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.
Interview with Charles Tait Graves
By Camilla A. Hrdy
Interview Date: October 25, 2025
Updated November 15, 2026
What is "Ready Ascertainability"?
CAH: First, could you say a bit about what the readily ascertainable requirement is, and what it means for information to be readily ascertainable? I have called this the "readily ascertainable bar." You call this "ready ascertainability." Either way: What’s the basic standard?
CTG: Let me back up a bit. This is an odd phrase in the statute because it’s not defined. It’s a term of art, so it doesn’t have an obvious lay meaning, and it’s expressed in the negative: “not generally known” and, in most jurisdictions, “not readily ascertainable.” “Not readily ascertainable” can’t be a synonym for “not generally known,” or it would be redundant, so it must have independent meaning.
I started, frankly, from curiosity. We don’t see the phrase litigated much; it doesn’t come up in practice often, and yet there it is as an independent requirement for what counts as a trade secret. I didn’t know the answer when I began the article four years ago. I reviewed hundreds of cases and found that—particularly in customer-list cases—courts often use a kind of “stopwatch test”: could you “readily” gather these names with a little effort? And I found this concept has been largely cabined to customer-list scenarios, where there are endless cases, but not much used in other contexts.
So, to answer your question: “generally known” means it’s just out there—you can point to a single source (such as a scientific article). “Readily ascertainable” means it’s not in one place, but with a minimal amount of time and effort, one can gather it.
CAH: Great—and a quick follow-up. Who is the relevant “person” for whom the information cannot be readily ascertainable?
CTG: Good question. There’s a split in the case law about whether the test is objective or subjective. Some cases ask, even if the defendant actually took the information (say, downloaded it), should we apply an objective test—how long would it take a person in the field to gather it? Or a subjective test—did this defendant do it? I think the latter is a mistake: it conflates readily ascertainable with independent development. The standard should be objective: a person in the industry.
CAH: Pop quiz: if it would take two days to figure out the secret, is that readily ascertainable?
CTG: We can’t answer without the context. The type of information is crucial. We can imagine time-sensitive information where even two days would destroy its value; in that scenario, two days might not be “readily.”
CAH: In deriving a standard, should the relevant factor be (a) how much the plaintiff invested, or (b) how long it would take an average person in the field to ascertain it? In other words, is this requirement just a stand-in for the plaintiff’s level of investment?
CTG: The statute is silent; courts aren’t clear. But it shouldn’t be just plaintiff’s level of investment. Investment is a poor proxy. Comparing a large, cumbersome firm to a fast-moving startup shows why investment is unreliable.
CAH: Because the investment required by the plaintiff to develop the information might not be the investment that would be required by most others. We would over protect if we said it’s not “readily ascertainable” just because the plaintiff spent a lot in development.
CTG: Yes, so in my view it should be an objective, industry-focused test—not an investment proxy. To be clear, though, cases aren’t settled on this, and the statute doesn’t say.
Variation in Adoption Among the States
CAH: The DTSA and the UTSA’s uniform version include the not-readily-ascertainable requirement, but not all states do.
CTG: Right. California famously left it out. Illinois did as well, and Nebraska has its own unusual version.
CAH: And New Jersey has a unique provision we flagged on email. New Jersey Trade Secrets Act Section 5 states:
"A person who misappropriates a trade secret shall not use as a defense to the misappropriation that proper means to acquire the trade secret existed at the time of the misappropriation."
So in New Jersey a person who misappropriates may not use as a defense that proper means were available? Even though the statute contains the language. The legislative history is thin and doesn’t explain how those terms got in.
No one really knows about this New Jersey provision! But California's stance is very well-known. And one of your interesting points is that California’s approach may have affected the rest of the law. Because California permits protection for readily ascertainable information, you suggest this has stymied development elsewhere.
CTG: Yes. California generates an enormous number of trade secret cases. Because California left out “readily ascertainable,” the issue doesn’t surface much there. When the jurisdiction with the most case law isn’t discussing the concept, that leaves a gap—and other courts don’t have California opinions to consult on this particular question.
CAH: In my research, I also found lots of California customer-list cases, but not tech cases (unsurprisingly). You say the limitation is overlooked—but only in some kinds of cases. Why?
CTG: It tends to be applied in “small town customer list” cases—older cases where courts cite each other using the same pattern. But in technology cases, there isn’t a tradition of using the language, so courts may not be spotting it. I suspect judges look for similar fact patterns and replicate what they see.
Confusion Between "Readily Ascertainable" and Reverse Engineering as Defense
CAH: Let’s talk about conflation of ready ascertainability with other concepts. You suggest courts conflate “readily ascertainable” with other ideas, especially reverse engineering - both reverse-engineerability and whether the defendant did reverse engineer. Say more?
CTG: This was one of the biggest descriptive findings. Courts use “readily ascertainable” to mean other things: independent development, reverse engineering, even the plaintiff’s own publication of the information (which properly goes to reasonable measures). I’ve seen it used in the “employee general skill and knowledge” context, too. So, courts appear confused: they use the phrase repeatedly for different, established concepts with their own meanings.
CAH: You also hone in on confusion between a subject-matter bar (trade secret status) and how the defendant acquired the information. In theory, “how and from whom” shouldn’t matter to the subject-matter question—but courts sometimes let it matter.
CTG: Exactly. Reverse engineering and independent development are lawful acquisition modes even if the information is a trade secret. Some courts use “readily ascertainable” and mix up the objective definition of what qualifies as a trade secret with subjective questions about how the defendant got it. In one case you noted (AvidAir), the court essentially hedged: uncertain about secrecy, it pivoted to how the defendant didn’t get it and used that to support the result. There’s also surprisingly little independent-development case law, which may contribute to the confusion.
Combining Sources as Means of Establishing Ready Ascertainability
CAH: A key theme in your article is that combining sources should be a means of establishing ready ascertainability.
You suggest courts are not applying the readily ascertainable bar frequently enough, and that the problem is most acute for so-called "combination secrets," which are trade secrets made up of combinations of pre-existing elements.
So imagine the trade secret is "ABCD." Each element, A, B, C, and D is available in the literature on its own, but it’s the combination that is claimed as the trade secret.
Or, relatedly, imagine you have a "modification secret," where the trade secret encompasses a very minor difference on a public source, like an update on a prior version of a document that is now public.
You discuss modifications in Avid Air v. Rolls-Royce, where the court held a minor modification on a public version of a document describing an FAA-approved overhaul process could be protected, even though the changes were “relatively minor in the context of the overhaul process as a whole.”
You suggest the primary purpose of the readily ascertainable limitation is to avoid protecting information that, even if not all contained in a single public source, could be combined from multiple sources.
Just to spell this out, the public source is “generally known," it's all in one place. The combination or modification on the public source isn't generally known, but it still can be deemed “readily ascertainable through proper means” if it's easy for someone in the industry to figure it out through reviewing what's known.
Have you always held that view?
CTG: I think it has to be part of the concept of ready-ascertainability. In customer-list cases the question is: could you go to more than one place and easily put it together? That said, technology cases are harder. I’m not challenging valid combination trade secrets; I’m targeting misuse in litigation. The readily-ascertainable requirement can help courts police that.
CAH: This sounds a lot like patent law’s obviousness standard to me, when you phrase it that way. Because obviousness is also, per cases like Hotchkiss v. Greenwood, about combining sources, right? You have a doorknob in the prior art. You have the material porcelain in the prior art. The plaintiff tries to patent a porcelain doorknob. It’s "obvious" because people with ordinary skill in the art would think to make that combination.
And yet you are very skeptical of this analogy. You say that the patent law 35 USC § 103 standard—whether the person having ordinary skill would find it obvious to combine the sources—is actually not really helpful here. Why not? How do they differ?
CTG: Many in our audience know patent law better than trade secrets, so they instinctively map patent concepts over. But the bodies of law differ. I collected cases where courts explicitly reject patent analogies. Obviousness asks a hypothetical PHOSITA question about whether a claimed invention would have been obvious given defined prior art. Readily ascertainable asks how easily someone could put together public sources with minimal time and effort. A Ninth Circuit case involving KFC (Hutchison) explains why they’re different. They’re not night-and-day opposites, but they’re not the same. Litigators can’t cite patent cases and expect traction in a trade secret dispute; you have to speak trade secret’s language.
Note (Nov. 15, 2026): The case he referred to is Hutchison v. KFC, 51 F.3d 280 (9th Cir. 1995). The court states, in relevant part:
"The threshold for "ready ascertainability" is not a high one: the process need not satisfy the "non-obviousness" standard necessary for patent protection. Nevertheless, if a process is generally known in the industry, or readily ascertainable from public sources, [the Nevada UTSA] does not protect it even if it was previously unknown to the recipient."
CAH: I’d add: ready ascertainability in trade secret feels more literal. It's: Can someone go get and combine the sources? In patent law obviousness is a purely theoretical assessment. Once the § 102 prior art is set, you ask: Would the differences from the prior art be obvious to a person having ordinary skill in the art, the PHOSITA, in theory, given that this PHOSITA, in theory, has all the sources that are known?
CTG: That’s fair, and with one important caution: Readily ascertainable in trade secret law doesn’t mean you can "Frankenstein" together 20 unrelated papers post hoc during litigation and declare the “combination.” The connection among sources matters. The sources should be reasonably related and easily obtainable; we’re not just rewarding clever reconstruction after the fact.
CAH: Your normative proposal seems modest: use readily ascertainable to shine a flashlight on improperly constructed combination trade secret claims.
CTG: Exactly. I’m not rejecting valid combination trade secrets. I’m concerned about litigation tactics that string together unconnected elements (A, H, L, P) and call it a “combination” simply because both sides share a few isolated features. Readily ascertainable helps courts address that misuse.
CAH: You call these “gerrymandered” trade secret claims, where the plaintiff highlights similarities (ABCD) and ignores differences in larger, complex products. But why is that wrong if misappropriation ultimately asks whether the defendant derived the trade secret from the plaintiff? We use a sort of fruit of poison tree assessment, as Mark Lemley calls it.
CTG: Because “ABCD” suggests a rational interrelationship. The gerrymander move is to pluck unrelated elements that don’t interrelate scientifically and re-label them a “combination.” That departs from real-world unified processes and invents a litigation-driven combination. If each isolated element is easily found in public sources, the readily ascertainable lens helps expose the tactic—even if a court declines to adopt a special “unified process” rule for combinations.
The History of the Readily Ascertainable Bar
CAH: I loved your section on the UTSA drafting history, but I disagree with some of your conclusions about what the drafters were thinking. You note that UTSA drafters sometimes treated reverse-engineerability and readily ascertainable as the same thing. I think that was by design. I’ve argued in my new historical paper that the drafters were very focused on products that could be easily reverse engineered, and that’s originally why the language was included.
I have argued, in another article forthcoming in BTLJ, that the drafters conceptually compared this issue to an on-sale bar, and were thinking of Kewanee, because the Sixth Circuit had held that a trade secret that is placed on sale and so barred by patent law cannot be protected as a trade secret. The Supreme Court said, this is ok, so long as the sale doesn't really render the trade secret public. The dratfers wanted to avoid all the new state laws being preempted by patent law, so they made sure easily-reverse-engineerable trade secrets would be held "readily ascertainable by proper means." That is why they included the language in the first place!
So we may disagree on the history here. I do not disagree with your argument that ability to easily combine sources can also render trade secrets readily ascertainable, but I don't think the drafters were confused. I think they were focused on trade secrets that could be reverse engineered from products because of all the common law cases they were addressing.
CTG: I agree they were working in the shadow of preemption and wrestling with core definitions. My criticism is that even in the 1960s–70s, reverse engineering (taking something apart to learn how it’s made) and readily ascertainable (how quickly one can connect related public sources to show the information isn’t a secret) are distinct. Yet the drafters sometimes used the terms interchangeably—much like courts have since. Greater definitional clarity would have helped. Some UTSA variants explicitly mention reverse engineering elsewhere in the statute, confirming the terms aren’t synonyms—even if the model didn’t define both.
CAH: Good. Ok. So speaking of the history, who exactly bears the burden to prove information is not readily ascertainable? What did the drafters say? I found they originally included a provision relaxing the burden, but deliberately removed it, because many drafters were worried plaintiff would then not have the burden to prove their trade secret was not generally known or readily ascertainable. They removed it.
CTG: The burden should be on the plaintiff—both production and persuasion—to show a trade secret exists, which includes not being readily ascertainable. California, which omits the requirement, treats “readily ascertainable” more like an affirmative defense; the UTSA drafters at one point toyed with burden-shifting but ultimately did not adopt it. The legislative materials are opaque; you can assemble puzzle pieces (industry letters, bar comments), but there’s no clear statement that the drafters intended an affirmative defense. My reading: they didn’t.
CAH: 100% agree.
Trade Secrets & Artificial Intelligence
CAH: Turning to artificial intelligence. David Levine, myself, and others suggest generative AI could render many trade secrets readily ascertainable.
Levine, in an absolutely fantastic and original article called Generative Artificial Intelligence and Trade Secrecy, suggested that generative AI could potentially spell the end of many trade secrets, because employees may use generative AI tools on the job and reveal the trade secrets of their employers, or because "bad actors" might strategically query generative AI tools to extract competitors' secrets.
For example, Levine observes that,
"if a trade secret owner places confidential information into ChatGPT’s training data (wittingly or unwittingly) through a query or some other action, that could deviate sufficiently from “reasonable efforts” to maintain a trade secret as to vitiate trade secret protection all together." (Levine, 564).
To me that makes sense. Some companies, some employees, just might not be careful enough. They could spill secrets to ChatGPT, just like they can spill secrets to the internet. That might fail "reasonable measures," as Levine suggests, and also render the secrets "readily ascertainable," yes?
We’ve talked about this before, and you mentioned you were skeptical that employees are spitting out their employers secrets to the world, yes?
CTG: I am more skeptical about this concern than some others. Companies just buy Enterprise Licenses. Those have terms that ensure the models won't be trained on the information in a way that threatens confidentiality. Their data isn’t used to train public models or leaked to third parties.
When take that into consideration, the concept that this going to vitiate trade secrets is doubtful.
Could an employee go rogue with a consumer tool? Perhaps. But we need empirical evidence that this actually exposes secrets in replicable ways.
CAH: OK, yes, I agree about all of that. Sophisticated companies will get Enterprise Licenses and obtain additional confidentiality promise.
I've written a lot about the confidentiality obligations in the ChatGPT Enterprise License, which I think are very important for preserving secrecy. (here).
But the "readily ascertainable" concern does not necessarily rely on an employee spilling the precise secrets and then others learning about that.
This concern is more apocalyptic, right? It’s that the generative AI tool can "readily ascertain," and tell humans, information that would otherwise be someone's trade secret.
I've made this argument a bit in my own work, building on Levine (and Ryan Abbott in the patent realm) though without testing it empirically.
The question I've asked it, is
"What if generative AI models get so good at searching available data and piecing together information that they can figure out companies’ trade secrets—even without being fed inside information, and even without 'bad actors' making deliberate attempts to discover trade secrets? Could that render companies’ trade secrets suddenly 'readily ascertainable'?" (Hrdy, 330).
This has special application for combinations, right? The concern is: "Well, you have ABCD combination secret, but AI cancombine data and even predict the precise combination that's said to be protected. So it's now readily ascertainable."
Do you agree with this concern?
CTG: I’m skeptical of the “mass extinction of trade secrets” claim. Historically, you would have to show objective sources that anyone can consult to establish something is readily ascertainable.
CAH: Like a patent or a dictionary.
CTG: Yes. With AI, a one-off chatbot output isn’t necessarily replicable or reliable. Hallucinations happen. Without objective, repeatable evidence tying outputs to public sources, I’m doubtful that a chatbot answer alone proves information is readily ascertainable.
CAH: Fascinating point. I am not sure I have heard that precise objection. What about using AI in court with expert testimony—feeding known sources and asking the model to synthesize? Would that be relevant?
CTG: Maybe as a demonstrative aid, but it’s risky. You can’t run uncontrolled experiments live. And even if you did, you’d still need to ground the result in objective, replicable sources that everyone can examine. Competing experts already disagree on “two weeks vs. two months.” The key is that the sources are stable and public; a unique, non-replicable chatbot output doesn’t meet that bar. Separate scientific questions (like whether AI accelerates discovery) are different from evidentiary ones about what counts as objectively, readily ascertainable.
CAH: This is great—and timely. I haven’t seen much on the evidentiary value of LLM outputs for ascertainability. It's unsettled. So I take your point that how we use AI as a basis for evidence of non-secrecy is up for grabs.
CTG: Until models are demonstrably reliable and outputs are repeatable from identified public sources, I’d be cautious about using them to prove ascertainability.
Note (Nov. 15, 2026): We do now have at least one district court suggesting that feeding to public-facing AI with no confidentiality measures destroys trade secrecy. There is also a case suggesting on summary judgment that if AI can figure the information out it may be readily ascertainable. In this case, the asserted secret was code. The defendant (former employee who sent code to himself on way out) pointed out that "ChatGPT could now recreate the Python Code that he put together. There is no genuine dispute that the Python code [defendant] sent himself is readily ascertainable. Thus, the court concludes that the Koch's Python code scripts do not qualify as trade secrets."
Conclusion
CAH: To wrap up, your article is both descriptive and I think very normative. First, it traces how courts have used—and misused—the phrase “readily ascertainable,” and second, it proposes using the concept to police improperly constructed combination claims in tech cases.
CTG: Exactly. The main value is clarifying use and misuse across the case law and offering courts a practical tool—without requiring radical doctrinal change.
CAH: Wonderful. Thank you so much—this was terrific.
