Case 3:25-cv-08133-RFL         Document 110        Filed 06/15/26      Page 1 of 7




                                  UNITED STATES DISTRICT COURT

                              NORTHERN DISTRICT OF CALIFORNIA


    X.AI CORP., et al.,                               Case No. 25-cv-08133-RFL
                    Plaintiffs,
                                                      ORDER GRANTING MOTION TO
            v.                                        DISMISS WITHOUT LEAVE TO
                                                      AMEND
    OPENAI, INC., et al.,
                                                      Re: Dkt. No. 91
                    Defendants.

         xAI accuses OpenAI of misappropriating its trade secrets through xAI’s former

employees. On OpenAI’s original motion to dismiss, xAI’s claim under the Defend Trade

Secrets Act (the “DTSA”) was dismissed because xAI failed to sufficiently allege a connection

between OpenAI and the alleged misappropriation of xAI’s former employees. (See Dkt. No. 73

(the “Prior Order”).) xAI has now filed an amended complaint, and OpenAI again moves to

dismiss. (See Dkt. Nos. 91 (the “Motion”), 92-1 (Second Amended Complaint (the “SAC”)).)1

For the reasons set forth below, the Motion is GRANTED WITHOUT LEAVE TO AMEND.

This order does not recount the factual background of the case, which was discussed at length in

the Prior Order, and assumes that the reader is familiar with the facts of the case, the applicable

legal standards, and the parties’ arguments.2

A.       Analysis
         OpenAI moves to dismiss on the basis that xAI insufficiently alleges misappropriation.

xAI counters that it has plausibly pled two theories of misappropriation: (1) OpenAI induced

xAI’s then-employee Li to misappropriate xAI’s trade secrets; and (2) Li disclosed xAI’s trade

secrets to OpenAI in a presentation that he delivered during the recruitment process.



1
    In the SAC, xAI no longer brings a claim under California’s Unfair Competition Law.
2
    All citations to page numbers in filings on the docket refer to ECF page numbers.



                                                  1
       Case 3:25-cv-08133-RFL            Document 110       Filed 06/15/26      Page 2 of 7



          1.     Inducement
          xAI insufficiently pled inducement in the prior complaint because it offered no non-

conclusory allegations allowing a reasonable inference “that OpenAI told or encouraged” xAI’s

former employees to exfiltrate its confidential information. (See Prior Order at 7-9.) To address

that deficiency, xAI now focuses on the alleged interview presentation delivered by Li during his

recruitment process and points to several new allegations: (1) OpenAI specifically targeted Li

for recruitment because of his role as “a senior engineer responsible for the reinforcement

learning and post training techniques for Grok 4,” which were “areas in which OpenAI was

lagging”; (2) OpenAI typically “asks candidates to deliver a presentation about a prior project”

which, for engineers like Li, “typically addresses technical aspects of the project”; (3) “OpenAI

knew or should have known that asking [Li] to present on his prior work was tantamount to

asking Li to discuss xAI’s reinforcement learning and post training techniques”; (4) Li met with

OpenAI’s head of research to understand Li’s professional interests to determine his “best fit” at

OpenAI, and during that conversation, “Li described his work on post-training and reinforcement

learning at xAI”; and (5) as part of the recruiting process, “OpenAI scheduled Li to meet with

key leaders of its post-training and reinforcement learning teams.” (See SAC ¶¶ 63-64, 66-67,

69-70.)

          These allegations do not suffice. In essence, xAI equates asking a candidate about their

prior work experience with encouraging the candidate to divulge trade secrets obtained during
that prior work experience. Without more, however, merely asking Li to discuss his previous

work—a routine part of the hiring process—does not allow a plausible inference that OpenAI

induced Li to reveal anything confidential or secret about that work. To hold otherwise would

potentially expose employers to liability any time they inquire about a candidate’s past work.

          xAI distinguishes between “asking a candidate to deliver a technical presentation about

the proprietary methods used by a competitor” and “asking a candidate about prior work

experience.” (See Dkt. No. 100 at 15.) Yet it alleges only that OpenAI asked Li to present on
“technical aspects” of a project he worked on at xAI, and OpenAI knew or should have known



                                                  2
       Case 3:25-cv-08133-RFL           Document 110        Filed 06/15/26      Page 3 of 7




that Li would have accordingly discussed reinforcement learning and post training techniques in

his presentation. (See SAC ¶¶ 66-67.) xAI does not identify any facts supporting an inference

that presenting on these topics would necessitate the disclosure of trade secrets, and nothing in

the SAC supports xAI’s speculation that OpenAI’s presentation request to Li included a request

that his presentation cover xAI’s “proprietary methods.”

       xAI also argues that OpenAI induced Li to steal xAI’s trade secrets by showing continued

interest in his candidacy even after he revealed xAI trade secrets in his presentation, which

“should have been a red flag against hiring him.” (See Dkt. No. 100 at 15.) Showing continued

interest could potentially demonstrate inducement only if OpenAI knew or should have known

that Li had disclosed xAI’s trade secrets in his presentation. In the prior complaint, “xAI

allege[d] no facts that permit[ted] a reasonable inference that OpenAI knew that the information

[disclosed in the presentation] was a trade secret, that Li had improperly acquired the

information, or that Li was improperly or mistakenly disclosing the information.” (Prior Order at

9.) To address that deficiency, xAI points to the following new allegations about Li’s slide deck

and interview presentation: (1) OpenAI engineers “who were focused on post-training and/or

reinforcement learning” interviewed Li during his presentation; (2) the presentation “focused

specifically on xAI’s reinforcement learning techniques” and Grok 4, and given that Grok 4’s

“strong performance [had been] widely publicized,” the “interviewers would have known the
value of the information that [Li] was communicating”; (3) the first page of the slide deck was

marked “confidential material”; (4) because confidentiality agreements are “prevalen[t]” in the

AI industry, listening to Li’s presentation should have notified the interviewers (who themselves,

on information and belief, had signed similar agreements) that Li “was violating his contractual

confidentiality duties to xAI”; (5) the slide deck included “details of xAI’s training recipes and

checkpoints (slide 2), the sources of problems in post-training AI models (slide 3), xAI’s

experience with different training recipes and which methods provided the best results (slides 5-

7), details of xAI’s experience with different training methods including specific characteristics
of xAI’s internal training (slides 8-10), details of xAI’s training for model behavior (slide 13),


                                                  3
         Case 3:25-cv-08133-RFL           Document 110         Filed 06/15/26      Page 4 of 7




and details of problems identified in xAI’s model training (slide 14)”; and (6) “[a]n xAI engineer

confirmed that th[e] slide deck . . . disclosed multiple xAI trade secrets relating to xAI’s

reinforcement learning approach vis-à-vis Grok 4.” (See SAC ¶¶ 68, 71.)

          These allegations are insufficient to support a reasonable inference that OpenAI knew or

should have known that Li disclosed xAI trade secrets during his presentation. It is not clear at

what level of detail Li discussed xAI’s reinforcement learning techniques. Similarly, while xAI

refrains from alleging that Li actually displayed the slide deck during his presentation (see id. ¶

68), even assuming that he did so, the level of detail included in the slide deck is unclear.3

Accordingly, the allegations are not sufficient to plausibly infer that it was obvious to OpenAI

engineers that the information disclosed was an xAI trade secret. That an xAI engineer

“confirmed” that the slides disclosed xAI trade secrets does not supply the inference that an

OpenAI engineer, based on their industry experience, would know that the slides disclosed xAI

trade secrets.

          Nor is the “confidential” label on the first page of the slide deck a sufficient basis to draw

such an inference. xAI’s argument relies on a multi-step inferential chain. First, xAI asks the

Court to infer that, although it is unknown whether Li used the slides “visually or by verbally

sharing [its] contents” (id.), Li would have either displayed the first page or read the

confidentiality label aloud. Then, xAI proposes a further inference that it would have been
obvious to OpenAI engineers that the word “confidential” referred to “xAI confidential”

material, rather than Li’s promise to OpenAI to keep the recruitment process confidential

pursuant to his nondisclosure agreement with OpenAI. (See id. ¶ 86 (Li “believed it would

violate a nondisclosure agreement to reveal the name of his future employer,” i.e., OpenAI).)

Finally, xAI seeks yet a further inference that it would have been apparent to the OpenAI

engineers interviewing Li that the xAI “confidential” information at issue rose to the level of a

trade secret and was not simply sensitive or internal corporate information. Each of these steps

3
    xAI did not file a copy of the slide deck with the SAC.



                                                    4
       Case 3:25-cv-08133-RFL           Document 110         Filed 06/15/26      Page 5 of 7




involves rejecting an obvious alternative explanation: that the slide deck was used as notes for a

verbal presentation in which the confidentiality label was not read aloud; that the OpenAI

engineers could reasonably have understood any “confidential” label to refer to the nondisclosure

agreement between Li and OpenAI; and that the OpenAI engineers could reasonably have lacked

sufficient information to deduce that any xAI information labeled as “confidential” was a trade

secret as opposed to merely sensitive. “When faced with two possible explanations, only one of

which can be true and only one of which results in liability, plaintiffs cannot offer allegations

that are merely consistent with their favored explanation but are also consistent with the

alternative explanation. Something more is needed, such as facts tending to exclude the

possibility that the alternative explanation is true . . . .” In re Century Aluminum Co. Sec. Litig.,

729 F.3d 1104, 1108 (9th Cir. 2013) (citations and quotation marks omitted). The SAC fails that

requirement because xAI alleges no facts excluding the obvious alternative explanation at each

inferential step.

        Genentech, Inc. v. JHL Biotech, Inc., No. 18-cv-06582-WHA, 2019 WL 1045911 (N.D.

Cal. Mar. 5, 2019), on which xAI relies, presents a useful contrast. There, the defendant

“allegedly received a confidential Genentech Technical Report that was clearly labeled as

‘GENENTECH Pharm R & D Technical Report – CONFIDENTIAL’ and clearly marked

‘Confidential’ and ‘Internal Only’ at the bottom of the cover page” and “was told not to show the
document to others.” See id. at *12 (quotation marks omitted); see also Bal Seal Eng’g, Inc. v.

Nelson Prods., Inc., No. 13-cv-01880-JLS, 2016 WL 11523446, at *1, *3 (C.D. Cal. Oct. 17,

2016) (defendant company allegedly received “drawings and other technical works papers” from

prospective customers with “legend[s] identifying the material as proprietary to” plaintiff

company). Here, however, xAI does not allege that the confidentiality marking on the slide deck

identified the slide deck as belonging to xAI or that Li told his interviewers not to share the

contents of the presentation with others. That the slide deck referred to xAI and Grok “in several

places” does not compel a different outcome (see Dkt. No. 100 at 13-14) because OpenAI
allegedly asked Li, who at the time was employed by xAI, to present on a work project.


                                                  5
       Case 3:25-cv-08133-RFL           Document 110        Filed 06/15/26      Page 6 of 7




       Finally, xAI argues that the Court may reach an adverse inference from Li’s invocation of

his Fifth Amendment privilege in refusing to answer in a separate litigation whether he shared

any xAI trade secrets with OpenAI. (See SAC ¶ 100.) Any adverse inference that Li did share

xAI trade secrets with OpenAI, however, is an insufficient basis to draw a plausible inference of

OpenAI’s knowledge.

       2.      Disclosure by Li
       Disclosure of xAI’s trade secrets to OpenAI by Li could constitute misappropriation only

if OpenAI allegedly knew or had reason to know: “(1) before the use or disclosure that the

information was a trade secret and . . . that [Li] had acquired it through improper means or was

breaching a duty of confidentiality by disclosing it; or (2) that the information was a trade secret

and that the disclosure was a mistake.” (See Prior Order at 9 (citations and quotation marks

omitted).) As discussed above, xAI does not sufficiently allege that OpenAI knew or should

have known that Li disclosed xAI’s trade secrets in his presentation.

       Even if xAI sufficiently alleged OpenAI’s knowledge, disclosure by Li would still not be

enough to state a misappropriation claim because xAI alleges only passive receipt of trade

secrets by OpenAI. Under the DTSA, misappropriation includes acquisition, disclosure, or use.

(See id. at 7.) At most, disclosure by Li could constitute only acquisition by OpenAI and not

disclosure or use. Courts interpret acquisition under the DTSA to require active conduct. The
mere passive receipt of trade secrets is not enough. See, e.g., Bus. Sols. LLC v. Ganatra, No. 18-

cv-01426-DOC, 2020 WL 1279209, at *6 (C.D. Cal. Jan. 22, 2020); see also Silvaco Data Sys. v.

Intel Corp., 184 Cal. App. 4th 210, 223 (2010) (interpreting same statutory language in

California’s trade secret act) (“One does not ordinarily ‘acquire’ a thing inadvertently; the term

implies conduct directed to that objective. The choice of that term over ‘receive’ suggests that

inadvertently coming into possession of a trade secret will not constitute acquisition.”). Indeed,

were passive receipt enough, that would run dangerously close to imposing liability for the mere

possession of trade secrets. Yet “mere possession of trade secrets is not sufficient to constitute
misappropriation.” (Prior Order at 11 (citations and quotation marks omitted).) JustMed, Inc. v.


                                                  6
       Case 3:25-cv-08133-RFL            Document 110         Filed 06/15/26       Page 7 of 7




Byce, 600 F.3d 1118 (9th Cir. 2010), on which xAI relies, does not require a different result.

There, in interpreting acquisition under Idaho’s trade secret act, the Ninth Circuit relied on the

following dictionary definition: “to come into possession, control, or power of disposal of.” Id.

at 1129 (citation omitted). That definition does not address or foreclose the passive-active

distinction that applies under the DTSA.

B.      Leave to Amend
        xAI has failed to cure the deficiencies identified in the Prior Order. Allowing further

amendment would be futile. In moving for a six-month stay of this action to allow time to obtain

discovery in other litigation, xAI argued that it would suffer prejudice in the absence of a stay

because that discovery would allow it to address the deficiencies identified in the Prior Order, so

“xAI should have the opportunity to incorporate such information into its amended complaint.”

(See Dkt. No. 74 at 8.) “But xAI was required to have completed [its] investigation [of its

claims] before filing suit, not after, and if it lacked the requisite information to allege [its] claims

in the manner required . . . when [it] filed suit, [it] should not have sued [OpenAI] in the first

instance.” (Dkt. No. 79 at 2 (citation and quotation marks omitted) (emphasis in original).) xAI

has already been provided multiple opportunities to state a valid claim. Accordingly, dismissal is

without leave to amend. See, e.g., Snapkeys, Ltd. v. Google LLC, No. 19-cv-02658-LHK, 2020

WL 6381354, at *7 (N.D. Cal. Oct. 30, 2020).

C.      Conclusion
        For the foregoing reasons, the Motion is GRANTED WITHOUT LEAVE TO

AMEND.

        IT IS SO ORDERED.

Dated: June 15, 2026


                                                                RITA F. LIN
                                                                United States District Judge




                                                   7