Apple warns AI 'learning' makes trade-secret theft irreversible

Apple's lawyers have a warning for the AI era: once a trade secret goes into a model, it may never come out. In a supplemental brief filed Monday in support of its request for expedited discovery in its trade-secret lawsuit against OpenAI, Apple argued that when confidential information is fed into an AI agent that "learns" from it, that learning "may create irreversible and continually propagating uses of the trade secret - harm that, at a minimum, is uniquely challenging to undo and requires prompt investigation." The word "irreversible" is doing heavy lifting here, and that is the point: unlike a stolen document, a secret absorbed into a model's weights cannot simply be handed back.

The filing centers on a former Apple employee who, according to Apple, used company secrets while employed by OpenAI and used AI agents to learn to run simulations. Apple's attorneys framed the conduct as "extending beyond ordinary document theft," arguing that AI systems present a fundamentally different problem from a departing employee who simply copies files. With AI, the secret is not just taken; it is absorbed, processed, and potentially redeployed in ways that are difficult to trace, let alone reverse. The company is asking a court to treat this as a new category of harm, one that existing legal frameworks were not designed to address.

The legal question is not entirely new. Companies have long fought over employees who carry knowledge in their heads to a competitor. But Camilla Hrdy, a law professor at Rutgers who studies trade-secret law and generative AI, told Business Insider that AI changes the calculus. "Employees are already real loose cannons, walking around with knowledge in their heads," Hrdy said. "Now they're taking that knowledge and plugging it into AI, and that could be a real loss of control. That is new." The novelty cuts both ways: it makes the harm harder to prove and harder to remedy.

Apple is not the only company wrestling with this problem. Elon Musk's xAI sued OpenAI last year, alleging that former xAI engineer Xuechen Li had the company's entire codebase in his personal cloud storage and connected his personal ChatGPT account as a source. The lawsuit, which a judge dismissed in June, claimed OpenAI had a means to access the stolen source code through its ChatGPT service. The dismissal suggests that courts are still figuring out how to handle AI-adjacent trade-secret claims, leaving companies with more questions than answers about their legal recourse.

The technical remedies, meanwhile, are uneven. Sijia Liu, a computer science professor at Michigan State who co-authored a paper on "machine unlearning," told Business Insider that the fix depends entirely on how the secret was used. If a sensitive document sits in a repository that an AI system retrieves from, deleting the file may be enough. But if the information was used to train or fine-tune a model, the process becomes far more complicated. "The second case could be more difficult because the influence of something is really difficult to evaluate," Liu said, adding that "you have to precisely define the boundary of unwanted capability."

Liu suggested a more practical near-term approach: build detection systems that flag sensitive requests or information passing between agents, triggering a hard stop before the data propagates further. He acknowledged this is not "true unlearning," but it is a workable stopgap. Apple did not specify in its filing how the former employee may have used the trade secrets with AI, and an Apple spokesperson did not return a request for comment. That silence leaves a critical gap in the record, one that could matter if the case proceeds to discovery.

For boards and executives, the filing is a reminder that the tools employees use every day are creating new categories of legal exposure. The old playbook of non-disclosure agreements and exit interviews assumes secrets can be contained once identified. AI breaks that assumption. Companies may need to consider how their data is being used in AI systems, what happens to that data if an employee leaves, and whether their legal teams are prepared to litigate disputes that hinge on how a model was trained.

The strategic stakes extend beyond the courtroom. If courts accept the argument that AI "learning" constitutes irreversible harm, companies could face pressure to disclose AI usage in due diligence, employment contracts, and vendor agreements. Conversely, if courts are skeptical, companies may have little recourse when their secrets end up embedded in a competitor's model. Either way, the Apple filing signals that the next frontier of trade-secret law will be fought not in the server room, but inside the weights of the models themselves.