Apple says a former engineer fed its circuit data into an AI agent at OpenAI

Apple just escalated its trade-secrets fight with OpenAI again, and this time the filing is about more than hiring drama. According to Apple, a forensic look at a MacBook used by former Apple engineer Chang Liu shows he did not just take confidential hardware files with him. Apple says he used at least one of those files in his work at OpenAI, including by training an AI agent to run electrical simulations around it.

Here is the timeline. Apple sued OpenAI and former Apple employees Chang Liu and Tang Yew Tan in July 2026 in the U.S. District Court for the Northern District of California. Liu spent about eight years at Apple as a senior system electrical engineer before leaving for OpenAI in January 2026. Apple's original complaint alleged that Liu kept Apple equipment, did not complete normal exit steps, and later obtained confidential engineering material after he was already gone, including by exploiting access Apple says he should not have had.

The new filing, reported August 31, 2026, focuses on what Apple claims it found after Liu's lawyers turned over the MacBook he used after leaving Apple. Apple says an initial forensic analysis supports four points: Liu downloaded a confidential Apple circuit schematic and used it in his OpenAI work; Liu and others at OpenAI allegedly knew about unauthorized access to Apple's third-party cloud storage; after learning Apple was investigating him, Liu allegedly told an OpenAI colleague to destroy evidence and she confirmed she would; and Liu used a tool at OpenAI with the same name as an internal Apple engineering application.

The AI angle is specific. Apple alleges Liu downloaded the circuit schematic in March, two months after leaving Apple, then ran a simulation with it in LTspice, a common electrical engineering tool. In messages around that time, Liu described training an AI agent to run LTspice, inspect results, and tune compensation parameters. One quote attributed to him in the coverage says the agent learned to run LTspice and look at results within an hour, cutting work that once took him a day down to about two hours. Apple says it learned about that use because the schematic was run on a Mac mini that later synced, via iCloud, to the MacBook Apple inspected. Apple now wants access to that Mac mini as well.

This matters for AI beyond the courtroom theater. Apple is arguing that once trade-secret material is fed into an AI agent or model that learns from it, the use can become harder to unwind, what the company frames as potentially irreversible and continually propagating. That is a serious claim, and it is still a claim. Courts have not ruled that Apple's account is proven. OpenAI has asked the court to dismiss the lawsuit as meritless, and it has publicly said it does not have or want Apple's trade secrets. A hearing is set for October 1, 2026 before Judge Edward J. Davila.

My take is pretty simple. If Apple's forensic account holds up, this is not a vague "people left and vibes leaked" story. It is an allegation that confidential hardware design data moved into an AI-assisted engineering workflow at a competitor. That is exactly the kind of AI risk people should take seriously: not sci-fi takeover, but ordinary trade secrets getting copied into agents that iterate faster than a human alone. Separate Apple's allegations from proven facts until discovery and the court say otherwise. But if the MacBook evidence is as concrete as Apple describes, the industry should pay attention to how AI agents handle proprietary files, not just how impressive the demos look.