Field Notes / Breach Analysis
An Engineer Left Apple for OpenAI, Then Walked Back In Through a Bug Nobody Closed. Here Is What Your Offboarding Misses.
Apple is suing OpenAI and two former staff. The court filing describes a departing engineer who kept his work laptop, found an authentication bug, and used it to reach Apple’s internal file storage months after he left, pulling down dozens of confidential files. OpenAI says the claims have no merit. Whoever is right, the security failure is one almost every company shares: access that outlives the employee.
01 What Apple alleges
pple filed in federal court in Northern California, naming OpenAI and two former employees: Chang Liu, a senior electrical engineer, and Tang Tan, an Apple veteran who worked on the iPhone and Apple Watch and is now OpenAI’s chief hardware officer. Apple’s language is blunt. It says the theft happened, in its phrasing, at every level, from technical staff to a chief hardware officer, and in coordination with business partners CNBC 2026 .
The backdrop is OpenAI’s push into hardware, the effort led by former Apple design chief Jony Ive after OpenAI bought his startup. Apple’s case is that people who left to build a rival device carried Apple’s confidential work with them Bloomberg 2026 . Tang Tan is accused of using Apple’s internal product codenames to draw information out of OpenAI job candidates. Chang Liu is accused of something more mechanical, and more instructive, which is where we will spend the rest of this piece.
Read all of it as allegation. A complaint is one side’s story, and OpenAI disputes it. But the mechanism Apple describes is specific enough to learn from on its own terms, whether or not a jury ever agrees the intent was theft.
02 How the access survived
Here is the chain Apple lays out. Read it the way an operator reads an attack path, one failed control at a time.
| Step | What Apple alleges happened |
|---|---|
| Retain | The engineer kept his work-issued Apple laptop after leaving the company |
| Discover | He found a rare, previously unknown authentication bug, which Apple calls a zero-day |
| Reach | He used it to reach Apple’s internal network file storage while employed at OpenAI |
| Take | He accessed and downloaded dozens of files, many marked confidential |
| Persist | The access kept working long enough that he joked, in a message, that he could still get in |
Apple quotes one message to a former colleague in which the engineer wrote that he had found he could still reach the network storage, and that it was funny TechCrunch 2026a . Set the humour aside and look at what that one line proves. Weeks or months after a departure, a former employee’s route into internal storage was still open, still usable, and still quiet enough that the discovery came as a surprise to him and, apparently, to nobody watching on the other side.
03 The control that failed
Every organisation has an offboarding process. Someone leaves, HR files the paperwork, IT disables the account, the laptop goes back in a drawer. On the org chart it looks clean. In practice it leaks, because offboarding is a checklist run by tired people under time pressure, and a checklist only covers what someone remembered to put on it.
Look at what had to go wrong here, in order, for the access to survive. The device was never collected or remotely wiped. The accounts and credentials tied to it were never fully revoked. The authorisation path that the device and identity used still resolved to real data. And nothing flagged a former employee’s identity reaching internal storage weeks after they left. Any one of those controls, held properly, would have closed the door. All four failed together, which is the ordinary way breaches happen: not one dramatic hole, but a row of small ones that line up.
04 What it changes for you
The instinct after a story like this is to reach for a policy. Write a stricter offboarding rule, add a line to the leavers’ checklist, and move on. That is necessary and it is not enough, because the failure here was not a missing policy. Apple is not a company without an offboarding process. The failure was that nobody tested whether the process actually closed the door, and a departing engineer did the test for them by accident.
This is the same gap we find across sizes and sectors. Controls exist on paper. Nobody has tried to beat them. A red team engagement is that try. An operator takes the posture of someone who just left with a grudge and a laptop, and looks for the account that still authenticates, the token that never expired, the internal service that trusts a device it should have forgotten. The point is not to catalogue every setting. It is to reach the data and prove the path, so you fix the one that is real instead of the fifty that are theoretical.
05 The quieter lesson
There is a second thread worth pulling, because it will matter more over the next year, not less. The engineer in Apple’s account did not need malware or a criminal toolkit. He needed a device he was allowed to keep, an identity nobody switched off, and one bug. That is a low bar, and it is getting lower as the tools for finding bugs get better and cheaper. The same week this suit was filed, the field was still absorbing OpenAI’s own disclosure that its models had chained an authentication weakness into a live intrusion with no human driving. Whether the hand on the keyboard is a departing employee or an automated agent, the way in is the same class of flaw.
The defensive answer to both is the same, and it is not a product. It is testing your own environment the way the attacker will, on a schedule that matches how fast your surface changes, with a human confirming what is real. Leavers, tokens, trusted devices and forgotten services are not a once-a-year concern. They change every week you hire, fire, ship and integrate.
06 What to do this quarter
This maps onto frameworks your board already tracks. NIST CSF puts Identify and Protect ahead of Detect for a reason, and the identities and devices tied to leavers belong in that inventory. ISO 27001 asks specifically that access rights be removed on termination and that assets be returned. The OWASP work on broken authentication and broken access control names the exact flaw class at the centre of this case. The list below is what turns those requirements into something you have actually tested.
- Every leaver has device, account and token access revoked on their last day, and the device recovered or remotely wiped
- Credentials and service tokens are scoped to least privilege and expire, so a forgotten one cannot reach internal data indefinitely
- Monitoring alerts when a departed or disabled identity authenticates or reaches storage, and someone owns that alert
- Authentication and authorisation paths are tested for broken access control, not just assumed correct because a policy exists
- An operator tests retained access from a former-insider posture, and every finding is confirmed by a human before it reaches your queue
An engineer left one of the most security-conscious companies on earth and, Apple says, still walked back into its files through a door nobody shut. Your estate deserves the same look before an ex-employee, or an automated agent, runs the same simple playbook against you. We map the access that survived, re-run it through RTP Robin as your people and systems change, and a human confirms each path before you act. Enterprise-grade. Not enterprise-priced. Start with a free audit.
References
Sources
- TechCrunch. Apple says former employee exploited a rare bug to download confidential files after leaving for OpenAI. 13 July 2026. techcrunch.com
- CNBC. Apple sues OpenAI alleging trade secret theft, says scheme was at every level. 10 July 2026. cnbc.com
- TechCrunch. OpenAI pushes back on Apple trade secret lawsuit. 14 July 2026. techcrunch.com
- Bloomberg. How Apple’s lawsuit threatens to disrupt OpenAI’s bid to rival the iPhone. 13 July 2026. bloomberg.com