Rarely in the technology industry does a court declaration reveal so much at once. That is precisely what just happened after Elon Musk acknowledged before a federal court in California that xAI used man-made intelligence models from OpenAI to train the first versions of Grokits star chatbot.
The most curious thing is that this admission did not come from a leak or a journalistic investigation, but from Musk’s own mouth, during the trial that he himself initiated against Sam Altman and OpenAI.
The scene is hard to ignore. The man who presents himself as one of OpenAI’s harshest critics, who accuses Altman of having betrayed the company’s new mission By converting it into a for-profit entity, he ended up confessing that his own company took advantage of its rival’s models to build Grok from scratch. The irony could not be more evident.
The technique that everyone uses but no one confesses
During his testimony, one of the lawyers asked Musk directly if xAI had applied the distillation technique on OpenAI models to train Grok. Musk first responded that this is a widespread practice throughout the industry, avoiding the focus of the question. But when they insisted with an uncomplicated “does that mean yes?”, he responded with a single word: “Partially.”
Distillation, for those unfamiliar with the term, is basically a process by which a smaller AI model learns to imitate the behavior of a larger one. The “student” model absorbs the response patterns of the “teacher” model, reproducing its predictions and capabilities without necessarily accessing its direct training data. In other words, you can build a competent AI by spying on how another AI answers a lot of questions.
The problem is that companies like OpenAI and Anthropic have been denouncing this practice for some time, especially when it comes from Chinese companies that create open source models almost as capable as theirs, but at a fraction of the cost. What no one had openly confirmed until now is that the American laboratories themselves also apply this technique to each other. Musk just broke that silence without anyone forcing him.
A lawsuit against OpenAI that ends up exposing xAI
The trial in which this revelation occurred is the one that Musk filed in 2024 against OpenAI, Sam Altman and Greg Brockman, accusing them of diverting the founding mission of the organization, which was originally born as a non-profit entity committed to developing AI for the benefit of humanity. According to Musk, Altman and his partners betrayed him by seeking financial benefits behind his back, something that Altman himself categorically denies.
The case will last approximately a month and includes testimonies from relevant figures in the tech world, including Satya Nadella, CEO of Microsoft. But Musk’s statement already stole the headlines before the trial reaches its most intense moments. Because basically the plaintiff ended up giving the defendant ammunition without anyone asking him to.
And if that were not enough, later in his testimony, Musk was asked about a statement he made last summer, in which he boasted that xAI would soon surpass any company in the world except Google. His response was surprising: he placed Anthropic in first place in the man made intelligence ratingfollowed by OpenAI, Google and Chinese open source models, and described xAI as a much smaller company, with just a few hundred employees.
Where is the line between learning and copying?
This is the question that remains floating after this entire episode. Musk argued that using other AIs to validate your own is common practice in the industryand you are technically right that distillation is not explicitly made illegal in most jurisdictions. What can happen is that it violates the platforms’ terms of service, which generally prohibit the systematic use of their APIs and chatbots to train competing models.
The Frontier Model Dialogue board, an initiative involving OpenAI, Anthropic and Google, is already working on mechanisms to detect and stop mass distillation attempts, which generally manifest as suspicious high-volume queries to models. However, If the leading laboratories themselves are using this technique against each other, the debate becomes much more complex than it seemed.
What this episode makes clear is that the race for manmade intelligence is not as clean as the press releases make it seem. Behind the announcements of record-breaking benchmarks and statements about responsible security, companies observe, copy and learn from each other, sometimes in ways that their own policies would not allow outside users. Elon Musk, almost unintentionally, has just confirmed what many already suspected.
Keep reading:
• Elon Musk makes good on his threat and sues Apple
• OpenAI reveals Elon Musk secrets in response to billionaire’s lawsuit
• Sam Altman and OpenAI want their own “SpaceX” to compete with Elon Musk in the AI space race
