Apple says ex-engineer used its schematic for OpenAI hardware work
A forensic review of Chang Liu's Apple-issued MacBook has become the iPhone maker's main argument for accelerating discovery.
By Ryan Merket · Published
Primary source: MacRumors
Why it matters
Apple is trying to establish that an AI agent can carry a trade secret beyond the original file, widening the discovery and containment risks for companies using agents in engineering work.

Apple said former engineer Chang Liu used a confidential power-converter schematic in his work at OpenAI and trained an AI agent to operate the software used to simulate it, according to a supplemental federal court filing submitted Monday.
The August 31st filing gives Apple its most specific allegation yet that information Liu accessed after leaving Apple was applied inside OpenAI's hardware operation. The filing, first reported by MacRumors, is Apple's latest attempt to persuade U.S. District Judge Edward J. Davila to authorize expedited discovery before an October 1st hearing.
Apple sued Liu, OpenAI hardware chief Tang Yew Tan, OpenAI and io Products on July 10th. Liu had spent eight years as a senior systems electrical engineer at Apple before joining OpenAI in January. Tan spent 24 years at Apple and worked on the iPhone, Apple Watch and iPod before co-founding io with Jony Ive, Scott Cannon and Evans Hankey. The io team merged with OpenAI in July 2025 to develop AI-native consumer devices.
The court has not ruled that Liu, Tan or OpenAI stole or used Apple's trade secrets. The allegations in Monday's filing come from Apple's initial forensic examination of an Apple-issued MacBook that Liu retained after his departure and returned after litigation began.
A schematic, a simulation and an AI agent
Apple said the MacBook contained synced evidence showing that Liu downloaded a confidential circuit schematic in March, roughly two months after leaving Apple. According to the filing, Liu opened the schematic on a Mac mini and used LTspice, a circuit-simulation program, to run power-conversion development work.
Messages cited by Apple show Liu describing an AI agent that learned to operate LTspice, inspect results and adjust a compensation parameter. In one message, Liu wrote that he was "feeling AI all day long" and said the agent had compressed work that previously took a day into about two hours, including the time required to learn the process.
Apple argues that the combination changes the urgency of the case. Its lawyers contend that confidential engineering information used with an agent may produce learned behavior that is difficult to identify, contain or remove later. The public filing does not establish which AI system Liu used, whether information from the Apple schematic entered a model's training data, or whether the agent retained details from the circuit after completing the task.
Those distinctions will matter. Running engineering software through an agent is different from training a general-purpose model on a proprietary design. Apple's filing links the schematic, the simulation and the agent-assisted workflow, while leaving discovery to determine what OpenAI systems received and retained.
The Mac mini is now a central target of Apple's discovery request. Apple said activity on that device synced through iCloud to the returned MacBook, allowing its investigators to reconstruct parts of Liu's work. Apple wants access to the Mac mini and other devices still held by Liu.
Apple alleges evidence was at risk
Apple also accused Liu of discussing the destruction of evidence after learning in June that Apple was investigating him. The filing says Liu and an OpenAI colleague discussed restoring and then using Apple-owned devices, actions Apple argues could overwrite logs, metadata and usage records needed for a forensic examination.
Apple further alleges that Liu and others at OpenAI knew he retained access to Apple's third-party cloud storage. That claim directly contests OpenAI's earlier account, which described Liu's post-departure access as a consequence of Apple's offboarding failures and said former employees could retain access without realizing it.
OpenAI said on August 3rd that Apple's lawsuit was "careless, aggressive and oddly personal." OpenAI said Apple employees contacted Liu after his departure to request help locating files, and that Tan had repeatedly instructed staff not to use confidential information from previous employers.
OpenAI and the individual defendants moved to dismiss the case on August 5th, arguing that Apple had failed to identify protectable trade secrets with sufficient precision or plausibly show that OpenAI and io misappropriated them. OpenAI also said Apple had not established an ongoing injury.
Monday's filing is designed to answer that last point before the court reaches the merits. Apple now claims it has traced a specific schematic from post-employment access to work performed while Liu was employed by OpenAI, along with communications describing the simulation and the agent that helped run it.
The immediate dispute is procedural: whether Apple gets early access to devices, documents and testimony. The underlying fight reaches OpenAI's attempt to build consumer hardware with a group of engineers and designers drawn heavily from Apple. The forensic record from Liu's devices could determine whether that hiring campaign remains a talent dispute or becomes evidence that proprietary Apple engineering entered OpenAI's product-development process.