What Apple v. OpenAI Teaches Every Dev Team About Offboarding and Insider Threat

On July 10, 2026, Apple sued OpenAI. Not over AI. Over hardware. And when you read the actual complaint, it doesn't read like a patent fight. It reads like a heist.

But here's the thing I want you to take away, as someone who builds software and probably manages access for a small team: the failures at the center of this case are not exotic. They're the same boring gaps almost every company has. Unreturned laptops. Access that outlives employment. A departing colleague who becomes a channel. If a trillion-dollar company with locked buildings, guards, and code-named projects can get burned by this, your five-person startup absolutely can.

So let's walk through it as an engineering post-mortem, not gossip. The full case file is linked at the bottom if you want to read the primary source yourself.

What Apple is actually alleging

The case is Apple Inc. v. Chang Liu, Tang Yew Tan, OpenAI, and io Products (N.D. Cal., Case No. 5:26-cv-07078). Apple brings two federal trade-secret claims under the Defend Trade Secrets Act plus breach-of-contract claims against two former employees.

Two people sit at the center. Tang Tan spent about 24 years at Apple, most recently VP of product design for iPhone and Apple Watch, and is now OpenAI's Chief Hardware Officer. Chang Liu spent eight years as a senior system electrical engineer and left for OpenAI in January 2026. Apple says more than 400 former Apple employees now work at OpenAI, and that a pattern of theft runs through that migration. OpenAI's public response is a flat denial: it says it has no interest in other companies' trade secrets.

Everything below is an allegation. None of it is proven, and OpenAI disputes it. But the alleged mechanics are worth studying regardless of who wins, because they map onto real controls you either have or don't.

The security anatomy: five failure modes

1. Offboarding that doesn't actually revoke anything

Apple says Liu simply didn't return an Apple-issued laptop when he left, ignored outreach to confirm he'd returned his devices, and skipped the exit interview. That device was still authenticated to Apple's network. The lesson is uncomfortable and universal: on most teams, "offboarding" is a checklist someone gets to eventually. The gap between someone's last day and the moment their access is truly gone is where breaches live. Access revocation has to be same-day and verified, not a ticket that ages in a backlog.

2. Access that outlived employment (the deprovisioning gap)

This is the part that should make every engineer wince. Apple alleges Liu found he could still reach Apple's cloud file storage weeks after leaving, thanks to an authentication bug, and that instead of reporting it he exploited it — allegedly reacting with a message that amounted to "this is so funny" — and downloaded dozens of confidential files, including a compilation running over a thousand pages. Whether or not that's what happened, the underlying risk is one you can check for today: does your identity provider truly kill every session, token, and integration when someone leaves? SSO logout is not the same as revoked refresh tokens. Disabled email is not the same as rotated API keys. Deprovisioning gaps are silent until someone walks through one.

3. The insider who is still on the inside

The most instructive thread isn't the person who left — it's the one who stayed. Apple alleges Liu kept a close line to a still-employed Apple colleague who fed him updates on active projects for months before she, too, left for OpenAI, and that he coached her on copying files in ways designed to "avoid trouble with the security team." No firewall stops this. It's social, not technical. The mitigations are behavioral and observational: least-privilege access so any one person can only see their slice, and logging and alerting on anomalies like bulk downloads from a file store or a repo. If one account pulls a thousand pages, something should page someone.

4. Weaponizing your own security runbook

Here's the twist that's pure security theory made real. Apple alleges Tan retained an internal "Need to Know" document describing Apple's employee-departure security procedures, and that it was shared with new hires so they could preview — and evade — Apple's exit checks before giving notice. Sit with that. Your offboarding playbook, in the wrong hands, becomes an evasion manual. The takeaway isn't "hide the runbook better." It's Kerckhoffs's principle applied to insider risk: assume adversaries know your process, and design controls that hold up even when the other side knows exactly what you check for. Security that depends on the secrecy of the procedure is already broken.

5. The supply chain is part of your attack surface

Apple also alleges OpenAI leaned on Apple's partners — including getting a trusted supplier to run a proprietary metal-finishing process by misleading them into thinking Apple had approved it. For those of us who ship software, the parallel is direct: your vendors, contractors, and shared suppliers inherit your trust and your secrets. Confidentiality terms in a contract are necessary, but they don't enforce themselves. Trust extended down the chain is trust you have to actively verify.

The builder's takeaway: an offboarding checklist that actually holds

Forget the billion-dollar drama. Look at how ordinary the failures were, and steal the fixes. Here's what I'd make non-negotiable, and none of it needs an enterprise budget:

• Revoke on day one, not "soon." Deprovision the identity in your SSO/IdP first — that cascades to most apps. Then explicitly kill VPN, cloud file storage, and any service the IdP doesn't cover.
• Rotate shared secrets on every departure. API keys, service-account credentials, database passwords, signing keys. Anything the person could have seen is now compromised until rotated.
• Treat an unreturned device as an active incident. A laptop that's still authenticated to your network isn't an HR loose end — it's a live credential in someone else's hands.
• Run access reviews after people leave. Deprovisioning gaps are invisible unless you go looking. Periodically confirm that ex-employees actually have zero access.
• Log and alert on bulk access. Mass downloads from a repo or file store, or access to folders outside someone's project, should trigger a human review.
• Default to least privilege and need-to-know. The blast radius of any single insider is exactly the scope you granted them.
• Extend the same rigor to contractors and vendors. Offboard them too. Their access is your access.

IP agreements and NDAs matter — they're what makes this a lawsuit rather than a shrug — but paper doesn't stop exfiltration. Technical controls do the enforcing. The contract is the consequence; the access control is the prevention.

A note on fairness

Everything here is drawn from Apple's complaint and reporting on it. These are allegations Apple must still prove, OpenAI denies them, and the separate agreement that puts ChatGPT inside Apple Intelligence is explicitly not part of this suit. I'm not here to tell you who's right. I'm here to point out that the security failures described are the same ones sitting quietly in most codebases and IT setups — including plenty back home — waiting for the wrong person to leave on the wrong terms.

Read the complaint. Then go check your own offboarding. That's the whole assignment.

Sources

• Full case file (primary source): Apple Inc. v. Chang Liu et al. — Complaint (PDF)
• CNBC — Apple sues OpenAI alleging trade secret theft, says scheme was 'at every level'
• TechCrunch — Apple sues OpenAI over alleged trade secret theft
• 9to5Mac — Apple sues OpenAI, accuses ex-employees of stealing trade secrets
• Fortune — Apple's lawsuit against OpenAI reads like a corporate spy thriller
• NBC News — Apple sues OpenAI and two former employees
• Quartz — Apple is suing OpenAI, saying it stole trade secrets