 5D intended the files on secret websites, accessible only through data scraping techniques prohibited by terms of use, to be confidential. With respect to Facebook, Planner 5D acknowledged that Facebook did not directly scrape data from the Planner 5D website and instead acquired the data from Princeton. But the court found that the complaint sufficiently alleged “many close connections” between Facebook and Princeton that created an inference that Facebook knew the data was a trade secret and that it had been acquired through improper means. Specifically, the complaint alleged that Facebook and Princeton shared connections between researchers and that a Princeton student who had received a “Facebook Fellowship” authored an article detailing Princeton’s acquisition of the Planner 5D dataset.

Although merely a ruling on a motion to dismiss, the decision arguably gives credence to an argument that information that is available on the internet can still be subject to reasonable efforts to protect that trade secret information (under certain circumstances). A plaintiff does not need to ensure absolute perfection, but rather must use only reasonable efforts to protect its information. Part of those reasonable efforts may be terms of use that restrict others’ right (even if not technical ability) to scrape or otherwise obtain the information off the internet. The case also serves as a caution that courts may allow trade secret cases to continue even if you were not the first one to obtain the trade secrets. Courts may consider the defendant’s relationships with others before deciding that a downstream recipient of trade secret information can get out of the case.

Copyright. The court also dismissed the copyright claim because the plaintiff did not show that it had registered the correct versions of the works.