Apple has escalated its trade-secret lawsuit against OpenAI and former Apple engineer Chang Liu, saying an initial forensic review of Liu’s post-Apple MacBook uncovered evidence that confidential Apple engineering material was used in work at OpenAI.
The claims appear in Apple’s latest push for expedited discovery—a court process that would require earlier access to documents, devices and witness testimony. They are allegations in a filing, not findings by a court, and OpenAI has previously sought dismissal of the case.
What Apple says it found
Apple sued in July, alleging that former employees took trade secrets for OpenAI’s benefit. Liu, a former senior system electrical engineer, joined OpenAI in January. Apple alleges he exploited a security flaw after departing to download confidential engineering files.
According to the new filing, Liu’s lawyers recently produced a MacBook used after his Apple employment ended. Apple says its preliminary analysis indicates that:
- Liu downloaded a confidential Apple circuit schematic and used it in OpenAI work.
- Liu and others at OpenAI knew of his access to Apple’s third-party cloud storage.
- After learning of Apple’s internal investigation, Liu instructed an OpenAI colleague to destroy evidence; Apple says the colleague agreed.
- Liu used a tool at OpenAI with the same name as an internal Apple engineering application.
Apple says Liu ran a simulation using the schematic in LTspice, an electronics simulation tool, in March. It also points to messages in which Liu said an AI “agent” had learned to run LTspice and review its results.
A further technical wrinkle is central to Apple’s request: it says the schematic’s use became visible because activity on a Mac mini later synced through iCloud to the MacBook. Apple is now seeking access to that Mac mini as well.
Why the AI angle changes the dispute
The most consequential argument in Apple’s filing is that trade-secret data supplied to an AI agent or model may be difficult to contain once the system has learned from it. Apple argues that this could produce “irreversible and continually propagating uses” of the information.
That is a sharper claim than conventional concerns about an employee copying documents to a personal drive. If sensitive design information is placed into an AI workflow, investigators may need to establish not only where the original file traveled, but also whether it was retained in prompts, retrieval systems, agent logs, fine-tuning datasets, tool outputs or downstream artifacts.
For AI companies and teams adopting agentic tools, the case underscores a governance gap: a model or agent can become an additional route by which proprietary information is processed, transformed and potentially redistributed. Existing endpoint monitoring and access controls may not provide a complete record of those interactions.
The operator takeaway: controls must extend beyond offboarding
The allegations—if substantiated—highlight several practical priorities for companies handling sensitive technical IP:
1. Close access paths immediately. Offboarding should cover cloud-storage permissions, personal-device syncing and overlooked application access, not just corporate accounts. 2. Set explicit AI data boundaries. Define what proprietary materials can enter internal agents, external models and engineering tools, and require auditable logging where possible. 3. Preserve evidence quickly. Litigation holds need to include laptops, phones, cloud accounts, agent histories, collaboration platforms and connected machines. 4. Review cross-device synchronization. Consumer sync services can both create risk and provide forensic traces. Security teams should understand what is syncing, where it is retained and who controls it.
What to watch next
The near-term question is procedural: whether the court grants Apple’s request for expedited discovery and access to additional devices and records. The broader test will be whether the evidence supports Apple’s claims that its information entered OpenAI workflows and whether any AI system materially retained or used it.
Regardless of the case’s outcome, it is a warning that AI-related IP disputes will increasingly turn on forensic detail: device histories, cloud synchronization, agent logs and the chain of custody for data that may no longer exist as a single identifiable file.
