The linchpin of his efforts have revolved around “full self-driving” (FSD), a feature that remains in beta after three years, 11 different iterations, and over half-a-billion miles of real-world testing.

Yet despite repeated delays in validating the software, eventual success would mean Musk finally possesses the underlying technology he believes he needs to make Tesla the most valuable company in the world.

Those plans are now potentially in jeopardy, after Tesla confirmed in a quarterly filing to the Securities and Exchange Commission published on Monday that its controversial automated driving software is the subject of a U.S. Justice Department probe. 

“The company has received requests for information, including subpoenas, from the DoJ. These have included requests for documents related to Tesla’s Autopilot and FSD features,” it said.

Previously the wording in its various 10-Q risk reports have referred only to requests, never mentioning an official subpoena. 

Tesla added that no government agency in any ongoing investigation has concluded that any wrongdoing occurred to the best of its knowledge, before including the common legal disclosure: “Should the government decide to pursue an enforcement action, there exists the possibility of a material adverse impact on our business.”

Tesla did not disclose the nature of the information sought by the U.S. Justice Department. Neither party could be reached by Fortune for further comment.

Autopilot is the most basic form of the software and comes standard on all Tesla vehicles. FSD is currently a $12,000 optional extra, following a $3,000 price cut. 

Under Musk, Tesla has pursued a technological path to autonomy that is unique in the industry. Whereas others such as GM’s Cruise and Alphabet’s Waymo are cautiously experimenting with robo-taxis confined to areas—often meticulously mapped to the inch and equipped with expensive laser scanners—Musk is attempting to use a sophisticated onboard computer fed with camera data only.

The idea behind this approach is that Tesla would enjoy unmatched scalability, rolling the technology out far more rapidly than any competitor, should it prove reliable. 

Billion-dollar investment on top of buying Nvidia AI chips

Instead of instructing his engineering team around Ashok Elluswamy to finish developing the software before launching it commercially, Musk made it available to paying customers. However, it comes under the stipulation that customers know they are being asked to validate the technology.

Musk is already way behind his initial ambitious schedule that foresaw 1 million autonomous Teslas on the road by the end of 2020, from his existing stable of “S3XY” models. This has led to some angry buyers, like Apple cofounder Steve Wozniak, who has accused Musk of being dishonest. 

Last week the CEO nonetheless said he remained as convinced as ever that AI has “the potential to make Tesla the most valuable company in the world by far,” after Tesla completed assembly of a 10,000 GPU cluster powered with Nvidia’s latest H100 processors used for neural net training.

This comes on top of its additional $1 billion investment into its Dojo supercomputer, which utilizes custom-designed silicon for faster AI training of visual data. Musk has said he could license out its use to other companies, and believes it may one day be worth hundreds of billions of dollars alone.

Shares in Tesla traded 1.5% higher on Monday, recouping some losses after suffering their worst trading week so far this year on the back of a grim investor call and weaker-than-expected Q3 sales and profits.