Today I Learned · Forensic Audit
Apple v. Liu et al. · 5:26-CV-07078 · N.D. Cal. · filed 07.10.26

The Apple–OpenAI Fault Line

Six months of signal, one lawsuit, read against the actual complaint, not the coverage of it. Twelve pieces written before the complaint existed, then checked against the filed document. This is the calibration system, run in public.
score before → check after → correct in public
Why this page exists

The value in a forecast is not the call, it is the scored, checkable call behind it. So this analysis carries a belief ledger: every major claim was scored when it was first written, then re-scored against the filed complaint, with the evidence that moved each number, and the claims that did not survive kept on the record. That is the same score-before, check-after discipline the ARCS calibration system runs on, applied to a real case.

The one claim, held

Apple was never racing to win the model. It was defending the layer the models plug into. Apple owns the OS, the silicon, the sensors, and the manufacturing IP; it rents out the model slot. A partner that tries to win the rented slot and take the owned layer stops being a partner. The design lineage that left Apple (2019–2022) and reconvened at io (2024–2025, acquired by OpenAI for $6.5B) is not backstory to that thesis; it is the same thesis, five years early, before anyone had a reason to write it down. On July 10, 2026 the partnership did not end. It bifurcated: ChatGPT-in-Apple-Intelligence continues, carved out of the suit; the hardware fight is now in federal court.

The belief ledger: what moved, and what got struck

Scored before the filing · 3 claims struck on the record · 0 retrofitted

Every score is a conviction from 0 to 100, tracked from when the claim was first made (T0) to after the complaint was read (NOW). The honest version of “I called this” includes the claims that did not survive contact with the filing.

Q.ai is a hardware bet, not chatbot panic
T0 (Jan 30): 55 · speculative
NOW: 85 · patent claims + Tan’s title confirm it
io’s founders are the Apple design lineage
T0: 70 · “Tan and others,” vague
NOW: 85 · 4 named; Hankey = Ive’s own successor
Ive’s Apple exit terms
T0: unverified “4-year noncompete”
NOW: 90 · 3-yr paid consulting exclusivity, not a legal noncompete
Apple defends the owned layer, rents the model
T0 (Feb onward): 85
NOW: 90 · held and sharpened by the complaint’s own hardware/software split
Titan cancellation was internal decision-drift
T0 (Feb): 75 · internal framing
NOW: 30 for that framing · 80 for the corrected version (external talent loss)
The lawsuit is retaliation for the Gemini/PCC snub
First draft: 60 · asserted as cause
NOW: 5 · struck; the complaint (fn.13) disclaims the connection
io built “in the shadow of” Humane’s failure
Draft: 50 · broad causal claim
NOW: 10 for the broad claim · 60 for a narrower one
Supplier overlap proves intent
Appendix pass: not yet scored
NOW: 95 the fact is real · 30 for reading it as proof of intent

How this was checked

  • Read against the filing. Every claim checked against the actual complaint, not summaries of it.
  • Stress-tested. Read for where a claim was framed more confidently than the evidence earned.
  • Argued from every side. The strongest version of Apple’s case, OpenAI’s defense, and a neutral read, built separately.
  • Pressure-tested. Found the single fact that, if wrong, breaks the argument, then checked it.
  • The cynical read. The facts held. The story wrapped around them is what got corrected.

The corpus: twelve pieces, one thread

Every piece this audit draws from was published on LinkedIn in the ordinary course, before the complaint existed. The thirteenth and final primary source is the filed complaint itself.

Disclosure

Independent analysis. No compensation, access, or coordination from any named party. No position in AAPL or OpenAI-linked securities. Every allegation is exactly that, an allegation from a filed complaint, not an adjudicated fact. Not investment advice, not legal advice. Where an inference is drawn rather than a fact reported, it is marked as such.