AI Will Force Trade Secret Calculus Shift, Escalate Tactics

The rise of artificial intelligence is changing the landscape for creating, defining and protecting trade secrets just as the advent of the internet did a quarter-century ago.

The ways generative AI can create, destroy, protect, and steal trade secrets will inevitably alter legal thresholds for defining and litigating that intellectual property, attorneys say. The leaps in AI’s ability to process mass quantities of data will broaden what information is considered too easily obtained by others to be legally protected, trade secrets attorney John F. Marsh of Bailey Cavalieri said. That could neutralize entire categories of secrets.

“Many of us believe that the notion of a customer list as a trade secret is on borrowed time,” he said. “How long will it be until AI figures out from the basic ingredients on the can how to replicate Coca-Cola?”

AI-related secrets are already at the center of lawsuits that mirror more traditional trade secrets cases, as in X.AI’s lawsuit accusing OpenAI Inc. of poaching its staff. But more interesting questions arise when AI intersects with different stages in the lifecycle of trade secrets, affecting what information is protectable and how hard companies have to work to keep it secret—two key questions in trade secrets law—attorney Sarah Tishler of Beck Reed Riden LLP said.

“These big picture questions will take a while to percolate through the courts, but it’s just a matter of time before we see more definitive case law on it,” Tishler said. “In the 1990s, commentators feared the internet would also destroy trade secrets. But they raised the bar for protection, and I think AI is going to have the same effect.”

Courts will soon wrestle with questions such as how exactly AI affects what constitutes “readily ascertainable” information that therefore isn’t secret enough for protection. Companies and their attorneys may not have to wait long to start to get those answers.

“It’s just a matter of time,” Avery Williams of McKool Smith said. “If we don’t see something in the next six months I’ll be quite surprised.”

Evolving Landscape

A trade secret is proprietary information that derives its economic value from remaining secret. Companies forfeit legal protection if they fail to take reasonable measures to maintain that secrecy, a standard that requires examining context in a world where context is changing rapidly. A theft claim requires proof that the information was acquired by improper means. Merely reverse-engineering a product or piecing together public information—things AI can do prolifically—isn’t a violation.

AI is the latest in a string of technologies that routinely change this trade secret calculus. The Fifth Circuit in 1970 found hiring an aerial photographer to capture the layout of a plant’s construction site might constitute “improper means” of obtaining secret information. Since then, online maps and other advances have made formerly concealable information “readily ascertainable.”

In 2024 the Eleventh Circuit said “scraping and related technologies” to acquire information online may be “perfectly legitimate” in some cases, trade secrets attorney John F. Marsh of Bailey Cavalieri noted.

AI platforms also represent a place where secrets can go to die. One district court already suggested inserting a trade secret into public-facing AI tool means the “property right is extinguished”, IP law professor Camilla Hrdy of Rutgers University noted. OpenAI Inc. thwarted a trade secrets claim in January, as a California judge reasoned the plaintiff disclosed her secret “to others under no obligation to protect” it when she put it into ChatGPT.

Companies have begun reevaluating how they protect their secrets, trade secrets attorney Matthew D. Kohel of Saul Ewing LLP said. After companies including Samsung reportedly had multiple employees enter proprietary information into AI chatbots, many have shifted to bespoke enterprise AI platforms and banned using public ones, he said.

“People are becoming pretty sophisticated about building their own sandboxes,” Pooley said.

Start-to-Finish

The black-box nature of AI—even models’ owners and creators don’t know how the constantly evolving platforms arrive at a particular answer—will further complicate questions around the creation, protection, and misappropriation of trade secrets.

More effort will inevitably be needed to show sufficient steps were taken to reasonably protect secrets, trade secrets attorney James Pooley of James Pooley LLC said. He pointed to cybersecurity’s impact on trade secrets over the last few decades as a template for forthcoming litigation. As secret information became easier to obtain through hacking, the threshold for what cybersecurity measures convey legal protection increased.

Since the 1990s, bad actors and cybersecurity professionals have pushed each other to increase their sophistication. That dynamic will also be true of AI, Pooley said. “One lesson is there will be no final answer, because, given the incentives, the arms race will continue.”

Data loss-prevention systems will be crucial to protect the building blocks that AI tools can organize to create—and potentially undermine—secrets with minimal effort, he said. But AI can also help protect trade secrets by helping track and manage proprietary information, he said.

Tishler said there are already powerful monitoring systems that can aid early detection of potential trade secrets theft. Whereas before you’d have to comb through access logs and print histories after the fact, now certain activities can trigger notifications or even shut down access before the theft can be completed, she said.

Creating a trade secret, unlike a patent or copyright, doesn’t require human involvement, meaning AI could also become a prolific creator of trade secrets.

But determining whether those secrets are too “readily ascertainable” to be protectable and whether the use of AI to discover somebody else’s secret is improper will often be “a pretty fact-intensive inquiry over what inputs were used to produce the result,” Williams said.

“If it’s a guy coding in his basement using Claude,” it would be hard to claim the information wasn’t “readily ascertainable,” Williams said. “On the other hand, a team of research scientists using their own GPT tweaked specifically for this is a totally different story.”