Imagine a plaintiff in a trade secret case (you can call them OpenAI) claims that its "training data," which it used to train its top AI model, constitutes a trade secret.
The plaintiff concedes the individual components of this training data are disclosed in public sources or generally known in the field. But it alleges that the entire combination of training data is not known. It's a "combination secret."
Combination secrets, sometimes referred to as "compilation" secrets, are often protectable, so long as the owner shows the entire combination derives "independent economic value" from remaining secret. As I've discussed, training data might potentially be protectable as a combination secret.
Still plaintiff concedes that in theory a person in the field could figure out the whole combination. For example, they could exhaustively prompt the AI model itself in order to discern the training data.
Now the question becomes: Is this combination of training data "readily ascertainable through proper means"? Does it matter that the defendant in this case did not obtain the information that way, but instead acquired it through their former employer, the plaintiff? Does it matter that the only way to figure out the whole combination is through using means that are only questionably "proper"? For example: What if a computer scientist could, in theory, figure out all the training data, but only using automation or some form of advanced artificial intelligence? (The Eleventh Circuit has suggested this might itself be "improper.")
These are extremely difficult questions that surprisingly few people have tackled. In this post, I interview a major exception: Charles Tait Graves.
Graves is a parter at Wilson Sonsini, an adjunct professor at UC Law San Francisco, and an expert on trade secret law. Graves has published numerous law review articles in the field. One of his latest articles, "What is Readily Ascertainability?," was published in Spring 2025 in Tulane Journal of Technology and Intellectual Property and can be accessed here.
I also have a new article on the history and case law development of the readily ascertainable bar, "The Readily Ascertainable Bar: Then and Now." It's just been published as part of a symposium, which Ryan Vacca and I co-organized last spring, for the tenth anniversary of the Defend Trade Secrets Act. The entire issue can be accessed here. I highly recommend several excellent doctrinal trade secret law articles in this issue, by scholars such as Deepa Varadarajan (whistleblower provision), Jonas Anderson (identification), and Gregory Dickinson (independent economic value), and Shuba Ghosh (DTSA's effect on contractual restrictions on reverse engineering).
I hope you enjoy our discussion.
