1 SEP 2026 — Apple has filed new evidence against a former engineer now at OpenAI, including text messages it says show he knew he still had access to Apple files. The filing is being read as a story about one employee. The relief Apple is asking for is an injunction against OpenAI's hardware development, which is a far larger ask than the conduct described supports on its face.
What the filing says
Apple sued in the Northern District of California on 10 July 2026. The new material, filed on 31 August, concerns Chang Liu, a former Apple employee who now works at OpenAI, and a colleague there named Yu-Ting Peng.
Apple alleges Liu used a confidential Apple circuit schematic in his work at OpenAI, and a tool sharing a name with an internal Apple engineering application. It alleges the two conspired to destroy evidence in June once Liu learned Apple was investigating him. It points to text messages in which Liu acknowledges continued access to Apple files, punctuated with crying-laughing emoji. Apple says he kept that access by exploiting a rare, previously unknown authentication bug.
OpenAI's position is that Liu accessed the files only after leaving Apple, to help former colleagues, and that the access existed because Apple failed to remove it when he departed.
The remedy is the story
A preliminary injunction blocking a competitor from developing hardware is an extraordinary request. It is granted before trial, on an incomplete record, and it requires the moving party to show a likelihood of success on the merits and irreparable harm that damages could not cure.
Trade-secret cases usually seek something narrower: an order that specific documents be returned and destroyed, that a named individual be walled off from a particular project, or that the defendant certify what was used and where. Those remedies map onto the alleged conduct of one engineer.
To stop an entire hardware programme, a plaintiff must argue that the secrets are so central the work cannot proceed without them. Apple is trying to make that case on a circuit schematic and an internally named tool, and the gap between the evidence and the claim is where the motion will be decided.
The defence has a real point buried in it
OpenAI's answer sounds like deflection, and it contains an uncomfortable fact: if a departed employee retained access to internal files, someone failed to remove it.
Apple's response is that Liu exploited an authentication bug to keep access, which shifts the conduct from passive retention to active circumvention. That distinction matters legally: unauthorised access carries its own exposure under computer-misuse statutes, quite apart from the trade-secret claims.
For other employers the practical difference is just as stark. Most organisations discover offboarding gaps the same way: an ex-employee mentions still being able to log in. Very few could prove afterwards whether the access was a provisioning failure or was actively obtained, because the logging needed to tell those apart is rarely retained for the months between a departure and a discovery.
Why 400 makes this fragile for both sides
More than 400 former Apple employees work at OpenAI. That number is the reason this case will not stay about one person, and it cuts against each side in a different way.
The number cuts against OpenAI because a pattern of conduct is easier for a court to enjoin broadly than a single incident. It cuts against Apple because a hiring flow that large makes it harder to argue that any particular piece of knowledge is a protectable secret rather than the ordinary experience an engineer carries between jobs. California does not enforce non-compete agreements, which is precisely why trade-secret litigation is the instrument that gets used there.
The line the courts draw is between general skill and specific confidential information. A schematic sits clearly on the protected side. A tool that merely shares a name with an internal application does not, unless Apple can show the implementation came with it.
What to take from this if you employ engineers
The operational lesson here is in the departure process, not the litigation. Access revocation should be verified, not merely triggered. A ticket that fires on an employee's last day proves nothing about whether the account is dead; the only real check is an authentication attempt after the fact.
Retain authentication logs long enough to answer a question raised months later. If Apple's account is right, the evidence that distinguishes circumvention from oversight is in logs from a period when nobody was looking.
And treat the alleged evidence destruction as the more serious half of the allegation. Courts have wide latitude to sanction spoliation, including instructing a jury to assume the destroyed material was unfavourable. Deleting messages after learning of an investigation lets a court infer the worst about their contents, which is why the alleged destruction in June may end up mattering more than the schematic. We reported on the vacatur of economic-espionage counts in the Linwei Ding prosecution, a reminder that the evidentiary bar in this area is high and that early characterisations often do not survive it.
