Apple vs. OpenAI: When an AI Agent Becomes a Vehicle for Stolen Secrets

The Real Victim in Apple v. OpenAI May Be the Concept of Trade Secrets

Apple has submitted new forensic evidence in its lawsuit against OpenAI, and it paints a picture of intellectual property theft that would have been unthinkable just five years ago. The case centers on Chang Liu, a former Apple engineer who apparently downloaded a confidential circuit schematic after joining OpenAI, then did something far more troubling than merely store it: he trained an AI agent to use it.

This is not a simple case of someone copying files to a personal drive. Liu allegedly used the stolen schematic to run electrical engineering simulations through LTspice, messaged colleagues about the work, and described training an AI agent to autonomously execute the same simulations. He wrote in one message, “In the past hour, my AI agent learned how to run LTspice, look at result, tune compensation parameter.” Another colleague reportedly replied: “Oh man! Why do they even need you then?”

The question that will now occupy courts, boards, and compliance officers is whether an AI agent that has absorbed proprietary knowledge can itself be considered a vessel for trade secret theft — and whether OpenAI can be held liable for what its engineers teach machines.

How the Evidence Unfolded

Apple filed its theft of trade secrets lawsuit in July against Liu and Tang Tan, who served as OpenAI’s chief hardware officer and previously spent 24 years at Apple as VP of product design for iPhone and Apple Watch. The allegations are specific and escalate quickly.

Liu downloaded the confidential circuit schematic in March, two months after leaving Apple. He ran it through LTspice, an electrical engineering simulation tool. He then described training an AI agent to perform the same work autonomously. When Apple launched an internal investigation, Liu sent instructions for destroying evidence to an OpenAI colleague who confirmed she would comply. Liu also used a tool at OpenAI bearing the same name as an internal Apple engineering application, according to the filing.

The laptop Liu took from Apple was eventually recovered. Forensic analysis revealed the schematic had been synced from a Mac mini to the MacBook through Apple’s cloud infrastructure. Apple is now seeking to examine the Mac mini and other devices Liu still possesses, arguing that expedited discovery is essential because evidence is actively being destroyed and trade secrets remain in use.

OpenAI’s Response: Denial and Deflection

OpenAI has moved aggressively to counter the allegations. The company filed a motion asking the court to dismiss the lawsuit entirely, calling the accusations “meritless.” In a public statement released hours after Apple’s initial filing, OpenAI characterized the lawsuit as “careless, aggressive and oddly personal” and stated it does “have, nor want, any of their trade secrets,” adding that it is “much more interested in building innovative products and technologies that push the frontier.”

Lawyers for OpenAI have argued that Apple’s suit mischaracterizes the actions of its employees. The company has not disputed that Liu accessed Apple’s cloud storage after departing, but maintains that any such access was unauthorized by OpenAI and does not constitute corporate liability.

Judge Edward J. Davila will hear arguments on both the expedited discovery request and the motion to dismiss on October 1. The outcome of that hearing could set early precedent for how federal courts treat AI-related trade secret cases.

What This Case Actually Tests

The Liu case arrives at a moment when the legal framework for trade secrets is straining under the weight of generative AI. Traditional IP law was built around the idea that stolen documents live in stolen folders. It was never designed to handle a scenario where proprietary knowledge is fed into a model and then retrieved by an autonomous agent that no single person fully controls.

If Apple wins expedited discovery and produces evidence that the AI agent is still executing LTspice simulations based on the stolen schematic, the legal questions multiply. Does OpenAI bear responsibility for instructing an agent to operate on proprietary data? Is the agent itself a “contraband” item that should be sequestered? Can a company be enjoined from using a model that was partially trained on stolen inputs?

These are not hypothetical questions. They will shape how every AI company handles post-employment data risks for the next decade.

Who Wins, Who Loses, Who Changes

If Apple prevails, the ruling could force OpenAI and its competitors to implement far stricter data isolation protocols. Engineering teams working on hardware-aware AI models would need to prove that no proprietary schematics, simulation parameters, or internal tooling was ingested during training. The compliance burden on AI labs would increase dramatically, and the cost of hiring engineers from hardware companies like Apple would rise accordingly, as firms would need to conduct deeper forensic audits of candidate data histories.

If OpenAI wins — or if the case is dismissed — the precedent would effectively signal that trade secret law has not kept pace with AI capability. Companies would face weaker recourse when former employees carry institutional knowledge into models that can replicate, refine, and autonomously execute on that knowledge. The incentive structure would shift: the marginal cost of hiring a well-placed engineer would look increasingly attractive compared to the marginal cost of building equivalent capability from scratch.

There is also a human dimension that the legal filings do not fully capture. Liu’s messages reveal someone genuinely excited about what the agent could do. The phrase “Everything ideal” suggests a breakthrough, not malice. That does not absolve the alleged theft, but it does illustrate why this case matters beyond corporate grievance. The people building these systems are often the ones who understand the value of the data best — and the ones most positioned to move it.

The Bigger Picture

Apple’s lawsuit is about one engineer, one schematic, and one agent. But the implications reach far beyond Cupertino and San Francisco. Every AI lab, every hardware company, every compliance department is watching to see whether the courts treat AI agents as tools or as liabilities. The hearing on October 1 will not resolve the case, but it will reveal how seriously the judiciary intends to treat this new category of intellectual property threat.

What happened in this case would have been a straightforward theft of documents ten years ago. The novel element is that the stolen knowledge was absorbed, automated, and set to work — possibly without any human directly touching the schematic anymore. The law has not yet caught up to that reality. This case may be the first test of whether it will.