OpenAIWriting the comprehensive article Hits Back At Apple Trade Secrets Lawsuit

Have you ever watched two companies that used to smile for the cameras suddenly start circling each other like they forgot they once shared a product announcement? That is the feeling hanging over this fight. One side says its hardware secrets walked out the door. The other side says the door was left half open on purpose, then someone pointed at the people who walked through it. I have covered enough talent wars to know this pattern. The public story is always about stolen files. The private story is usually about who gets to build the next device people actually want to carry.

Why This Trade Secrets Clash Suddenly Matters

In late summer filings, the ChatGPT maker asked a federal judge in Northern California to refuse a preliminary injunction. That order, if granted, would have blocked the company and two former Apple specialists from accessing, using, or disclosing alleged confidential material while the case moves forward. The response was blunt. This dispute, the filing argued, is a mess of Apple’s own making.

Apple sued in July. It named the AI lab and two ex-employees: Tang Tan and Chang Liu. Tan spent roughly 24 years inside Apple and once served as vice president of product design for the iPhone and Apple Watch. He is now chief hardware officer at OpenAI. Liu was a senior systems electrical engineer. He now sits on OpenAI technical staff. Those titles alone explain why Cupertino is tense. Hardware is not a side project anymore. It is the next battlefield after models, chat windows, and voice assistants.

Apple’s complaint claimed Tan used confidential project code names in interviews and asked at least one candidate to bring “actual parts” from prior work for a show-and-tell. It also said Liu kept reaching company files after he left by exploiting what Apple called a rare, previously unknown authentication bug. OpenAI denied the core accusations and tried to reframe them as ordinary offboarding friction plus lawful employee movement under California rules.

This dispute is a mess of Apple’s own making, and it is trying to blame everyone else.

That sentence is doing a lot of work. It is not just legal color. It is a strategy. If the judge accepts that Apple encouraged personal cloud accounts for work and rushed some leavers out the building, then “trade secrets walked away” starts to look like a process failure instead of a heist. I am not saying that framing wins the case. I am saying it is a smart way to turn a morality play into an operations debate.

The Two People At The Center Of The Complaint

Tan is the headline name because design leadership is scarce. Twenty-four years is not a short tour. People who spend that long inside one product culture carry instincts that no slide deck can copy. How a hinge feels. How a thermal budget gets negotiated with industrial design. How a supplier meeting actually ends when the room goes quiet. Those instincts are not automatically illegal to take to a new job. The line is between general skill and protected detail. Courts spend years drawing that line, and they still argue about the pencil marks.

OpenAI said Tan returned prototypes before he left and kept only nonconfidential material. Any parts brought into interviews, it added, were old or publicly available and used as demonstration aids. That is a careful phrase. Demonstration aids can look innocent in a courtroom and still look reckless in a recruiting room. If you have ever sat in a hiring loop, you know how quickly a physical object becomes a conversation piece. The legal question is whether the object carried protected information or just nostalgia with a serial number filed off.

Liu’s story is messier because it involves access after departure. Apple says he used a bug to keep reaching internal material, including a confidential power-converter circuit schematic, and that he used proprietary information to train an AI agent in March. Some of that claim, Apple said, surfaced from a MacBook OpenAI turned over on August 21 after weeks of delay. Large parts of the public record are redacted, which is normal and also convenient for both sides. Redactions keep recipes hidden. They also keep the audience guessing.

OpenAI’s answer on Liu is almost administrative. Any Apple files he touched after leaving were retrieved because former colleagues asked for help finding documents or answering questions about old work. If that is true, it is a classic offboarding hole. Knowledge leaves with a person, then the people who stay ping the person who left because the wiki is a graveyard. If it is not true, it is a cover story. Discovery exists to sort those two possibilities, not press releases.

What Apple Calls New Evidence

On the same Monday as OpenAI’s pushback, Apple told the court it had new material against Liu and wanted faster discovery. The MacBook, it argued, was only a sliver of what defendants had produced, and only after delay. Apple said it was not on a fishing trip. It said secrets were in use and evidence was being destroyed. That last clause is the kind of sentence that makes judges sit up. Destruction claims change timelines. They also raise the temperature in a way that can backfire if the proof is thin.

OpenAI treated the same laptop episode as proof of Apple’s offboarding gaps. The company said Apple encouraged personal iCloud accounts for work, which blurs the line between a family photo library and a schematic. It also said some departing staff were escorted out quickly, leaving little time to return devices, transfer files, or hand off responsibilities. I have seen that ritual in more than one campus. Security loves the escort. Teams hate the vacuum it creates the next morning.

Apple blames employees who left the company for taking their personal accounts with them even though Apple encouraged them to use personal accounts for work.

There is a practical lesson here that has nothing to do with who wins. If your company tells people to mix personal cloud storage with official work, you are building tomorrow’s exhibit list. That is not a legal opinion. That is just pattern recognition. Mixed accounts create mixed custody. Mixed custody creates stories that sound like theft even when the human on the other end thought they were helping a former teammate find a spreadsheet.

From Friendly Partnership To Hardware Rivalry

Remember when this relationship looked almost sweet? In June 2024 the two firms announced a deal that put ChatGPT inside Apple products. Siri and other Apple Intelligence features could hand some requests to the chatbot. It was presented as complementarity. Apple kept the device and the privacy pitch. OpenAI got distribution that money alone cannot buy.

Then the hardware ambition stopped being theoretical. In May 2025 OpenAI acquired io Products, a startup tied to former Apple design chief Jony Ive, in a deal reported around $6.5 billion. Ive and his circle moved into a close working relationship with the lab as it chased a new class of AI-first gadgets. Tan left Apple, joined io, and later became OpenAI’s chief hardware officer. Apple’s lawsuit does not name Ive as a defendant. That absence is loud. It keeps the case pointed at files and two employees rather than at a design legend with a global brand of his own.

Apple did not sit still on models either. In January it locked in a multiyear arrangement to use Google Gemini as a foundation for future Apple Intelligence features, including a rebuilt Siri. Partnerships, in this industry, are seasonal. They last until a device roadmap makes them awkward. Perhaps the most interesting aspect is how fast “partner” can become “someone who might ship a competing object that sits on a nightstand.”

The July complaint went further than employee files. Apple also alleged that OpenAI approached some of Apple’s business partners using confidential information. OpenAI denied wrongdoing and cast the suit as an effort to slow a rival and scare people who want to leave. It said roughly 400 former Apple employees have joined its hardware efforts. That number is doing political work. It says this is a labor market, not a burglary.

California Law And The Right To Walk Out

OpenAI leaned hard on a simple idea. California generally lets workers move between competitors. Apple cannot treat ordinary departures as unlawful just because the destination is a rival building AI devices. “Apple cannot stop the next 400 employees from leaving by filing baseless lawsuits,” the filing said. It added that people can leave a company that has struggled to adopt AI and join a startup building new products. Apple may dislike those choices. Dislike is not the same as a trade-secret claim.

In my experience, that argument plays well with engineers and less well with supply-chain lawyers. Engineers hear freedom. Lawyers hear customer lists, yield data, fixture drawings, and the unofficial map of which vendor can hit a tolerance next quarter. Both groups can be right at the same time. Skill travels. Spreadsheets should not. The case will live in the gap between those two sentences.

• General know-how and craft usually travel with the employee.
• Specific schematics, yields, and unreleased code names often do not.
• Interview theater with physical parts creates avoidable risk.
• Post-exit file access is the fastest way to turn a clean departure into a complaint.
• Personal cloud accounts used for work make custody arguments ugly.

None of that list decides the injunction. It does explain why this file will keep producing headlines. Every discovery dump can look like a smoking gun if you squint, or like leftover clutter if you do not.

What A Preliminary Injunction Would Actually Do

A preliminary injunction is not a final verdict. It is a pause button with teeth. Apple wants the court to freeze access and use of alleged secrets now, before a full trial. OpenAI wants the opposite: no freeze, because the request is overbroad and the facts are contested. Judges weigh likelihood of success, irreparable harm, balance of hardships, and public interest. That sounds tidy. In practice it is a gut check about whether waiting until trial would let a rival ship something that cannot be unshipped.

Hardware timelines make that gut check harder. A language model can be swapped. A tooling path for a new gadget cannot. If Apple can show that a unique converter design or a manufacturing trick is already inside an OpenAI workflow, delay looks dangerous. If OpenAI can show that the disputed material is old, public, or retrieved only to help leftover colleagues, a freeze looks like a competitive weapon dressed as emergency relief.

I keep coming back to the MacBook. Limited production after delay is a gift to the plaintiff because it supports a narrative of concealment. It is also a gift to the defense if the machine mostly shows messy personal storage rather than a curated vault. We do not have the unredacted picture. Until we do, both stories remain plausible, which is exactly why both sides filed on the same Monday.

The Offboarding Problem Nobody Wants To Own

Let me be plain. Big tech is sloppy at goodbyes. Badge cutoff is fast. Knowledge transfer is slow. Managers promise a two-week handoff and then security shortens it to an afternoon because someone is joining a competitor. The departing engineer still has a personal laptop that once synced a work folder “just for the weekend.” A teammate later texts, “Where did you put the power board notes?” Helpfulness turns into a log line. That log line turns into a paragraph in a complaint.

OpenAI’s filing tries to make that culture Apple’s problem. Fair enough as advocacy. It is also an industry problem. If you run a hardware team, the boring fixes matter more than the courtroom poetry.

1. Separate personal and corporate cloud identities before an employee ever touches a schematic.
2. Collect devices with a documented chain, not a hallway scramble.
3. Give leavers a short, supervised window to return files instead of a sudden escort with no checklist.
4. Record who asked a former colleague for help, and why.
5. Train interviewers never to request physical parts from a candidate’s last employer.

Those steps will not thrill a keynote crowd. They will keep your next discovery letter shorter. That is the unglamorous side of trade-secret hygiene, and it is the side this lawsuit keeps exposing.

Why Hardware Secrets Feel Different From Model Weights

Software people sometimes shrug at these cases. Weights leak. Prompts leak. A good researcher reconstructs a lot from papers and vibes. Hardware is less forgiving. A power-converter circuit is not a vibe. It is a drawing with tolerances, vendor notes, and failure modes that cost real money when a factory line stalls. Supply-chain knowledge is even stickier. Who can hold a micron. Who will miss Christmas. Who quotes a fantasy price until week twelve.

That is why Apple’s complaint talks about design, manufacturing, and operations in the same breath. The product is the object plus the path that makes the object at scale. OpenAI’s consumer-device push only makes that bundle more valuable. If you are building a new class of gadget, you need people who have already paid the tuition of shipping millions of units. Those people used to sit in one campus. Now they sit in several. The market calls that mobility. The former employer calls it risk.

Does that mean every departure is suspect? Of course not. Treating every exit as theft would freeze the whole Bay Area. It would also collapse the story OpenAI is telling about California public policy. Workers move. Startups form. Incumbents adapt or they sulk in court. The honest version is less slogan and more accounting: list the files, date the access, explain the purpose, and stop recruiting with show-and-tell props that look like contraband.

Recruiting Theater And The Show-And-Tell Problem

I have sat through interviews where a candidate pulled out a personal side project and the room lit up. I have also seen rooms go cold when someone described an unreleased fixture in too much detail. The difference is obvious in hindsight and foggy in the moment, especially when the interviewer used to work on the same family of products. Familiarity feels like shop talk. Shop talk can still be a disclosure.

Apple’s allegation that Tan used internal code names and invited physical parts into interviews is therefore not a small color detail. It is an attempt to show intent and sloppiness at the same time. OpenAI’s reply that the parts were old, public, or merely teaching tools is an attempt to shrink the moment back to ordinary hiring. A judge may never need to decide which culture is cooler. The judge may only need to decide whether the moment crossed a line that injunction law cares about.

If I were advising any hardware startup tomorrow, I would ban souvenir parts from interview loops. Not because every bracket is a secret. Because the photograph of the table will look terrible two years later in an exhibit binder. Optics are not law. Optics still move settlements.

Partners, Suppliers, And The Quiet Second Front

The claim that OpenAI used confidential information to approach Apple partners deserves more attention than the celebrity names. Partners are how phones and watches become real. If a rival can skip the awkward first year of learning who actually delivers, the clock changes. Apple has always treated that map as crown-jewel material. OpenAI denies misuse. Denial is required. Proof will live in emails, calendars, and the odd slide that should never have left a laptop bag.

This is also where the case could widen. A fight about two employees can become a fight about a network. Networks leave traces. They also create witnesses who do not want to pick a side in public. That is why both companies may end up disclosing more than they want about recruiting scripts, vendor conversations, and internal labels for unreleased work. Unusual transparency is the tax on unusual lawsuits between unusual partners-turned-rivals.

Tables flatten nuance, but they help you see the symmetry. Each row is the same fact pattern told in two dialects. Discovery is the translator. It is slow, expensive, and allergic to press cycles.

What Investors And Builders Should Watch Next

If you care about markets rather than motions, watch three things. First, whether the court grants any freeze that touches product work, not just two inboxes. Second, whether more former Apple specialists are named or quietly put on leave from sensitive work. Third, whether either company’s device calendar slips in a way that looks like legal friction rather than ordinary engineering delay.

Also watch the partnership layer. ChatGPT inside Apple software and Gemini under future Siri can coexist with a hardware lawsuit, at least on paper. In practice, legal teams start attending meetings that used to be about latency and battery curves. That changes tone. Tone changes shipping dates. Shipping dates change narratives about who is behind in AI.

I do not buy the idea that this case is only theater. I also do not buy the idea that it is a simple theft novel with a neat villain. Real companies are simultaneously brilliant and clumsy. They invent beautiful objects and then store the drawings in the wrong cloud. They hire the only people who can build the next object and then act shocked when those people already know how factories work.

A Longer View Of The Talent Map

Four hundred former Apple people inside one rival hardware effort, if that figure holds, is not a rounding error. It is a campus. It means entire informal networks moved: mechanical folks who know antenna trades, electrical folks who know power budgets, operations folks who know which line in Asia will tell you the truth. You cannot litigate a campus out of existence. You can try to fence the parts of their memory that still belong to the old employer. Fences leak. That is why contracts, clean rooms, and audit logs exist. It is also why they fail when culture treats them as optional.

There is a human register here that filings flatten. People leave because they are bored, blocked, or excited. They leave because a friend called. They leave because the next object might be weirder than another yearly phone. Calling those motives unlawful is a stretch. Ignoring the files that sometimes travel with the motives is also a stretch. Grown organizations hold both thoughts.

OpenAI’s line about workers fleeing a company that “struggled to adopt AI” is a jab dressed as policy. Apple’s line about rare bugs and destroyed evidence is a jab dressed as forensics. Readers can enjoy the theater and still ask the adult question: what, exactly, was accessed after the badge died, and what was done with it in March when an internal agent was being trained? Until those dates and documents are public in usable form, we are reading competing novels with the same character list.

How This Could End Without A Cinematic Trial

Most fights like this do not end with a jury gasping at a schematic on a giant screen. They end with narrowed claims, special masters, forensic images, and a quiet protocol about what certain employees may not touch. Sometimes money moves. Sometimes a hire gets walled off from a product surface. Sometimes both sides realize that a public trial would teach competitors more than it teaches the court.

A settlement would not mean the accusations were empty. It would mean the price of proving them exceeded the value of the headline. An injunction would not mean the entire hardware group is radioactive. It would mean the court thought some slice of harm could not wait. Either outcome will be spun as victory. You should ignore the spin and look at who still gets to open which folder.

Rough map of the live questions:
  Who accessed what after exit dates?
  Were parts in interviews protected or public?
  Did partner outreach use nonpublic maps?
  Is offboarding policy a defense or a confession?
  Does California mobility law shrink the remedy even if some facts cut against the defense?

Those questions are less glamorous than “who stole the future of the smartphone.” They are the actual case. Glamour is for keynotes. Folders are for discovery.

What The Rest Of The Industry Should Steal From This Mess

Steal the process, not the files. Write down where work product lives. Stop pretending a personal photo stream is an acceptable vault for a converter drawing. Tell recruiters that props from a prior employer are not clever. Tell remaining staff that “quick question for the person who left” should go through counsel or a recorded handoff channel when the destination is a direct rival. None of this is exciting. All of it is cheaper than a Northern California docket number.

Founders building AI devices should assume their first serious competitor already employs people who used to sit in the rooms they want to enter. That is the market now. Hiring those people is legal in broad strokes. Asking them to unpack last year’s unreleased stack is how you buy a lawsuit. There is a clean way to use their craft: put them on a fresh problem, give them a clean bench, and keep yesterday’s drawings on the other side of a wall.

Incumbents should assume some of their best people will leave anyway. Escorting them out in a panic without a file plan is how you manufacture the other side’s best defense. If you encouraged personal accounts, you do not get to act stunned when those accounts leave with the human. That is not sympathy for sloppy leavers. That is accountability for sloppy systems.

The Relationship That Will Stay Awkward Either Way

Software integration can continue while hardware lawyers sharpen pencils. Users do not care about docket numbers when they ask a phone a question and get a fluent answer. Executives do care, because every shared feature review now carries a shadow agenda. Who is learning whose interface habits? Who is stalling whose gadget? Who is collecting whose alumni?

That awkwardness may be the real product of this case, more than any injunction. Two firms taught the public they could collaborate. Then one started hiring the other’s hardware nervous system. Then the filings arrived. Consumers will keep using whichever object works. Courts will keep asking for logs. Reporters will keep shortening a custody fight into a cartoon about theft. The cartoon is easy. The custody fight is the truth-shaped thing underneath.

So where does that leave a reader who just wanted a clean scorecard? There isn’t one yet. Apple has alleged specific after-exit access, a schematic, agent training, interview props, and partner approaches. OpenAI has denied the sting of those claims, pointed at offboarding culture, wrapped itself in employee-mobility law, and asked the court not to hit pause on its hardware push. Both descriptions can contain real facts. Both can also contain performance.

I will keep watching the unglamorous pieces: the next production of devices, the scope of any freeze, and whether more names join the caption. If those pieces stay quiet, this may shrink into a sealed sideshow. If they get loud, we will learn more than either company intended about how modern gadgets are actually born. That, if I am honest, is why the case is worth the word count. Not the rivalry branding. The plumbing.

Apple may not like those choices. But it cannot claim those choices are unlawful.

That closing note from the defense is a dare. The plaintiff will try to prove that some choices were not mere career moves. Between those two sentences sits a pile of redacted pages, a MacBook, two careers, a former partnership, and a future device that still does not have a public shape. The interesting part is not who shouted louder on Monday. The interesting part is what the next folder shows when someone finally has to open it in a room without cameras.