Apple sues OpenAI over stolen hardware secrets — and names the two engineers it says carried them out the door

Information Technology 4 min 80 sources

Apple's lawsuit says former staff took confidential files to OpenAI's hardware team. Brussels declares Meta's autoplay and infinite scroll addictive. SK Hynix pulls off the biggest foreign IPO in US history. And America's top cyber agency admits it wrote its emergency plan mid-emergency.

Key takeaways

Apple sued OpenAI for allegedly stealing hardware secrets and named two former Apple engineers who moved over — but a trade-secret case can only chase the files that left, not the know-how in someone's head.
The EU formally declared Meta's autoplay and infinite scroll "addictive" under its online-harms law, the same week Meta pulled a new AI feature after backlash.
SK Hynix's record $26.5bn US listing and a looming memory shortage show the AI race shifting from biggest model to cheapest one that works.

Apple went to court on Friday against OpenAI, accusing the ChatGPT maker of a “pattern of theft” of Apple’s trade secrets to build rival hardware [1]. The complaint names two people by name: Tang Tan, now OpenAI’s chief hardware officer, and Chang Liu, who left Apple for OpenAI in January [3]. It also names IO Products — the hardware startup founded by Jony Ive, Apple’s former design chief, which OpenAI bought in 2025 [3].

The specific accusation is concrete. Apple says Liu accessed its systems after resigning and downloaded “dozens” of confidential hardware files [3]. Apple’s statement says “significant evidence has emerged suggesting individuals employed by OpenAI wrongfully took Apple’s secret and confidential information regarding our unreleased technologies” [3]. OpenAI denies it: spokesman Drew Pusateri said “we have no interest in other companies’ trade secrets” [3]. A second Apple filing describes the scheme as operating “at every level” of the hardware effort [2].

Why now: OpenAI is building a physical device — a pocket AI gadget, the reason it bought Jony Ive’s team — and it hired hardware people to do it. Apple has the deepest bench of consumer-hardware engineers on Earth. When a rival needs that skill fast, the shortest path runs through Apple’s alumni. The angle for anyone who’s changed jobs in a specialised field: a trade-secret case turns on what you can document leaving — the downloaded file, the copied spec. That’s the part courts can see. It is rarely the biggest part that actually moved.

Brussels calls Meta’s design addictive — by name

The European Commission issued a preliminary charge on Friday: Facebook and Instagram’s autoplay and infinite scroll are “addictive” and breach the EU’s Digital Services Act, the bloc’s law against online harms [4][8]. The regulator’s language was unusually blunt. Autoplay and endless feeds “shift the brain into ‘autopilot mode,’ contributing to unhealthy habits and compulsive use,” it said [8]. It singled out the time children spend on the apps at night, and features like Reels and Stories, as risks Meta failed to assess [7].

This is a preliminary finding, not a fine — Meta gets months to respond, and it’s already disputing it, pointing to “Teen Accounts” that cap under-18 screen time at 15 minutes a day [7]. But the stakes are real: DSA penalties can reach 6% of global revenue. The same week, Meta pulled a brand-new AI tool, Muse Image, from Instagram after days of backlash over privacy and copyright — Hollywood agencies among the complainers [10][11]. Two pressures, one company, one theme: the features that drive engagement are the features now drawing regulators.

The compute economy keeps repricing

The money kept flowing toward the chips underneath AI. SK Hynix — the South Korean firm that makes the high-speed memory AI accelerators depend on — raised $26.5 billion in a US listing, the biggest foreign IPO in American history [6]. Its shares jumped on their Nasdaq debut [6], and the US urged it to build fabs — chip factories — on American soil [6]. The company’s chief executive warned the memory shortage will be its “worst ever” in 2027, with demand outstripping supply beyond 2030 [5].

That scarcity is reshaping the AI race itself. Analysts at CNBC described a shift away from building ever-bigger models toward cheaper, more efficient ones — cost and control now matter as much as raw size [9]. Hugging Face’s chief executive Clem Delangue put the same trend from the buyer’s side: companies start on rented frontier models — paying per use through an API — but as they scale, the bill pushes them toward cheaper open models they can run themselves [12]. Roughly half the Fortune 500 already pull models from his platform [12]. What to watch: when the constraint is memory and the cost is inference, the winning move stops being “the biggest model” and becomes “the cheapest one that’s good enough.”

The people you hired to defend you

Two smaller stories rhymed. CISA — the US agency charged with defending federal networks — admitted in a post-mortem on Friday that when a real incident hit in May, it had no response plan ready and “had to spend time building [a playbook] during the early stages of the incident” [13]. The trigger: a contractor had publicly exposed sensitive keys and credentials to government systems, flagged not by CISA but by an outside journalist [13]. The agency’s own lesson: write the plan before you need it, not during.

And in Florida, a ransomware negotiator — the specialist companies hire to bargain with hackers holding their data hostage — was convicted for secretly working with the gang he was supposedly negotiating against, helping extort the very US companies that trusted him [14]. Both stories land on the same uncomfortable spot: the defense you outsource is only as sound as the person or plan behind it, and you usually find out which the hard way.

02 · Lesson · why it matters

The part you can sue over is the smallest part that left

When an expert changes sides, the files they copy are what the law can see — and the least of what actually moved.

Two names and a stack of files

Apple’s lawsuit against OpenAI is unusually specific. It names two engineers — Tang Tan and Chang Liu — and accuses one of them of downloading dozens of confidential hardware files on his way out the door. That is the shape of almost every trade-secret case: a person, a set of documents, a claim that the documents crossed a line they shouldn’t have.

It reads like the whole story. It is the small part of it.

Think about what OpenAI actually got when it hired those two. Not mainly a folder of specs. It got people who spent years learning how Apple builds a device — what fails on the assembly line, which materials warp, how a battery and a screen and a radio fit in a space too small for all three. That knowledge doesn’t sit in a file. It sits in a person’s judgment, built slowly, and it walks out the door the moment they do — legally, in plain sight, with no download required.

What the law can read

A court can only rule on what it can see. A downloaded file is visible: it has a timestamp, a name, a copy. So the law reaches for the file, because the file is the part it can hold.

But the file was never the point. The real transfer — the sense of what works, the memory of a thousand small problems and how Apple solved them — is inside a head, and no court can subpoena a head. The line between “what I learned to do” and “what I took that wasn’t mine” runs straight through a person, and it cannot be cut cleanly. Every skilled worker carries their last job with them. That is not theft. That is what expertise is.

So the case fights over the visible sliver. Apple points at the files because the files are provable. Whether OpenAI needed them at all is a separate question — you don’t hire the person who built the thing and then depend on their downloads.

The line was drawn by whoever holds the most

Here is the part that poses as plain fact. Trade-secret law feels like a neutral fence: it protects a company’s hard work from being copied. And it does — real theft happens, and the fence is not nothing.

But look at who the fence favors. The company with the deepest stockpile of secrets is the one with the most to protect and the most to sue over. Apple has the largest bench of hardware engineers on Earth; almost anyone OpenAI hires to build a device will have passed through it. The law that guards “innovation” also, quietly, guards the incumbent — and shadows the worker who wants to use, at a new job, the skill they spent a decade building. The fence protects and it tilts. Both are true at once, and the tilt is easy to miss because it wears the language of fairness.

The pool everyone draws from

Zoom out and the lawsuit stops looking like a clean fight between two companies. The whole race to build AI hardware runs through a tiny pool of people — a few thousand engineers who know how to make consumer electronics at scale. They move between a handful of firms, carrying what they know each time. Jony Ive’s team went from Apple to a startup to OpenAI. Chang Liu went from Apple to OpenAI in January. The knowledge circulates because the people do.

This isn’t only about famous engineers. If you have ever changed employers in a specialised field, you took something with you that no exit interview could recover — a way of seeing the problem, a set of instincts about what will break. You were the carrier. Most of us have stood near that blurry line without noticing it was there. The reader is inside this web, not watching it from above.

What no one in the room can see

Notice how little any single seat here actually knows. Apple cannot see inside Chang Liu’s head; it can only see the files, so it sues over the files. OpenAI cannot fully separate what its new engineers learned from what they remember of Apple’s documents — the two are tangled in the same minds. The court will rule on the sliver it can read and call it a verdict, while the largest transfer already happened in a way no rule reached.

The thing worth carrying out of today isn’t a side. It’s how much of what moves in the world moves like this — invisibly, inside people, past the fences we built to catch it. Companies guard what’s written down and lose what’s learned. When you next hear that someone “stole” an idea, it’s worth holding the word a little more loosely: the copyable part is almost always the smallest part that left.

03 · Lab · your turn

What Walks Out the Door

Rehearse trying to stop a departing expert's knowledge from leaving, and feel how the law can only lock the small copyable part while the larger learned part walks free.

04 · Hope · carry this

What no company can lock down, no company can hoard. Knowledge lives in people, so it travels — and every field gets better because what one mind learns rarely stays in one place for long.