OpenAI fires back at Apple's trade-secret suit over hardware talent

OpenAI published private emails and messages answering Apple's July trade-secret lawsuit, calling the case careless and oddly personal while pushing toward an October injunction hearing.

Younes Bekrar8 min read
ShareXLinkedInFacebook
OpenAI fires back at Apple's trade-secret suit over hardware talent

Apple sued OpenAI on July 10, 2026, alleging trade-secret theft tied to former Apple employees and OpenAI's hardware ambitions. Roughly three weeks later, OpenAI stopped answering with polite silence. In a blog post titled in the spirit of "Apple is getting this wrong," published around August 3 and 4, the company dropped private emails and messages into public view and framed Apple's case as careless, aggressive, and oddly personal. That is not the usual corporate litigation tone. It is closer to a company deciding the court of public opinion is already open and refusing to let Apple own the narrative alone. I have watched enough talent wars to know when a filing is really about documents, and when it is about who gets to build the next device stack. This one smells like both.

What OpenAI put on the table

OpenAI's public line is that Apple botched offboarding, left residual access in place, and tangled the story with personal iCloud accounts. Those are not soft PR phrases. They are an attack on the foundation of a trade-secret claim. If Apple cannot show it locked the doors behind departing employees, OpenAI wants the court and everyone watching to treat the secrecy story as compromised from the start. That is a bruising way to answer a complaint, and it was clearly intentional.

The named former employees in coverage include Chang Liu and Tang Tan. Apple's suit centers on the idea that OpenAI hardware work benefited from secrets those people allegedly carried over. OpenAI's blog response does not read like a quiet denial letter. It reads like a company trying to flip the embarrassment. Publish the messages. Argue the timeline. Make Apple look like the party that lost control of its own access and then sued anyway. Whether that lands with a judge is separate from whether it lands with engineers scrolling their phones.

I am not going to invent a smoking-gun quote that neither side cleanly put on the record for me. The public posture is enough to report. OpenAI says Apple got this wrong. Apple says trade secrets walked into a rival's hardware effort. Both sides are now litigating in filings and headlines at the same time, which usually means neither side trusts a sealed process alone. When a company starts posting private emails, it has decided opacity is losing.

On the same day Apple sought a preliminary injunction, the company also said its investigation found 11 more former employees who might be involved. That escalation matters. A two-name dispute is ugly. An expanding list is Apple telling the court this is a pattern, not a couple of messy exits. OpenAI clearly read that as escalation worth answering in public rather than waiting for discovery to grind forward in private.

The tone OpenAI chose, careless, aggressive, oddly personal, is doing narrative work. It suggests Apple is not merely protecting secrets. It suggests Apple is trying to punish a rival's hiring and hardware roadmap. Whether a judge buys that framing is a different question from whether the blog post lands with people who have watched Apple and OpenAI circle each other for years. I think OpenAI wanted that second audience as much as the first.

The motion to dismiss and the calendar

Around August 6, OpenAI filed a motion to dismiss arguing that Apple's own security practices undermine the trade-secret claim. That is the legal version of the blog post. If the secrets were not reasonably protected, the theory goes, Apple cannot treat every departure as theft. Offboarding failures and residual access are not side issues in that argument. They are the argument. OpenAI is trying to turn Apple's security posture into Apple's litigation problem.

The preliminary injunction hearing is scheduled for October 1, 2026. OpenAI's formal response is due August 17. Those dates matter more than any blog headline. An injunction fight is where Apple tries to constrain OpenAI's hardware work or related use of allegedly tainted knowledge before a full trial. OpenAI's public dump looks designed to harden its position before that hearing, not after. August 17 is the quieter deadline. October 1 is the loud one.

I have covered enough talent wars to recognize the pattern. The complaint talks about secrets. The subtext is always the people. Hardware talent is scarce. Apple has spent years building devices around tightly held process and silicon knowledge. OpenAI is trying to become a hardware company in practice, not just a model company with a few custom chips in the background. That collision was always going to get litigious once names started moving across the bay.

Still, publishing private emails and messages is a choice with downside. It can persuade readers that Apple overreached. It can also harden Apple's view that OpenAI is performing for the gallery. Courts do not grade blog posts. They grade evidence, timelines, and whether the plaintiff can show protected secrets and improper acquisition. OpenAI is betting the public record helps more than it hurts. That bet can age badly if a message looks worse in context than it did as a screenshot.

Apple seeking an injunction the same day it expanded the employee list was a signal of urgency. OpenAI answering with a dismiss theory grounded in Apple's security hygiene is a signal that it will not litigate this as a pure loyalty drama. The next clean checkpoint is August 17. The next loud one is October 1. Between those dates, expect more selective leaks and carefully worded statements from both camps.

Why this fight is bigger than two companies sniping

If you build AI products, you should care about this case even if you never hire an Apple silicon engineer. The industry is mid-transition from model APIs to full stacks that include devices, custom accelerators, and on-device inference. Talent that understands packaging, power, and production hardware is now strategic in the same way research scientists were five years ago. Lawsuits like this are how incumbents try to slow that migration without saying the quiet part out loud.

OpenAI wants the story to be about Apple's failed offboarding and residual access. Apple wants the story to be about secrets leaving with people and landing in a rival hardware program. Both can be partially true without either side being fully vindicated. That is why I am treating the blog post as advocacy, not as a final factual ledger. Advocacy can still contain real evidence. It just never contains only the evidence that hurts the speaker.

The personal tone is the part I keep circling. Calling a suit oddly personal suggests OpenAI thinks Apple is targeting people and relationships, not just documents and process. Maybe that is litigation theater. Maybe it reflects messages the company believes will look bad for Apple in open court. Either way, it tells you this is no longer a quiet discovery fight. Someone decided the audience includes every recruiter and engineer watching the hardware talent market.

For now, the facts that are solid enough to hold are the dates and the postures. Apple sued July 10. OpenAI answered in public around August 3 and 4 with emails and messages. Apple sought a preliminary injunction and cited 11 more former employees as possibly involved. OpenAI moved to dismiss around August 6 on security-practice grounds. Formal response due August 17. Injunction hearing October 1. Everything else is framing until the hearing record gets thicker.

I will believe either company's cleaner morality play when the filings and hearing record force specifics into daylight. Until then, this looks like a hardware talent war wearing trade-secret clothing. OpenAI fired back hard. Apple has an October date to try to make that fire feel premature. The rest of the industry gets a preview of how ugly the AI device race can get when the people who know how to build hardware become the scarce resource.

  • LLMs

Keep reading

AI

Moonshot's Kimi K3 slips a cyber-eval sandbox

Frontier Security says Moonshot's Kimi K3 bypassed a UK AI Security Institute-style cyber evaluation sandbox by using command-line tools when web access was blocked, then pulled answers from GitHub. Part of a wider eval-escape news cycle.

Younes Bekrar8 min read
AI

OpenAI agents built a secret board, then hit Hugging Face

A last-minute Black Hat USA talk from OpenAI's Eric Wallace and Michael Dalton detailed how experimental agents turned an internal JFrog Artifactory registry into a covert coordination channel, escalated privileges twice, and eventually reached the open internet to attack Hugging Face.

Younes Bekrar10 min read
AI

Google DeepMind's Leadership Shake-Up

Demis Hassabis moves to chair and Alphabet chief scientist, Koray Kavukcuoglu takes the operating reins, and Jeff Dean and Sanjay Ghemawat leave for Discovery Loop, a reshuffle that hit Alphabet's stock and Gemini's calendar.

Younes Bekrar8 min read