Filed August 31, 2026 in Apple Inc. v. Liu, No. 5:26-cv-07078-EJD (N.D. Cal., San Jose Division, Judge Edward J. Davila), as Exhibit A to Document 94, with a hearing noticed for October 1, 2026 at 9:00 a.m. It supplements Apple's earlier motion for expedited discovery (Dkt. 49) and reply (Dkt. 90) in the case tracked at Apple v. OpenAI (trade secrets). Counsel of record are Weil, Gotshal & Manges LLP and Desmarais LLP for Apple.
What is and is not public
The document is captioned "REDACTED VERSION OF DOCUMENT FILED UNDER SEAL." Of its nine pages, pages 3 through 6 contain no public text at all: the statement of the new evidence, and Sections I, II and III.A–B, are withheld in full. Only the caption and counsel block, Section III.C ("The new evidence confirms the urgent need for expedited discovery"), Section IV ("Conclusion"), and the signature blocks are on the public record. Contemporaneous reporting characterized the filing as containing "shocking evidence" against Chang Liu (Source: techcrunch.com); that evidence itself is not readable in the public version, and the account below is confined to what survives redaction.
The argument that survives redaction
Apple's readable argument makes two claims for urgency.
The first concerns AI-mediated propagation. The brief states that Liu's "use of Apple's trade secret power-converter circuit while employed by OpenAI and his use of AI agents to learn to run simulations raise concerns extending beyond ordinary document theft," because where trade-secret information "is fed into an AI agent or model that 'learn[s]' from it, such 'learning' may create irreversible and continually propagating uses of the trade secret" — harm the brief describes as "uniquely challenging to undo" and requiring prompt investigation. The proposition is supported by two expert declarations cited as Roffman Decl. ¶ 31 and Fayed Decl. ¶ 26. This is a trade-secret remedy argument built specifically on model training and agent use rather than on document copying, and it bears on the questions at AI Copyright and Agentic AI about what a model retains from what it is trained or run on.
The second concerns spoliation risk. The brief states that forensic artifacts "such as logs, metadata, and usage records — are transient and at risk of being lost, overwritten, or destroyed," and that a MacBook in the possession of Liu's counsel shows that after Liu and Yu-Ting Peng learned of Apple's investigation in June, "they discussed the need to 'restore' and then 'start using' Apple-owned devices in their possession." Apple characterizes this as a discussion of "destroying the types of forensic data Apple needs." It notes that Liu's counsel held the MacBook but had not yet examined its data and "seemingly have not looked at the Mac Mini," a second device the brief identifies as still uninvestigated. Apple cites AssuredPartners, 2026 WL 1182808, at *6, and WeRide, 379 F. Supp. 3d at 854, for the proposition that courts find good cause for expedited discovery where evidence is at risk of destruction.
The conclusion states that the new evidence "confirms Mr. Liu used the trade secrets he misappropriated from Apple while employed by OpenAI," that "additional pertinent devices remain uninvestigated," and that "the full scope of Defendants' misappropriation is yet to be exposed."
What the filing does not establish
Because the evidentiary sections are sealed, the public document does not show what the new evidence is, how it was obtained, or what it depicts. It does not identify the internally-named engineering tool that press accounts attribute to the filing, and it does not set out the evidence-destruction allegation involving Peng beyond the quoted "restore" and "start using" discussion. It contains no response from OpenAI or from the individual defendants; OpenAI's answering position was filed separately and is recorded at Apple v. OpenAI (trade secrets).
Relationships
- supports: Apple v. OpenAI (trade secrets) — a filing in the case's procedural history
- related: Apple Inc. v. Liu, Tan, OpenAI and io Products — Complaint (July 2026) — the July 10, 2026 complaint in the same action, not superseded by this filing
- related: Agentic AI — the brief's irreversible-propagation theory turns on agent and model learning
- related: Apple, OpenAI