File First, Pay Later: XAI’s Dismissal, Apple’s New Complaint, And The Potential Cost Of Getting It Wrong

Two trade secret complaints against OpenAI, both filed in the Northern District of California, now sit on opposite sides of the line that separates a viable DTSA claim from one that gets dismissed. xAI’s lawsuit against OpenAI was thrown out on June 15, 2026, following Judge Rita Lin’s order granting OpenAI’s motion to dismiss without leave to amend. Apple’s lawsuit against OpenAI, filed July 10, reads as though its drafters had Judge Lin’s order open on the desk beside them. And on July 13, OpenAI filed a motion for more than a million dollars in attorney’s fees against xAI, on the theory that xAI’s dismissed trade secret claim never should have been filed.

Read in sequence, the three filings are a lesson in what the DTSA requires a plaintiff to plead in a talent-raiding case, and what it costs to guess wrong.

xAI’s pleading failures

xAI’s theory narrowed, by the end, to a single event: former engineer Xuechen Li allegedly disclosed xAI’s reinforcement learning and post-training techniques for Grok 4 during an OpenAI recruiting presentation. Judge Lin found the theory deficient twice over.

First, on knowledge. OpenAI’s continued interest in Li could show inducement only if OpenAI knew or had reason to know he had disclosed trade secrets. xAI built that knowledge allegation on an inferential chain: that Li displayed or read aloud a slide marked “confidential”; that OpenAI’s engineers understood “confidential” to mean xAI confidential rather than Li’s own nondisclosure obligation to OpenAI about the recruitment process; and that they further understood the marked material to be a trade secret rather than merely sensitive. Each link had an innocent explanation. Under In re Century Aluminum Co. Securities Litigation, 729 F.3d 1104, 1108 (9th Cir. 2013), a plaintiff facing an equally plausible innocent explanation must plead facts tending to exclude it. xAI pleaded none.

Second, and with the longer reach, on the nature of the conduct. Even assuming disclosure, xAI alleged only that OpenAI received what Li presented. Under the DTSA, misappropriation requires acquisition, disclosure, or use; passive receipt is at most acquisition, and the court held that acquisition requires active conduct to be actionable. The reasoning tracks Silvaco Data Systems v. Intel Corp., 184 Cal. App. 4th 210, 223 (2010): one does not “acquire” something inadvertently, and treating passive receipt as acquisition edges toward liability for mere possession, which is not misappropriation. See also Bus. Sols. LLC v. Ganatra, No. SACV181426DOCKESX, 2020 WL 1279209, at *6 (C.D. Cal. Jan. 22, 2020) (“a person acquires a trade secret when they actively seek to gain ownership or control of the trade secret, but not when they are merely a passive recipient.”). A candidate who volunteers a former employer’s secrets creates exposure for himself; he does not, without more, create acquisition liability for the new employer.

How Apple overcame those obstacles

Apple’s complaint is captioned Apple Inc. v. Chang Liu, Tang Yew Tan, OpenAI Foundation, OpenAI Group PBC, and io Products, LLC, No. 5:26-cv-07078, assigned to Magistrate Judge Virginia K. DeMarchi. It pleads four separate DTSA claims—one each against Liu, Tan, OpenAI, and io—plus two breach-of-contract claims against the individuals under Apple’s Intellectual Property Agreement (there is no claim under the California Uniform Trade Secrets Act, keeping the trade secret theory entirely federal).

Where xAI alleged passive receipt, Apple alleges active acquisition. Liu, a former Apple electrical engineer, allegedly failed to return an Apple laptop authenticated to the company network, exploited an authentication bug to reach Apple’s shared network folders after his departure, and downloaded dozens of confidential files while working for OpenAI—celebrating the access to a still-employed colleague (“LOL, I found out I can access the [network storage], so funny”) and coaching her to copy files so as to “avoid trouble with the security team.” That is acquisition by unauthorized access, not inadvertent receipt.

Where xAI could not plead that OpenAI knew what it was getting, Apple pleads documented knowledge. Its scienter centerpiece is an internal Apple managers’ document marked “Need to Know,” describing the company’s security procedures for departing employees, which Chief Hardware Officer Tang Tan and his OpenAI colleagues allegedly circulated to recruits before they gave notice, previewing Apple’s forensic checks. The complaint uses that document to close the knowledge gap directly: Tan’s “possession and distribution of Apple’s internal departure procedures—which expressly reference Apple’s IPA obligations—foreclose any defense of ignorance” (¶ 107), with a parallel allegation against OpenAI (¶ 120).

And where xAI alleged only a routine interview presentation, Apple alleges an interview process that was designed to extract trade secrets. Tan allegedly used Apple’s internal project codenames to ask candidates “What’s the plan[?]” for unreleased products, and directed them to bring “Actual parts”—batteries, systems-in-package, logic boards—along with “CAD/design artifacts” and “prototypes” for “show and tell” (¶¶ 69–70, 75). One candidate reportedly noted he “didn’t even know we could take those from the office.” This is the “something more” that Century Aluminum demands: an alleged solicitation of identified, physical confidential material. Apple also pleads active use through a supplier—OpenAI, on its own or through io, allegedly had a trusted Apple partner perform Apple’s proprietary metal-finishing technique, “misleading the partner to believe they had Apple’s permission to do so” (¶¶ 12, 80–81)—and an alter-ego theory attributing io’s conduct to OpenAI (¶¶29–32).

None of this means Apple wins. Its allegations are still just that – allegations – and OpenAI’s strongest response sits in Apple’s own litigation history: Hooked Media Group, Inc. v. Apple Inc., 55 Cal. App. 5th 323 (2020), where Apple prevailed as the defendant on the principle that hiring an employee who retains a competitor’s knowledge is not, without an independent wrongful act, misappropriation. See id. at 331 (“Allowing an action for trade secret misappropriation against a former employee for using his or her own knowledge to benefit a new employer is impermissible because it would be equivalent to retroactively imposing on the employee a covenant not to compete.”) (emphasis in original). But Apple’s complaint contains a catalogue of the affirmative conduct—unauthorized access, documented knowledge, directed exfiltration, downstream use—whose absence defined the failure of xAI’s lawsuit.

OpenAI’s motion for attorney’s fees

On July 13, OpenAI moved for $1,041,859.90 in attorney’s fees against xAI under the DTSA’s fee-shifting provision, 18 U.S.C. § 1836(b)(3)(D), and CUTSA’s parallel provision, Cal. Civ. Code § 3426.4. The standard is the two-prong California test for a bad-faith trade secret claim: objective speciousness and subjective bad faith.

On objective speciousness, the motion quotes Judge Lin’s holding back at xAI—that its allegations “at most amounted to ‘passive receipt of trade secrets,’ which is ‘not enough.’” What was a ruling in June has now become the predicate for OpenAI’s requested fee award. On subjective bad faith, the motion leans on the pre-suit timing of OpenAI’s interactions with xAI. OpenAI says it wrote to xAI before suit, explained the absence of evidence, and asked for any factual basis; rather than engaging, xAI sued instead. OpenAI also notes that xAI’s CEO Elon Musk tweeted about the lawsuit before it had been served or a copy provided to OpenAI’s counsel, “illustrating that the public message was more important than the actual litigation.” After the first dismissal, xAI moved for a six-month stay to hunt for the evidence it lacked; a request that was denied, with the court noting that “a plaintiff must investigate the facts underlying their allegations before filing suit, not after.” xAI then filed a Second Amended Complaint reasserting the rejected theory, which was dismissed a second time.

OpenAI represents that its request for $1,041,859.90 is a deliberately conservative slice of its actual spend, excluding several timekeepers and its entire pre-litigation investigation, which includes partner rates up to $1,573 an hour. Whatever the court awards, the motion demonstrates how a failed misappropriation claim can transform into a price tag.

Important takeaways for talent competition

These new developments arrive at a time when several factors are converging to incentivize filing a trade secret lawsuit. Federal trade secret filings hit an all-time high in 2025—1,552 new cases, up roughly 20% over 2024 and the most since the DTSA took effect in 2016, according to Lex Machina’s 2026 Trade Secret Litigation Report. California is the center of gravity: the Central District of California was the single most active trade secret federal district in 2025, with 100 new filings, and AI-related trade secret cases have climbed sharply as the talent wars intensify.

Part of the pressure is due to California’s other legal frameworks. Most importantly, California voids employee noncompetes under Business and Professions Code § 16600, a ban the Legislature strengthened effective January 1, 2024 through SB 699 and AB 1076—adding a private right of action, reaching agreements signed out of state, and requiring employers to notify employees that their noncompete clauses are void. When a company cannot stop an engineer from walking across the street to a competitor, trade secret law is one of the few levers left. It is widely assumed that the noncompete ban therefore drives California’s trade secret docket, though the empirical support is thinner than the assumption suggests; the one study to test it directly found the apparent correlation fades once state population is taken into account (Giri, Dang & McKiernan, ABA Landslide, 2024). However, as an anecdotal observation, it does appear that employers who have lost the noncompete feel a gravitational pull toward the courthouse, when they feel their business interests are threatened. And this gravitational pull is strongest when the battle for top talent in the AI and tech industries is more ferocious than ever.

Nonetheless, the absence of other remedies is not an invitation to file first and investigate later, as Judge Lin made abundantly clear in her order dismissing the xAI lawsuit. The same conditions that make trade secret litigation a powerful tool in California—constant lateral movement, high-value technical knowledge, and no noncompetes—also make it tempting to move quickly to litigation. xAI and Apple both sued OpenAI on very similar talent-raiding theories. xAI pleaded receipt and hoped discovery would supply the rest; it ended up with a dismissal and a seven-figure fee demand. Apple, by contrast, pleaded the specific alleged conduct. The DTSA rewards the difference, and through its fee-shifting provision, it charges for the failure to observe it.