Trade Secrets Summoned to Protect AI Amid Noncompete Uncertainty

A Texas federal judge’s order halting enforcement of the Biden administration’s sweeping ban on noncompete agreements has, at least temporarily, cleared a complex path to safeguarding breakthroughs in artificial intelligence.

AI entrepreneurs have turned to trade secrets law and noncompete agreements in recent months to protect their proprietary artificial intelligence systems, and the products of AI-backed discoveries. But that strategy is threatened if the Federal Trade Commission’s ban on most noncompetes is allowed to go into effect, said Joshua Rich, partner and general counsel at McDonnell Boehnen Hulbert & Berghoff LLP.

“That’s why the combination of not being able to use noncompetes and the power of AI—making a research process that might previously taken months or years and collapsing it into hours—will create a real risk for parties who have the trade secrets now,” he said.

US District Judge Ada Brown, a Trump appointee, intervened on July 3 to reinforce, for the moment, the authority of noncompete agreements and companies’ increasing use of trade secrets to protect their AI tools from theft. The decision came two months before the FTC’s final rule against noncompetes would’ve taken effect, and some five months after the US Patent and Trademark Office issued guidance indicating AI-created tools can be patented only if a human is significantly involved in development.

The flurry of AI innovations has led to a litany of legal battles over tech ownership, especially between startups and big tech companies jostling for talent and their IP while racing to market, said Maggie Welsh, co-chair of Baker Botts’ artificial intelligence practice team.

Trade secret protections and noncompete agreements often go hand in hand. Trade secrets keep corporate IP, like new AI tools, private until a company is ready to publicly disclose their innovation, while noncompete agreements prevent employees from taking those tools with them to new jobs at competitors.

And the disputes are already here.

In March 2020, Meta Platforms Inc. had to defend itself against against claims it misappropriated AI trade secrets after hiring startup Neural Magic Inc.'s technology director who was under noncompete and nondisclosure agreements. Earlier this month, Insight Direct USA Inc. accused a former senior executive of stealing generative-AI tools and enticing coworkers to join his new startup, Fog Solutions LLC.

If an employee with access to trade secrets acts unethically by joining a competitor and offering an AI product without disclosing its origins,"it is a problem that your trade secrets will walk away,” Rich said.

Noncompetes & Secrets

Preventing employees from joining competitors is a key strategy to mitigate trade secret theft, Rich noted. He highlighted the risk posed by leveraging “human capital” and reverse engineering to exploit confidential information.

“The challenge often lies in detecting when trade secrets walk out the door,” Rich cautioned.

Some litigation outcomes may force companies relying on trade secrets to relinquish use of noncompete agreements when employees pursue new professional opportunities.

Clear employment agreements are essential to deter litigation over misappropriated trade secrets, said Matthew D’Amore, associate dean at Cornell Tech. He stressed the complexity of distinguishing between an employee’s general skill set and proprietary knowledge.

AI startup Neural Magic said it realized Meta misused its confidential information after the Facebook parent announced in December 2019 the creation of an open-source product, according to a complaint filed in March 2020 in the US District Court for the District of Massachusetts.

The startup said its technology director Aleksader Zlateski was under noncompete and nondisclosure agreements but left to work for Meta. Zlateski had access to trade secrets and confidential business plans, according to Neural Magic. It alleged Meta’s open-source product used the same algorithms as Neural Magic’s IP.

Meta denied the allegations. The companies notified the court in August 2023 they’d resolved the claims under “confidential terms.”

Jorden Rutledge, an associate at Locke Lord LLP emphasized the importance of tailored employment agreements that explicitly define trade secrets, particularly as companies navigate AI tools and the security implications of remote and hybrid work.

“With the rise of generative AI, the scope of what constitutes a ‘trade secret’ is likely expanding,” Rutledge added.

Human Involvement

When it announced its ban on noncompetes, the FTC suggested patents as another pathway companies could use to protect their inventions. It estimated the ban could lead to as many as 29,000 more patents each year for the next decade.

But courts and agencies have uniformly said the law requires significant human involvement during the invention or creation process for patent or copyright protection to apply, casting into doubt that avenue to prevent others from appropriating new AI creations.

Stephen Thaler, a computer scientist, has sought protections for AI-generated art and inventions around the world. The US Court of Appeals for the Federal Circuit in 2022 ruled definitively in Thaler v. Vidal that patent inventors must be human. The court rejected Thaler’s argument that his AI system could be considered an inventor.

Meanwhile, the US Patent and Trademark Office’s February 2024 guidance gives little clarity on what amount of input or human involvement is necessary, Rich said.

The patent office’s guidance is similar to that issued by the US Copyright Office in March 2023, stating sufficient human authorship is needed for works created by artificial intelligence to receive protection.

The US Court of Appeals for the District of Columbia Circuit in September will hear Thaler’s appeal of the US Copyright Office’s refusal to register his AI-generated artwork of a two-dimensional image of a train track running through an overpass. A district court judge in August 2023 affirmed the copyright office’s decision that the work lacked human authorship and therefore isn’t protectable.

The allure of trade secrets lies in their ability to protect confidential information without the public disclosure inherent in patent and copyright filings.

Moreover, trade secrets law doesn’t require “human authorship” like patents or for the information to be “novel” or have a “creative spark” like copyright, Rutledge said.

Companies seeking trade secret protections of AI outputs should demonstrate that the byproduct of AI has been substantively altered to create independent economic value, and that the company has taken steps to protect the information, he said.

“This all makes trade secret law a prime candidate to protect information that is derived from AI products,” Rutledge said. “However, no court has yet ruled on whether an AI’s output is sufficiently ‘secret’ to receive trade secret protections.”

That fluidity does present a risk, attorneys said. Although trade secret protections may present a temporary remedy to AI’s patentability problem, Rutledge anticipates it will face its own challenges.

“We can expect companies who are claiming trade secret protections to have their opponents argue ‘no, this can’t be a trade secret, it is from AI,” Rutledge said. That could result in “a protracted battle and discovery concerning how a company utilizes AI in their workflow, and how they used it in this specific instance.”