OpenAI Hiring Records Ordered Disclosed as Apple Trade Secrets Fight Widens

OpenAI must disclose hiring records within 14 days after a federal judge gave Apple limited early access to evidence in its trade secrets lawsuit. The order moves the dispute beyond competing allegations and into OpenAI’s interview files, private messages, forensic images, and sworn testimony.

The ruling does not establish that OpenAI stole anything. Instead, it gives Apple a faster route to evidence that could support or weaken its claims. The approved OpenAI hiring records cover interviews involving former Apple employees who applied to OpenAI’s consumer hardware and robotics groups.

That distinction matters. Apple argues that OpenAI treated recruiting as a channel for obtaining confidential hardware knowledge. OpenAI says Apple is using a flawed lawsuit to compensate for its problems retaining employees and competing in artificial intelligence.

The court accepted neither side’s full position. It authorized targeted discovery tied to Apple’s allegations, while rejecting broader access to OpenAI’s hardware research. The result puts pressure on OpenAI without giving Apple an unrestricted view inside a future competitor.

The OpenAI Hiring Records Order Is Narrow but Significant

The court gave Apple access to the recruiting process at the center of its allegations, but it did not hand Apple OpenAI’s entire hardware operation.

U.S. District Judge Edward Davila issued the order on October 13, 2026, in the Northern District of California. Apple filed the case on July 10 against OpenAI, io Products, Chang Liu, and Tang Yew Tan.

The court order grants part of Apple’s motion for expedited discovery. Expedited discovery allows evidence gathering to proceed faster than the ordinary case schedule.

OpenAI and the other defendants have five days to produce responsive materials that they already collected for the litigation. They have 14 days to respond with other materials covered by the order.

The production includes forensic images already collected from devices connected with Apple’s claims. A forensic image is a detailed copy of a device’s storage used to preserve files, metadata, and activity records.

The defendants must also produce documents used to guide interviews of Apple candidates for OpenAI’s consumer hardware and robotics teams. Those materials can include checklists, guides, instructions, manuals, and interview scripts.

The order reaches technical presentations that summarized those interviews. It also covers interview records that referenced Apple technology or physical components.

This is the most consequential part of the ruling. Apple has alleged that OpenAI’s recruiting process encouraged candidates to discuss confidential work, study internal materials, or bring Apple components into interviews.

The approved discovery can test whether those allegations describe documented interview practices or isolated conversations. Interview templates and presentation records may show what OpenAI formally requested, recorded, or circulated.

Judge Davila also ordered production of LINE messages between Liu and Alyssa Peng dating from January 1, 2026. LINE is a private messaging service used for direct and group conversations.

Apple can depose Liu, Peng, and Tan under the limits specified in its narrowed requests. Depositions are sworn interviews conducted before trial, with testimony recorded for later use.

All three worked at Apple before joining OpenAI. Tan spent 24 years at Apple and eventually became a vice president of product design. He now serves as OpenAI’s chief hardware officer.

Liu worked at Apple for eight years as a senior systems electrical engineer before joining OpenAI in January 2026. Apple says he retained access to sensitive files after leaving the company.

However, the court denied Apple’s request for expedited access to documents covering OpenAI’s research on metal finishes, power management, and battery functions. Those topics will remain available through ordinary discovery.

The judge also rejected several broad requests concerning OpenAI’s hardware research and work created by former Apple employees. He found that parts of Apple’s proposed definition for Apple-originated information were excessively broad.

That definition could have captured information involving third parties or ordinary knowledge acquired through employment. The court said compliance with the resulting requests could become extremely burdensome.

The ruling therefore separates two questions. Apple can investigate whether OpenAI’s recruiting and specific employees handled its information improperly. It cannot use expedited discovery to inspect every related part of OpenAI’s hardware program.

The initial coverage accurately described this as limited access. Calling the decision a complete victory for Apple would miss the boundaries that Judge Davila imposed.

Those boundaries also protect the case from becoming a general audit of OpenAI. For now, discovery must remain tied to specific allegations, people, devices, interviews, and communications.

Apple Says Recruiting Became a Route to Confidential Hardware Knowledge

Apple’s case depends on proving that ordinary hiring crossed into organized collection of protected information.

Apple’s complaint presents recruitment as part of the alleged misconduct, not merely the context surrounding employee departures. The company says OpenAI sought information about unreleased products while assembling its consumer hardware organization.

According to Apple, Tan used confidential project names when speaking with candidates from his former employer. Apple also alleges that he requested details about unannounced products and invited candidates to bring hardware components to interviews.

Those remain allegations. OpenAI disputes Apple’s characterization, and the court has not ruled that any interview request involved a protected trade secret.

The OpenAI hiring records may nevertheless provide unusually direct evidence. Recruiting documents could reveal whether technical interviews were designed around general engineering experience or around Apple-specific products and processes.

That difference is central to trade secrets law. Employees can carry general skills, experience, and professional judgment to a new employer. They cannot lawfully transfer protected technical information belonging to their former company.

Apple also accuses Liu of failing to return an Apple laptop when he left. It says he accessed and downloaded confidential hardware files after starting work at OpenAI.

The complaint claims Liu exploited an authentication flaw that allowed him to reach Apple systems after his employment ended. Apple says it later fixed the flaw and terminated his access.

Apple alleges that Liu obtained dozens of confidential files containing engineering presentations, technical specifications, and data about unreleased products. It further claims that he advised another Apple employee about materials to study before an OpenAI interview.

Later filings escalated those allegations. Apple said a forensic review showed Liu using a confidential circuit schematic while working at OpenAI.

OpenAI has contested that interpretation. Its experts reportedly found no evidence that Liu used the disputed file in connection with an AI platform on a separate Mac mini.

The judge declined to resolve that forensic disagreement at this stage. A discovery order determines what evidence the parties can obtain, not who has proved the underlying claim.

Apple also argued that temporary digital evidence could disappear before normal discovery progressed. Logs, metadata, and usage records can be overwritten as devices and systems continue operating.

Judge Davila found that concern persuasive when combined with messages discussing the restoration of Apple-owned devices. The order says early discovery is justified partly by the risk that evidence could be lost.

The court did not find that Liu or Peng destroyed evidence. It explicitly avoided weighing the parties’ competing forensic accounts.

This careful distinction is important. A risk of lost evidence can support faster production without establishing that anyone intentionally deleted or concealed records.

Apple’s broader concern is that disputed information could become embedded in OpenAI’s hardware work. Untangling independently developed engineering from misappropriated material becomes harder as designs change and teams collaborate.

Apple is seeking a preliminary injunction, which would impose restrictions before a final trial. The company argues that early discovery can help define any restrictions and prevent alleged harm from becoming irreversible.

An independent legal analysis notes that the complaint invokes the federal Defend Trade Secrets Act. Apple also brings contract claims against the individual defendants.

To prevail, Apple must do more than identify files marked confidential. It must establish protectable trade secrets, reasonable efforts to preserve their secrecy, and unlawful acquisition, disclosure, or use.

OpenAI can challenge each element. It can argue that disputed information was too general, already known, independently developed, or poorly protected by Apple’s own systems.

The interview evidence matters because it connects individual conduct with corporate practice. A retained file on one employee’s device presents one kind of case. Standardized recruiting instructions present a much broader institutional question.

Apple and OpenAI Are Fighting Over Talent as Much as Files

The primary conflict is whether OpenAI competed fairly for Apple engineers or converted hiring into access to a rival’s internal knowledge.

More than 400 former Apple employees work at OpenAI, according to Apple’s allegations. That figure does not indicate wrongdoing by those workers, but it explains the scale of Apple’s concern.

Technology companies routinely recruit from competitors. California also protects employee mobility more strongly than many jurisdictions, making an employer’s attempt to control former workers especially sensitive.

Apple cannot treat every departure as theft. OpenAI cannot assume that rapid hiring from one company eliminates the need for strict information controls.

The dispute therefore tests the boundary between competitive recruitment and trade secret misappropriation. That boundary often depends on conduct, documentation, and the specificity of the information involved.

OpenAI’s response portrays Apple as a company struggling to retain talent while falling behind in AI. Its motion to dismiss describes the lawsuit as pretextual and attacks Apple’s handling of employee access.

The company argues that Apple mixed personal and corporate information by encouraging staff to use personal iCloud accounts for work. OpenAI also says former employees were not necessarily aware that they retained access after leaving.

In its public defense, OpenAI says Apple created confusion through its own systems and policies. Its published response includes selected communications intended to challenge Apple’s timeline and account of the dispute.

Those materials do not independently settle the factual questions. OpenAI selected them to support its defense, just as Apple selected evidence for its complaint and injunction request.

OpenAI also says Apple’s lawyers contacted the company in February but did not clearly present the later lawsuit’s allegations. It highlights an email that Apple’s outside counsel mistakenly sent to the wrong recipient.

The email mistake is embarrassing but not dispositive. It neither proves that OpenAI sought trade secrets nor establishes that Apple’s claims lack merit.

The more substantive defense concerns security controls. If Apple failed to remove credentials or separate personal accounts from company systems, OpenAI can use that record to question Apple’s protective measures.

However, weak offboarding controls would not automatically authorize former employees to retrieve confidential data. System access and legal permission are different questions.

The motion to dismiss also argues that Apple has not adequately identified a protectable trade secret. That is a common and important challenge in early trade secrets litigation.

Companies sometimes describe secrets at a level so broad that defendants cannot understand what they supposedly took. Courts often require greater specificity before allowing expansive discovery.

Judge Davila’s order reflects that concern. He accepted targeted production while rejecting Apple’s broad definition of information created for or learned through Apple work.

This partial result pressures both sides. OpenAI must expose parts of a recruiting process that it would ordinarily keep private. Apple must build its case without immediate access to broad areas of OpenAI research.

The rivalry also contains an unusual commercial tension. Apple and OpenAI became prominent partners in 2024 when Apple integrated ChatGPT access into its devices.

Their relationship now combines cooperation, competition, and litigation. OpenAI is still associated with software services on Apple platforms while developing hardware that might reduce dependence on smartphones.

That shift raises the stakes around technical talent. Hardware expertise involves manufacturing limits, battery systems, materials, radio performance, supply chains, thermal design, and human interaction.

A company entering consumer hardware can recruit people who already understand those constraints. The legal problem begins when expertise becomes inseparable from protected documents, components, or confidential project details.

For engineers and hiring managers, the lesson is practical. A candidate’s ability to explain prior work does not give a prospective employer permission to solicit another company’s unreleased designs.

Interviewers need documented boundaries. Candidates need clear instructions about what they must not disclose. Employers also need procedures for handling suspicious documents or physical components that appear during recruitment.

That kind of evidence trail can become decisive. A clear warning supports an employer’s claim that it tried to prevent misuse. A request for proprietary details can support the opposite inference.

The Court Did Not Decide Whether OpenAI Stole Trade Secrets

The ruling gives Apple evidence, not a judgment on liability, and its most serious accusations remain contested.

Early procedural victories often acquire more meaning in public coverage than they carry in court. This order does not hold OpenAI, Liu, Tan, Peng, or io liable.

It does not confirm that Apple identified a valid trade secret. It does not decide whether OpenAI used Apple information in a product.

It also does not establish evidence destruction. The judge cited the risk that transient records might disappear, while declining to decide the parties’ competing claims about deletion.

OpenAI therefore retains several defenses. It can argue that interview documents show conventional technical evaluation rather than requests for Apple secrets.

It can also contend that employees discussed general engineering knowledge, public information, or personal experience. Those categories usually remain available to workers after changing jobs.

Forensic evidence will need context. A file’s presence on a device does not necessarily show that it influenced a commercial product.

Investigators may examine access times, application histories, copies, communications, simulations, and design records. Each artifact can support several explanations unless connected to other evidence.

The court’s rejection of broader expedited discovery also limits Apple’s ability to trace alleged information directly into OpenAI’s research. That restriction may make immediate product-level conclusions harder.

Apple can revisit some requests through normal discovery. The judge also left room for narrower proposals if Apple can tie them more closely to its claims.

This creates a tradeoff for Apple. Narrow requests can move faster and face less resistance, but they may reveal less about how OpenAI developed its hardware.

OpenAI faces the inverse tradeoff. It avoided immediate exposure of broad research categories, but its interview process and selected communications now receive closer scrutiny.

The public record also lacks crucial context. Many technical details remain redacted because they allegedly concern confidential information.

Readers should be skeptical of categorical narratives from either party. Apple describes an organized pattern of theft. OpenAI describes a baseless response to competition for talent.

The available ruling supports a more limited conclusion. Apple identified enough specific concerns to justify accelerated access to selected evidence.

The judge also recognized that Apple’s requests could become a burdensome search through a competitor’s files. That concern prevented the company from obtaining everything it requested.

The order’s balanced structure weakens simplistic claims that one side has already won. Apple gained investigative leverage, while OpenAI preserved meaningful limits around its research.

An earlier case account framed the lawsuit as a major rupture between technology partners. The discovery ruling shows that the rupture now reaches daily operating practices.

Recruiting teams may need to preserve interview notes, scoring records, presentations, and candidate communications. Engineering teams may face questions about when they received particular ideas or files.

Executives could also face scrutiny over reporting lines and escalation procedures. A company’s response after learning about disputed information can matter almost as much as the initial incident.

For other employers, the case exposes risks on both sides of employee mobility. Departing workers can retain access accidentally, intentionally, or through poorly designed systems.

The former employer needs immediate credential revocation, device recovery, and clear separation between personal and corporate storage. The new employer needs safeguards against receiving protected material.

Neither process works through policy language alone. Courts examine messages, device records, actual interview practices, and the actions taken when concerns surfaced.

The OpenAI hiring records may provide that operational evidence. Their significance will depend on what they contain, not simply on the fact that Apple obtained them.

What Comes Next in the Apple and OpenAI Case

Three developments will determine whether this discovery order becomes a turning point or remains a limited procedural step.

The first signal is the content of the interview materials. They should clarify whether OpenAI used standardized safeguards or encouraged discussion tied to Apple technology.

Records showing general engineering questions and confidentiality warnings would strengthen OpenAI’s defense. Apple-specific prompts, component requests, or confidential project references would strengthen Apple’s account.

The production deadlines create the earliest test. Materials already collected must be produced within five days of the October 13 order, while other covered records are due within 14 days.

Public access will probably remain limited. Sensitive documents may be sealed, redacted, or exchanged under a protective order.

Even without public documents, later filings may reveal how each side characterizes the evidence. Readers should watch for specific descriptions rather than broad accusations.

The second signal is deposition testimony from Liu, Peng, and Tan. Their answers can connect documents and messages with decisions, instructions, and technical work.

Liu’s testimony should address device access, retained equipment, disputed files, communications with Apple employees, and any use of those materials at OpenAI.

Peng’s testimony can clarify the LINE messages and discussions about Apple-owned devices. Tan’s testimony can address how OpenAI interviewed Apple candidates and managed confidentiality.

Inconsistencies between testimony and digital records would increase pressure on the relevant witness. Consistent accounts supported by contemporaneous documents would weaken opposing allegations.

The third signal is Apple’s preliminary injunction request and any renewed discovery proposal. An injunction would matter more operationally than the current evidence order.

Apple wants restrictions that would prevent alleged trade secrets from influencing OpenAI hardware while litigation continues. To obtain that relief, it must establish more than suspicion.

The court will consider factors that include Apple’s likelihood of success, potential irreparable harm, competing hardships, and the public interest. The precise legal test depends on the requested relief.

If Apple narrows its hardware research requests and gains access, the case will move closer to product development. If the court continues denying them, the dispute may remain centered on individual conduct and recruiting.

OpenAI’s first consumer device provides another important context, but speculation should remain controlled. Its exact form, capabilities, and release schedule have not been fully established in the public record.

The lawsuit does show that Apple views OpenAI’s hardware effort as commercially relevant. OpenAI’s recruitment of experienced Apple designers and engineers reinforces that perception.

Yet recruiting talent does not establish that a viable product exists. A device must still solve difficult questions involving cost, reliability, privacy, battery life, interaction design, and daily usefulness.

The case could also influence how technology companies document interviews. Teams recruiting from close competitors may create stronger rules against discussing confidential projects.

Candidates could receive written reminders before technical presentations. Interviewers may need to stop sessions when a candidate begins describing protected systems.

Companies may also separate recruiting materials from product development repositories. That separation can reduce accidental exposure and create a clearer record of independent development.

Knowledge workers should care because the dispute reaches common career transitions. Employees often rely on examples from previous projects when demonstrating expertise.

The safer approach is to explain methods, responsibilities, and lessons without exposing protected names, files, measurements, designs, or unreleased plans. Confidential information does not become portable because it improves an interview.

Employers should care for a parallel reason. Restrictive warnings mean little when access remains active, devices stay unreturned, and personal storage mixes with company work.

The Apple and OpenAI dispute will not create one universal rule for technology hiring. It can still provide a detailed case study in how recruiting, security, and product competition collide.

For now, the clearest conclusion is procedural. Apple gained targeted access to OpenAI hiring records and testimony, while the court protected broader hardware research from accelerated review.

The next filings should show whether those records support Apple’s description of coordinated solicitation or OpenAI’s account of ordinary hiring distorted by litigation. Until then, neither narrative has been proved.

Watch the evidence rather than the rhetoric. The interview documents, sworn testimony, and injunction decision will reveal whether this fight stays narrow or reaches OpenAI’s hardware roadmap.