Apple has gained faster access to selected OpenAI recruitment records in its hardware trade-secrets lawsuit. The October 13, 2026 order directs production of interview materials, certain device evidence and private messages. It does not decide that OpenAI stole information, that Apple proved a valid trade secret or that anyone destroyed evidence.
The distinction is operationally important. Apple can investigate whether hiring crossed into collection of protected knowledge; OpenAI avoided an accelerated inspection of broad hardware research. The order advances evidence gathering while leaving the parties' central stories contested.
What the court ordered—and what it withheld
US District Judge Edward Davila issued the order in the Northern District of California. Apple filed suit on July 10 against OpenAI, io Products, Chang Liu and Tang Yew Tan. The ruling partially grants expedited discovery, meaning selected evidence can be obtained before the ordinary schedule.
Defendants have five days to produce responsive materials already collected for the litigation and 14 days for other covered materials. Production includes existing forensic images of relevant devices: detailed storage copies that preserve files, metadata and activity evidence.
The interview requests cover guides, checklists, instructions, manuals and scripts used for Apple candidates to OpenAI's consumer hardware and robotics teams. Technical presentations summarizing interviews and interview records referencing Apple technology or physical components are also included. Those documents could test what recruiters formally requested, recorded and circulated.
The order also reaches LINE messages between Liu and Alyssa Peng from January 1, 2026, and permits depositions of Liu, Peng and Tan within Apple's narrowed requests. All previously worked at Apple. Tan spent 24 years there, became a product-design vice president and is now OpenAI's chief hardware officer; Liu spent eight years as a senior systems electrical engineer before joining OpenAI in January 2026.
Davila refused expedited access to research on metal finishes, power management and battery functions, leaving those topics to ordinary discovery. Other broad requests concerning hardware work and former employees were rejected. Apple's proposed definition of Apple-originated information could capture third-party material or ordinary employment knowledge and impose excessive burdens.
The initial report therefore describes a limited order, not an unrestricted victory. Apple gained access tied to people, interviews, devices and communications, not a general audit of a rival's future products.
Apple's recruiting and file-access allegations remain allegations
Apple says Tan used confidential project names, sought details of unreleased products and invited candidates to bring components to interviews. If documented, standardized instructions would raise a different institutional question from an isolated employee retaining a file. The court has not found that an interview request involved a protected trade secret.
Apple also alleges Liu failed to return a company laptop and accessed and downloaded confidential hardware material after joining OpenAI. It says an authentication flaw allowed continued access, later fixed, and describes dozens of engineering presentations, specifications and unreleased-product files. Apple further alleges he advised another employee on material to study before an OpenAI interview.
In later filings Apple said forensic review showed Liu using a confidential circuit schematic at OpenAI. OpenAI contested that interpretation, with its experts reportedly finding no evidence the file was used in connection with an AI platform on a separate Mac mini. Davila did not resolve that forensic disagreement.
Transient logs and metadata can disappear as devices operate. The judge found this preservation concern persuasive alongside messages about restoring Apple-owned devices. That supports earlier access without establishing intentional deletion: the court did not find Liu or Peng destroyed evidence and declined to weigh competing forensic accounts.
As the legal analysis of the complaint notes, Apple invokes the federal Defend Trade Secrets Act and brings contract claims against individuals. Its case requires more than a confidential label: protectable information, reasonable secrecy measures and unlawful acquisition, disclosure or use remain disputed issues.
OpenAI challenges both the alleged secrets and Apple's controls
OpenAI portrays the case as a response to talent losses and AI competition. Apple alleges more than 400 former employees work at OpenAI; that count does not imply wrongdoing by those workers. General skills and experience are different from protected files, components and confidential project details.
In its published response, OpenAI argues Apple blurred personal and corporate information through personal iCloud accounts and failed to make continuing access clear. It presents selected communications challenging Apple's timeline, including February contacts and a lawyer's misdirected email. These are advocacy materials, not independent findings.
An incorrectly addressed email does not decide either side's case. Neither would weak offboarding automatically grant permission to retrieve confidential information. Technical access and lawful authorization are different questions, even while access design may bear on Apple's claimed protective measures.
The motion-to-dismiss account describes OpenAI's argument that Apple has not adequately identified protectable secrets. Information might be general, already known or independently developed. Davila's refusal of an expansive Apple-information definition reflects the need to connect discovery to specific claims, rather than all knowledge learned at a former employer.
Partnership and hardware competition create the wider stakes
Apple and OpenAI became prominent partners in 2024 through ChatGPT access on Apple devices. Their relationship now combines cooperation, recruiting competition and litigation as OpenAI builds a consumer hardware organization. The earlier case report framed the suit as a rupture between technology partners.
Hardware recruitment brings expertise in manufacturing, batteries, materials, radios, supply chains, heat and interaction design. That expertise can move with a worker; alleged protected documents and unreleased details create a separate question. The order's significance will depend on actual interview practices and the handling of disputed material, not the mere presence of former Apple staff.
A file on a device does not alone show product influence. Access times, applications, copies, messages, simulations and design records need context. Redactions also limit what outsiders can infer. Apple may seek ordinary discovery or narrower renewed requests, while OpenAI must expose selected recruitment practices without immediately opening its broader research.
The next meaningful evidence is documents, testimony and injunction review
Interview materials could show general engineering questions and confidentiality warnings, or Apple-specific prompts, component requests and project references. Depositions can connect those records with Liu's device access, Peng's messages and Tan's recruiting instructions. Consistent contemporaneous evidence matters more than either party's categorical description.
Apple also seeks a preliminary injunction to restrict alleged misuse while litigation continues. That would carry different operational consequences from an evidence order. Likelihood of success, potential irreparable harm, competing hardships and public interest form part of that review; this order alone does not grant the requested restrictions.
Sensitive production may remain sealed, redacted or subject to protection, so later filings may provide only partial descriptions. OpenAI's first device form and schedule also remain insufficiently established here; hiring and litigation do not prove a viable product or justify speculation about one.
The present result is narrow but consequential: Apple can test specific recruiting and handling allegations sooner, and OpenAI keeps meaningful limits around accelerated access to research. What the records say—not simply that disclosure was ordered—will determine whether this remains a hiring-and-individual-conduct dispute or reaches the hardware development process.