A fascinating February 26 decision issued by the Federal Court sitting in the Northern District of California in X.AI Corp. v. OpenAI, Inc. - and well before it addressed serious questions as to whether Generative AI can be protected as trade secret under New York law in the first instance - considering, after all, that the architects of these systems 1) aren't the ones continuing to generate it; and, 2) the designers/owners of these companies themselves aren't 100% certain how their own AI works.
I suppose that more pressing/vexing question will have to be addressed at a different time and date.
But make no mistake: that day is certainly coming, and sooner than most of us think.
In order to better appreciate the problems, and evidentiary hurdles, that this creates, it's best to take a step back and analyze these challenges against the backdrop of the legal requirements in a garden-variety trade secret theft case.
How a Former Employer Usually Establishes Trade Secret Theft
While trade secret theft claims may afford claimants an assortment of potential legal avenues to pursue recovery (or, in legalese, "causes of action"), they tend to fall primarily under the umbrella of unfair competition, which, in general terms, requires that the plaintiff demonstrate the following:
1. That they possessed a trade secret;
2. That the defendant used that trade secret in breach of an agreement, confidential relationship or duty or as a result of discovery by improper means; and,
3. That the plaintiff therefore suffered damages.
See, e.g., Photonics Industries International, Inc. v Xiaojie Zhao, 185 AD3d 1064, 127 NYS3d 568 (2d Dept 2020); Schroeder v Pinterest Inc., 133 AD3d 12, 17 NYS3d 678 (1st Dept 2015).
That said, all of the foregoing is contingent on the plaintiff being able to establish that this information was, in the first instance, a trade secret, which is no small feat.
Consider:
In order to qualify as a trade secret under New York law, the plaintiff must satisfy a multi-part test that was laid out by New York State's Court of Appeals (the highest court in the State), including the following:
1) the extent to which the information is known outside of the business;
2) the extent to which it is known by employees and others involved in the business;
3) the extent of measures taken by the business to guard the secrecy of the information;
4) the value of the information to the business and its competitors;
5) the amount of effort or money expended by the business in developing the information; and,
6) the ease or difficulty with which the information could be properly acquired or duplicated by others.
Ashland Management Inc. v Janien, 82 NY2d 395, 604 NYS2d 912, 624 NE2d 1007 (1993).
Practically speaking, the courts have generally grouped trade secret claims into one of two different categories: customer lists, or special knowledge or information that relates to the manufacturing or business operation itself.
For our purposes in this article, AI clearly falls into the latter category.
How Plaintiffs Usually Prove Trade Secret Theft
Traditionally, plaintiffs seeking to establish that defendants should be held liable for trade secret theft rely on one or more theories that are grounded in circumstantial evidence to demonstrate that it is more than reasonable to infer that but for a misappropriation of their confidential, trade secret information, the defendant could not have developed its competing product and/or certainly not in the amount of time they managed to do so.
But that avenue seems to be far more difficult to navigate when it comes to AI, considering that which was set forth above, i.e., that the architects of these AI systems 1) aren't the ones continuing to generate it; and, 2) the designers/owners of these companies themselves aren't 100% certain how their own AI works.
But, at least as of last month, the Federal Court didn't need to reach that legal issue.
What the California Court Held in X.AI Corp. v. OpenAI
In taking the unusual step of dismissing a case at its initial pleading stage, the Court noted that while a number of seemingly connected facts might point to liability for the plaintiff's former employees, the complaint, as drafted, lacked sufficient allegations to sustain a claim against OpenAI, who was the only defendant named in the case, stating:
Notably absent are allegations about the conduct of OpenAI itself. xAI does not allege any facts indicating that OpenAI induced xAI’s former employees to steal xAI’s trade secrets or that these former xAI employees used any stolen trade secrets once employed by OpenAI. While xAI may state misappropriation claims against a couple of its former employees, it does not state a plausible misappropriation claim against OpenAI, which is the sole defendant in this case.
The Court continued as follows:
xAI does not offer any nonconclusory allegations that OpenAI itself acquired, disclosed, or used xAI’s trade secrets. Instead, its allegations of direct misappropriation appear to proceed entirely on a theory that OpenAI induced xAI’s former employees to misappropriate its trade secrets. But xAI offers no nonconclusory allegations that permit a reasonable inference of inducement ... when taken together, as true, and with all reasonable inferences drawn in favor of xAI, xAI’s allegations do not state a plausible claim based on inducement. Yes, several employees took its confidential information at around the same time and before leaving to work for OpenAI, and two of those employees communicated with the same OpenAI recruiter using an encrypted messaging application around the time of their exfiltration and their exploration of a move to OpenAI. Without more, though, those allegations are insufficient to support a plausible inference that OpenAI encouraged the alleged theft.
At its conclusion, the Court gave the plaintiff leave to amend its complaint to try to remedy the shortfall of facts in its pleadings. I suspect, however, that if they had "the goods" on that score, they would have included it in the first go-around.
