Apple Says AI Can 'Learn' Trade Secrets—And Never Forget Them

Apple is warning that when trade secrets are fed into an artificial intelligence system, the damage may be permanent. The company made the argument in a court filing on Monday, as part of its lawsuit against OpenAI. Apple alleges that former employees used its confidential information to benefit OpenAI. In the filing, Apple’s lawyers wrote that if an AI agent or model “learns” from trade secret data, that learning “may create irreversible and continually propagating uses of the trade secret.” They said this type of harm is “uniquely challenging to undo.” The case centers on a former Apple employee who, according to Apple, used company secrets while working at OpenAI—including using AI agents to run simulations. This is not the first time a tech company has raised AI-related concerns in a trade secret dispute. Elon Musk’s xAI previously sued OpenAI, claiming an ex-engineer had connected his personal ChatGPT account to cloud storage containing xAI’s source code. That lawsuit was dismissed in June. Legal experts say the law already offers remedies in these cases. Courts can order a company to stop using the stolen secrets, bar further disclosure, and pay damages or royalties. But the technical side is more complicated. Sijia Liu, a computer science professor at Michigan State University, studies “machine unlearning”—the process of removing data’s influence from an AI model. He said that if a secret is stored in a file an AI can retrieve, deleting that file may be enough. However, if the secret was used to train or fine-tune a model, removing its influence is much harder. “You have to precisely define the boundary of unwanted capability,” Liu said. A more practical short-term fix, he suggested, is building detection systems that flag sensitive requests and stop the AI from responding. Apple has not said exactly how the former employee used its secrets with AI. The company declined to comment. What is clear, experts say, is that AI is creating new ways for companies to lose control of their most valuable information.