Apple has submitted new material in its trade secret lawsuit against OpenAI and former Apple engineer Chang Liu. Apple calls the evidence “shocking.” The public filing alleges that Liu used a confidential Apple power converter circuit while working at OpenAI, used AI agents to run related simulations, and discussed actions that could destroy forensic evidence after learning of Apple’s investigation.

These are allegations in an active civil case. Much of the underlying evidence is redacted or filed under seal, and the court has not decided that Apple’s claims are true. OpenAI disputes Apple’s account and says it does not want or use Apple trade secrets.

What Apple says it found

The latest filing followed the inspection of Liu’s former Apple work laptop, which his legal counsel provided during the dispute. Apple argues that the device connects confidential company information to work performed after Liu joined OpenAI.

According to Apple’s supplemental brief, the evidence raises three concerns:

  • Use: Apple alleges that a confidential circuit schematic was used in work performed while Liu was employed by OpenAI.
  • Propagation: Apple argues that feeding trade secret information into an AI agent or model may create uses that are difficult to reverse or trace.
  • Preservation: Apple says messages about restoring and using Apple owned devices show a risk that logs, metadata, or other forensic evidence could disappear.

Apple is using these claims to support a request for expedited discovery. That procedural point matters. The filing is an argument for faster access to evidence, not a final judgment on liability.

What OpenAI says in response

OpenAI has published its own account. It says Apple employees contacted Liu after his departure and asked him to help locate files and information for ongoing Apple work. OpenAI also argues that Apple failed to remove residual access properly and that its lawsuit mischaracterizes the communications.

OpenAI says it has no interest in competitors’ trade secrets and that its hardware leadership instructed teams not to use confidential information from other companies.

Two competing narratives can exist in the same record. The court’s job is to test them against evidence, access logs, devices, communications, and the law.

Why AI changes the risk model

Traditional trade secret cases often focus on a copied folder, an emailed document, or a physical prototype. AI introduces a harder question: what happens if confidential information is used to prompt, train, fine tune, evaluate, or guide an agent?

A company may need to determine:

  1. Which confidential input entered the system.
  2. Which model, agent, or tool processed it.
  3. Whether that information was stored in logs, memory, embeddings, or training data.
  4. Which outputs or engineering decisions were influenced by it.
  5. Whether the use can be isolated, deleted, or independently reproduced.

This is why provenance is becoming a core engineering control. Teams need to know where data came from and why a model was allowed to use it.

The operational lessons are immediate

Even before a court decides this case, technology companies can learn from it.

1. Offboarding must close every path

Returning a badge and disabling one account is not enough. Companies need a complete inventory of laptops, cloud sessions, shared drives, personal sync services, tokens, developer tools, messaging systems, and privileged credentials.

2. New employers need clean room onboarding

When someone joins from a competitor, the employer should document that confidential material is prohibited. Engineers should be given clean devices, approved data sources, and a clear escalation path if old files or access appear unexpectedly.

3. AI systems need data lineage

Agent logs, model inputs, tool calls, simulation files, and generated artifacts should be traceable. If a disputed source enters the workflow, the organization must be able to identify where it went.

4. Legal preservation must be technically enforceable

Once a dispute is anticipated, devices and logs should be preserved quickly. Retention policies, backups, immutable audit records, and access controls should support the legal hold rather than depend on individual memory.

A case about more than two companies

Apple and OpenAI are competing to define the next generation of consumer hardware. That makes the dispute commercially significant. It also makes careful language essential. Apple has presented serious allegations and supporting material. OpenAI has offered a conflicting explanation. The truth will depend on evidence that has not yet been fully tested in court.

For every other company, the message is already clear: in an AI workflow, confidential data can move faster and become harder to unwind. Access governance, provenance, and clean onboarding are now product and engineering responsibilities, not paperwork for later.

Sources