Apple's 'Shocking Evidence': Ex-Engineer Trained an AI Agent on Stolen Circuit Schematics

The Apple v. OpenAI trade secrets fight just escalated again. On August 31, Apple filed what its own lawyers call “shocking evidence” in the lawsuit it brought against OpenAI in July — and this time the centerpiece isn’t a text message or a download log. It’s a forensic image of a MacBook, a confidential circuit schematic, and an AI agent that Apple says a former engineer trained to exploit both.

The filing lands three weeks before Judge Edward J. Davila is scheduled to hear arguments on October 1, and Apple is using it to push for two things: a preliminary injunction blocking OpenAI from building hardware based on Apple technology while the case proceeds, and expedited discovery — a fast-tracked evidence process — on the grounds that its trade secrets are “in active use” and evidence is actively being destroyed.

What the forensic analysis actually found

The backstory: Chang Liu, a senior systems electrical engineer who spent eight years at Apple working on iPhones, left for OpenAI’s nascent hardware division. Apple sued Liu and fellow ex-Apple engineer Tang Tan on July 10, alleging they used multiple methods — including a retained company laptop and what Apple describes as a “rare, previously unknown authentication bug” — to keep accessing Apple’s confidential systems after departure.

After the lawsuit was filed, Liu’s legal counsel finally handed his old Apple-issued MacBook back, and earlier this month Apple’s forensics team got to work. What they claim to have found goes well beyond residual access.

According to the filing, Liu downloaded a confidential Apple circuit schematic in March — two months after he stopped working at Apple. He then used that schematic to run a simulation in LTspice, a widely used electrical engineering tool, in connection with power conversion development work. And here the story takes a distinctly 2026 turn: Liu didn’t just run the simulation himself. He described, in messages Apple now has, training an AI agent to run LTspice for him.

The filing quotes him telling contacts he was “[f]eeling AI all day long,” and that “[i]n the past hour” his AI “agent learned how to run LTspice, look at result, tune compensation parameter.” He reportedly wrote: “I try to make a behavior buck. With voltage outer and current inner. Everything ideal. It still took me a day.. now it’s two hour. Including learning fresh.” The person he was messaging with replied, “Oh man! Why do they even need you then?”

That exchange — an engineer marveling that an AI agent compressed a day of simulation work into two hours, allegedly on top of a schematic that belonged to his former employer — is precisely the artifact Apple wants in front of a judge.

There’s also a forensics subplot: Apple says Liu ran the schematic on a Mac mini that later synced to the MacBook he had taken from Apple. That sync is how Apple knows the schematic was used at all. The company now wants to examine the Mac mini and other devices still in Liu’s possession.

The destruction-of-evidence allegation

The second prong is arguably more serious. Apple alleges that when Liu learned in June that he was under internal investigation, he enlisted an OpenAI colleague — named in the filing as Yu-Ting Peng — to help destroy evidence. MacRumors reports that the forensic record shows Liu sent instructions for destroying evidence, and that the colleague confirmed receiving them.

Apple’s filing frames the laptop as the thin end of the wedge: “The MacBook represents the very limited information Defendants provided so far (and only after weeks of delay), and shows Apple is not conducting ‘fishing expeditions’ but that its trade secrets are being used and evidence is being destroyed.” The company also alleges OpenAI was “well-aware” of Liu’s continued access to Apple data, and that more former employees may be implicated — a meaningful threat given that, per Apple’s initial complaint, more than 400 ex-Apple employees now work at OpenAI.

Much of the new evidence is redacted from public view, but earlier filings already included Liu’s text messages — punctuated, memorably, with crying-laughing emojis — indicating he knew he still had access to Apple files.

OpenAI’s counter-narrative

OpenAI has not conceded an inch. In early August it moved to dismiss the suit as “meritless” and published a defiant blog post titled “Apple is getting this wrong,” calling the lawsuit “careless, aggressive and oddly personal” and insisting OpenAI does not “have, nor want, any of their trade secrets” because it is “much more interested in building innovative products and technologies that push the frontier.”

On the substance, OpenAI argues Liu only accessed Apple files after leaving in order to help former colleagues who asked for his assistance, and that the “residual access” Apple complains about is a routine offboarding failure — “a common issue with Apple which is caused by them failing to properly manage system access when people leave.” In other words: if your departing engineer can still log in months later, that’s an IT governance problem, not a heist.

That argument now has to contend with the schematic simulation and the destruction instructions. Residual access explains a login; it doesn’t obviously explain downloading a confidential schematic two months after departure, running power-conversion simulations against it, or coordinating with a colleague to erase the traces once an investigation surfaced.

Why this matters beyond the courtroom

Three stakes stand out.

First, the AI-agent twist is genuinely novel. Trade secret law was built for humans carrying documents out the door. Liu’s alleged behavior — directing an AI agent to operationalize a stolen schematic, compressing “a day” of engineering iteration into “two hours” — represents a new category of harm: not just exfiltration of trade secrets, but acceleration of work built on them. Courts have not really wrestled with what it means for an AI agent to be the instrument of a trade secret’s use, and how any eventual damages or injunction framework accounts for that multiplier effect. Whatever Judge Davila does on October 1, this filing will get cited.

Second, the case is becoming a referendum on offboarding. Apple says a rare authentication bug was exploited; OpenAI says Apple’s access management is systematically sloppy. Every enterprise watching this case is implicitly asking the same question: could our departed engineers still get in? If Apple — a company legendary about security — couldn’t fully cut off a departed hardware engineer, the residual-access problem is likely endemic across the industry, and AI companies aggressively hiring hardware talent are where it bites hardest.

Third, there’s real business risk on OpenAI’s side. A preliminary injunction that stalls hardware work built on Apple-derived technology would land directly on OpenAI’s consumer device program — the Jony Ive collaboration it has staked considerable credibility on. Expedited discovery would also force internal OpenAI communications into the open far faster than the company would like, months before the IPO window it has been widely reported to be targeting.

For now, the ball is in Judge Davila’s court, literally. The October 1 hearing will determine whether OpenAI’s hardware program has to operate under a court-ordered constraint while the case grinds on, and whether Apple gets the fast-track discovery it says the destruction of evidence demands. One thing is already clear: in the first major trade secrets war of the AI era, the evidence is no longer just documents — it’s forensic images, sync logs, and the unmistakable trail of an AI agent doing in two hours what used to take an engineer a day.