Trade secrets in the age of artificial intelligence

In recent years the trade secret has become one of the tools most used by companies that build artificial intelligence to limit access to information about how their systems work. It was conceived to protect formulas, production processes and technical know-how that gave a competitive edge. It is now invoked more and more often to shield essential components of AI technology from independent verification.

In the context of AI, trade secrets can cover a wide range of elements: the algorithms used to train and run models, software architectures, internal parameters, optimisation procedures, content filtering techniques and, above all, training data. The ways in which that data is collected, selected and processed are also often treated as confidential. In many cases companies claim as trade secrets even the criteria by which a system takes certain decisions or produces certain outputs.

From a legal standpoint, trade secret protection answers a legitimate need. Companies invest heavily in ever more sophisticated models and have an interest in preventing competitors from appropriating the results of that investment for free. The problem arises when confidentiality collides with matters of public interest: transparency, accountability, and the ability to subject systems to independent scrutiny.

The risk is that a tool created to protect innovation becomes a barrier against external control. Without adequate mechanisms for access and oversight, industrial confidentiality ends up overriding transparency and makes technologies that are already complex even harder to understand, assess and challenge.

In the United States, however, federal courts asked to rule on trade secret misappropriation claims increasingly want clarity on what, precisely, is secret.

Trinidad v. OpenAI Inc. (No. 25-cv-06328-JST), decided in January 2026, is a telling example. Rebecca Trinidad, who represented herself, sued OpenAI in a complaint filed on 3 July 2025. She alleged that OpenAI had appropriated her "proprietary methodologies" and her protocols and frameworks for AI development, including those for "emergent identity" and "autonomous multi-agent collaboration", which she said she had developed through her interactions with ChatGPT. Her premise was that, by using ChatGPT, she had substantially improved its functionality, and that OpenAI, by monitoring her use, had then exploited and commercialised those innovations.

The court first examined the copyright claims. The plaintiff herself acknowledged that she had registered no work. Even had she shown that she owned the works, registration is a requirement for bringing a copyright infringement action, and there was none.

What remained was the claim of trade secret misappropriation under the Defend Trade Secrets Act. To succeed, a plaintiff must prove three things: that she possessed a trade secret, that the defendant misappropriated it, and that the misappropriation caused or threatened to cause harm.

The court dismissed the claim. When a party discloses her secrets to another, here ChatGPT, which is under no obligation to protect them, her property right is extinguished.

The case concerns a litigant without legal representation and should be read with caution. But it points to a concrete risk: developing or sharing proprietary information through AI platforms, without contractual confidentiality protections, can destroy trade secret status. It also shows that generic assertions about the secrecy of AI technologies are unlikely to hold up in court.

Every platform has its own Terms of Use. They should be read and understood before confidential information is entered into it, not after.