xAI accuses former employee Li Xuechen (transliterated) of stealing Grok 4 source code and switching to OpenAI. The case has entered judicial proceedings and may impact the March trial between Musk and OpenAI.
(Background summary: V神’s first review of LLM: Grok essentially saves the X platform “aids in truth dissemination,” but still has many hallucinations)
(Additional background: Musk announced that Grok 5 will challenge the strongest human team in LOL next year: if successful, it will prove a substantial breakthrough in AGI)

Table of Contents

• Current judicial progress of the case
  • FBI involvement and search
  • Career freeze (injunction)
• Netizen comments

In September 2025, former xAI senior engineer Li Xuechen was widely criticized online by Musk for allegedly downloading Grok 4 training algorithms and source code without authorization in mid-July, and attempting to delete server logs to conceal his actions.

At that time, xAI filed a civil lawsuit in California, with a maximum claim amount potentially reaching $1 billion. This case is viewed as the first commercial secret dispute in Silicon Valley under the generative AI arms race, possibly treated as a federal felony.

xAI accuses Li Xuechen, just before his resignation in July 2025, of abusing his position to:

• Steal core code: downloaded unreleased Grok 4 training algorithms, weights, and nearly “the entire xAI codebase.”
• Destroy evidence: after transferring data to personal devices, attempted to hide traces by deleting server logs, clearing browser history, and renaming/compressing files.
• Cash out: sold xAI stock worth about $7 million just before leaving for competitor OpenAI.

OpenAI offered him a salary of $3 million per year plus a signing bonus of $5 million. Prosecutors believe this high compensation was directly linked to Li Xuechen obtaining and carrying xAI confidential code.

The indictment shows that Li Xuechen, leveraging his high-level permissions as an early employee, packed and downloaded Grok 4 files late at night on July 25, 2025, and tried to erase access records. However, system backups and multi-layer logs still retained key timestamps, serving as primary evidence for xAI.

TBS News cited judicial documents indicating these files included model architecture and inference optimization details, classified as high-level trade secrets “directly affecting national competitiveness.”

Current judicial progress of the case

The focus of this case has shifted from civil damages between xAI and Li Xuechen to a much more serious situation involving federal law enforcement.

FBI involvement and search

According to court documents, the FBI has launched a criminal investigation into Li Xuechen. Search warrants were executed at his residence, vehicle, and hotel room, seizing his phone, computer, hard drives, and notebooks.

This criminal risk means he faces not only a huge civil liability to xAI but also potential federal charges under the Economic Espionage Act, similar to the case of former Google engineer Linwei Ding.

Facing criminal risks, Li Xuechen has adopted a defensive strategy in the civil lawsuit: his legal team advised him to exercise his Fifth Amendment rights during discovery, refusing to answer certain questions to avoid self-incrimination in the criminal case.

xAI attempted to force him to disclose all device passwords and cloud account access, but Li believed this would lead to self-incrimination, leading to intense procedural disputes in court.

Career freeze (injunction)

The court previously approved a temporary restraining order (TRO), which largely maintained a tough stance after a hearing at the end of 2025 (originally scheduled for December 2). This prohibits Li Xuechen from working on generative AI-related tasks for OpenAI or any other competitors until the case is resolved. So, OpenAI did not actually employ him.

The most recent and significant hearing is scheduled for January 20 (Tuesday) — the Preliminary Injunction Hearing.

This is a highly decisive hearing. The judge will determine whether the current TRO against Li Xuechen will be converted into a longer-lasting, broader preliminary injunction.

The court will decide whether to continue a comprehensive ban on Li Xuechen working on generative AI at OpenAI or other xAI competitors before the final trial. The TRO has been extended until that evening at 9:00 PM; if the judge rules to extend it after the hearing, Li Xuechen will face a long-term “career freeze.”

On the other hand, rumors in the US industry suggest that Li Xuechen has been blacklisted by major AI companies, including OpenAI, Google, and Meta, as no AI giant is willing to risk employee involvement in theft.

Netizen comments

This case has sparked analysis from Chinese netizen Eugeni on Zhihu:

Actually, I still have some doubts about whether he took all the code and how he did it. Let’s use some limited information to speculate and see if we can uncover something we don’t know.

1. Li Xuechen: PhD from Stanford, undergraduate from Beijing Zhongguancun Middle School, attended Beijing No. 4 Middle School, won second place in the 2013 National High School Physics Competition, so he was probably in his second or third year of high school in 2013, and should be under 30 years old. Joined xAI in 2024.

2. Before resigning in July 2025 (when the company’s valuation was about $40), he cashed out $7 million worth of xAI stock. Based on the common Silicon Valley 4-year vesting with a 1-year cliff, he would get 25% after one year.

According to this, his equity package should be less than 800,000 shares. There are some puzzling aspects here: possibilities include that he joined before the Series B round in early 2024.

His equity could have multiplied significantly, meaning his initial stake might have been around $2 million+. Such a package would typically correspond to a Director or Senior Engineering Manager level. If he joined after the Series B in 2024, the package would be large enough to be at VP level, which doesn’t quite match his age and low profile in the industry. This guess is verified by Li Xuechen’s personal info on X, indicating he indeed joined before the Series B in early 2024.

3. Joined on August 19: he had not yet joined when sued, so theoretically he hadn’t provided code to OpenAI.

4. Behaviors like deleting logs and renaming files are unlikely to be about deleting server logs; renaming files is very basic and looks more like defensive behavior—perhaps he was reluctant to give up the code he worked on at xAI but was also worried about non-compete clauses.

5. Musk’s demands are compensation plus an injunction. If you were Musk, and all the code was taken and given to a competitor, what would you do? Would you be so lenient? (Of course, the compensation amount could be huge.) So it looks more like a show of force, but Musk doesn’t want to go all the way.

In summary, I think: an excellent AI talent, for personal reasons, wanted to switch jobs and couldn’t bear to part with his code, so he downloaded it to take away. He knew it was wrong, so he used some basic concealment methods. Maybe colleagues did the same before, but since they were old friends or acquaintances, the company didn’t scrutinize much.

Recently, Musk and Ohtman have been fighting fiercely, and with several companies aggressively recruiting, Musk has targeted this case for a big show. The exaggerated claims about the entire codebase, plus his Chinese identity, caught media attention.

Musk has solid evidence and can use it freely, so OpenAI remains silent. But Li Xuechen has basically become a scapegoat. Of course, his actions are not justified or correct, but the punishment seems disproportionate to the offense.

(Generally, Li Xuechen’s career in the industry abroad is probably over.)