Apple lawyers up as OpenAI raids its AI bench

What happened

According to the Financial Times, Apple has sent legal letters to dozens of former employees who left for OpenAI, reminding them of their confidentiality obligations and, in some cases, non-solicit clauses that bar them from recruiting ex-colleagues. The letters are not lawsuits — yet — but they are the standard opening move of a company preparing to litigate if a former employee walks out with proprietary information or starts poaching their old team.

The departures have been steady and pointed. OpenAI has hired at least two dozen Apple researchers and engineers over the past year, including senior figures from the Apple Foundation Models group, the team behind the on-device LLM that powers Apple Intelligence. Ruoming Pang, who led that group, decamped to Meta's Superintelligence Labs earlier this summer for a package reportedly worth over $200M. The OpenAI-bound cohort is smaller per-head but broader, and it's the breadth — not any one name — that seems to have finally triggered Apple's legal department.

Apple's response — form letters from outside counsel rather than headline-grabbing lawsuits — is the corporate equivalent of leaving the porch light on: a signal that the house is being watched, without committing to a fight. The letters cite California's narrow but real carve-outs to its otherwise employee-friendly non-compete regime: trade secrets, confidential information, and non-solicit-of-employees clauses that most courts still enforce even when non-competes don't survive.

Why it matters

For the last decade, the story of Silicon Valley talent wars was Google versus everyone. Apple was a net importer, famously insular, and — until Apple Intelligence shipped in a limping half-state at WWDC 2024 — content to be a fast follower on AI. That posture is no longer viable, and the letters are the clearest signal yet that Cupertino knows it.

OpenAI is now the gravitational center for anyone who wants to work on frontier models with real deployment surface area, and Apple is discovering what Google discovered in 2022: you cannot retain researchers with vesting schedules alone when the shop across the street is handing out liquid equity at a $500B valuation. The Foundation Models team is a particularly painful loss because it's small, cohesive, and expensive to rebuild. On-device model work requires a specific blend of ML research and systems engineering that doesn't grow on trees, and every departure compounds — the next person out the door has one fewer reason to stay.

The legal angle is worth reading carefully. California Business and Professions Code §16600 voids most non-competes, which is why the letters lean on confidentiality and non-solicit-of-employees language instead. Non-solicits are still enforceable in California when narrowly drawn, and trade-secret claims under the Defend Trade Secrets Act give employers a federal cause of action with real teeth — see Waymo v. Uber, which ended in an $245M settlement and a criminal referral. Apple isn't threatening that outcome, but the letters put every recipient on notice that any recruiting DM to a former colleague, any reference to internal architectures, any reuse of pipeline code from memory, is now discoverable.

The community reaction on Hacker News has been predictably split: half calling this SOP for any large tech company defending its IP, half noting that Apple is famously aggressive about confidentiality even by FAANG standards, with alumni reporting years-long chill effects from vaguely-worded reminder letters. The second camp has a point. A form letter costs Apple nothing to send and imposes real friction on the recipient — most will forward it to their new employer's legal team, some will lawyer up personally, and a nonzero fraction will decline to recruit former colleagues even if the clause wouldn't actually hold up. The chilling effect is the product.

What this means for your stack

If you're an engineer weighing a jump from a big-cap incumbent to a frontier AI lab, the ground has shifted. Assume every clause in your offer letter and separation agreement is now live ammunition, not boilerplate: get a plaintext copy before you sign anything, have an employment lawyer read your existing NDA before you accept the new offer, and do not — under any circumstance — reference internal architectures, training data, or headcount from your old employer in any written channel at the new one. The cost of a two-hour consult with a Bay Area employment attorney is $500-$1000. The cost of getting named in a trade-secrets suit is your entire signing bonus and roughly eighteen months of your life.

For engineering managers on the hiring side, the operational implication is that your onboarding for anyone coming from Apple, Meta, or Google needs a formal clean-room protocol, not just a verbal reminder. That means documented no-touch periods on projects that overlap with the new hire's old scope, written confirmation they've handed back all devices and credentials, and — critically — a policy that they do not recruit from their former team for the duration of any applicable non-solicit. The alternative is discovery, and discovery is where clean intentions go to die.

For everyone else, this is a leading indicator worth tracking. The frequency and aggression of these letters correlates with how much the incumbent thinks it's losing. Apple sending dozens of them in a single wave is a tell that internal metrics on AI velocity are worse than the WWDC keynote suggested, and that senior leadership sees the talent gap — not the compute gap or the data gap — as the binding constraint on catching up.

Looking ahead

Expect this to escalate before it de-escalates. If OpenAI ships something in the next six months that visibly borrows from on-device optimization techniques Apple pioneered — quantization tricks, KV-cache management, mixture-of-experts routing on constrained hardware — a real lawsuit becomes plausible. If not, the letters will do their quieter work: slowing the next wave of departures by a quarter or two while Apple tries to fix the retention problem with money, org changes, or a genuinely competitive model roadmap. Whichever it is, the era of Apple treating its ML talent as a permanent asset is over.