On July 10, Apple sued OpenAI in the U.S. District Court for the Northern District of California for trade secret misappropriation and breach of contract. The complaint cites more than 400 former Apple employees now working at OpenAI, including hardware chief Tang Tan (24 years at Apple, VP of product design for iPhone and Apple Watch) and engineer Chang Liu, who Apple says downloaded dozens of confidential files before leaving.

That same weekend, Elon Musk and Sam Altman turned the filing into another round of their running public fight on X. Musk called Altman “Scam Altman” and bragged about SpaceX’s space-based data centers. Altman fired back, accusing Musk of selling public market investors a short-term story. It’s entertaining. It’s not the point.

The point sits two days earlier. Satya Nadella published an essay on what he called the “Reverse Information Paradox”: when you use an outside AI model, you have to hand it proprietary knowledge for it to work well, so you end up paying for your own intelligence twice. Nadella calls that leaked knowledge “intelligence exhaust,” the prompts, corrections, and daily inputs your team feeds a vendor’s model, and he proposes building a “trust boundary” around that combined pool of people and tokens.

Apple’s lawsuit is the older, more direct version of the exact same problem. Institutional knowledge doesn’t only leak into a model’s weights through a prompt. It walks toward the exit with the person who built it. And no trust boundary, no model architecture, no evals framework stops that.

What the complaint actually says

Apple isn’t accusing OpenAI of hiring its former employees. Hiring talent away from a competitor is legal and ordinary. What Apple alleges is a coordinated pattern of conduct to extract specific knowledge before and after those hires.

Per the filing: Tang Tan emailed himself information about Apple’s suppliers before resigning, used confidential Apple project code names during OpenAI’s recruiting process, and asked candidates still employed at Apple to bring hardware components to interviews for “show and tell” sessions. Chang Liu left Apple in January 2026 without signing the company’s confidentiality reminder, scheduling an exit interview, or confirming he’d returned his devices, and in that window downloaded detailed technical files on unreleased products. Apple also describes internal guidance it says OpenAI used to coach new hires on evading Apple’s exit-security checks.

OpenAI’s response was short: “We have no interest in other companies’ trade secrets. We remain focused on building innovative technology.” Apple says it sent a notice in February that went unanswered before it filed suit.

Nobody outside the courtroom knows how this resolves, and speculating on the merits isn’t the point of this piece. What’s confirmed in the public filing is the scale: more than 400 former Apple employees now working at OpenAI, plus Jony Ive himself, whose design studio io Products (also named in the suit) OpenAI acquired in 2025. Apple says the recruiting from its iPhone product design group was aggressive enough that it had to rebuild parts of that team.

The other half of Nadella’s paradox

Reread Nadella’s thesis. Every time your team feeds an outside model more context so it answers better, you’re handing over a piece of your edge. It’s a real and specific risk for this decade: knowledge leaves through ordinary tool use, without anyone stealing anything on purpose.

The Apple-OpenAI case describes the same risk with fifteen fewer years of new vocabulary. Knowledge doesn’t need a prompt to cross from one company to another. It needs an accepted resume, a signed offer letter, and a corporate laptop nobody asked back on time. It’s the oldest leak in any IP-intensive industry, and it’s more complete than any prompt-based intelligence exhaust: a person doesn’t just know what worked, they know why it worked, what was tried and failed, and which specific vendor solves which specific problem.

A well-designed trust boundary, the kind Nadella describes, can limit what enters a prompt. It can’t limit what an engineer remembers when they sign with your competitor. Those are two separate layers of risk, and companies only watching the first one (their AI vendor’s data policy, the data processing agreement, encryption at rest) have a sizable blind spot on the second.

The blind spot in your due diligence

Here’s the practical problem for any company hiring technical talent or bringing in outside AI vendors with system access. Most 2026 due diligence checks a vendor’s privacy policy, a data processing agreement, maybe a SOC 2 report. Almost none systematically check what specific knowledge and what concrete access leaves with every technical hire coming or going.

Three questions almost nobody writes down:

What access did the departing person actually have, and who is revoking it, on what timeline? Not just email. Code repositories, architecture documentation, vendor credentials, internal knowledge bases. Chang Liu reportedly went weeks without a formal exit interview, per Apple’s complaint. That process gap, not some sophisticated act of espionage, is what let confidential files sit outside anyone’s control.

What are you actually asking an incoming technical candidate for, and are you crossing a line? Asking a candidate to “bring examples of past work” is normal. Asking for unannounced project code names from their current employer, or physical components they shouldn’t have outside the building, isn’t. If your technical recruiting process doesn’t have an explicit line here, individual recruiters are drawing it on the fly.

Does your contract with an AI vendor cover what happens when their staff turns over? This is where Nadella’s paradox and the Apple case touch. You sign a data processing agreement with a model provider. It says what the vendor does with your data. It rarely says anything about what happens when the engineer who had access to your enterprise account, your historical prompts, or your custom configuration leaves that vendor for a direct competitor.

None of these three questions require litigation lawyers or Apple-scale corporate security budgets. They require someone in your organization to have the explicit authority and mandate to ask them before they become a problem, not after.

What IQ Source does about this

When we help a mid-sized company assess its AI exposure, the conversation almost always starts at the visible layer: which model to use, which data policy to sign, what technical boundary to put around prompts. That layer matters, but it’s half the story.

The other half is process, not technology: how you document the departure of someone with access to sensitive systems or knowledge, what a technical interview can and can’t ask for without crossing into unfair competition, and who in your organization actually has the authority to block a hire or an access grant when the risk justifies it. Those are talent governance questions, not model architecture questions, and they’re exactly the kind of gap we map during AI Maestro discovery before we touch a single tool.

If your company is scaling its technical team, bringing in AI vendors with access to internal systems, or has simply never formalized what happens when someone with sensitive knowledge leaves, this audit is worth doing before a case like this one forces it on you under pressure.

Book a talent and AI vendor governance conversation →

For more on why governance matters more than which model you pick, read the model is a commodity, governance is the moat. And for the deeper split in technical talent markets driving all of this, see the $570K engineer paradox and your team.