Elon Musk in court admitted that xAI partially trained Grok on OpenAI models

Elon Musk admitted in federal court in California that xAI did indeed use OpenAI models when training Grok. For the industry, this is one of the most uncomfortable admissions of the year: companies publicly fight distillation, but in reality, it seems they apply it to each other as well.

Court Admission

During Musk's testimony, he was directly asked whether xAI used distillation methods on OpenAI models to train Grok. He first responded that such practice is generally common among AI companies. When the question was clarified and he was asked to say it simply—does that mean "yes"—Musk replied: "partially."

For the first time at a public level, an admission was made that one of the most prominent American AI companies relied on a competitor's results when training its own model. This statement was not made at a specialized conference or on a podcast, but in court, where Musk himself is the plaintiff against OpenAI, Sam Altman, and Greg Brockman. He argues that OpenAI has abandoned its original non-profit mission and shifted toward a commercial structure.

Against this backdrop, the admission about Grok looks especially sensitive: the dispute is not only about OpenAI's values and structure, but also about how the race between laboratories is actually organized.

Why the Dispute Is Sharp

Distillation typically refers to training a new model on the responses of a stronger system through a public chat or API. This approach doesn't provide the original weights but helps reproduce the style of responses, logic, and some useful behavioral patterns. For smaller players, this is a chance to cut corners: instead of repeating the entire expensive research and training cycle from scratch, take the leader's behavior as a reference and more quickly achieve a nearly comparable result.

"Partially,"

Musk replied when asked whether xAI used distillation of OpenAI models to train Grok.

This is precisely why the topic has become so toxic for market leaders. OpenAI and Anthropic have been increasingly speaking in recent months about protecting their models from systematic requests through which competitors can collect large arrays of responses for training. Until now, public focus has often been on Chinese companies trying to build cheaper open-weight models this way. But Musk's testimony confirms what the industry has long suspected: American laboratories probably use similar techniques against each other as well.

What Musk Said

An important legal detail here: distillation itself may not be directly prohibited by law, but it can easily violate the terms of service of services. This is where one of the main future conflicts of the AI market will play out. If the practice is not considered unambiguously illegal, then major companies will have to defend themselves not so much through copyright courts but through API restrictions, monitoring of suspicious bulk requests, and industry agreements.

There are several other details that help understand the context:

• OpenAI, Anthropic, and Google reportedly exchange information on fighting distillation attempts through the Frontier Model Forum.
• One defense method is tracking suspicious bulk requests to models through chat and API.
• At the time of publication, OpenAI did not respond to a request for comment on Musk's statements.
• Later in the same testimony, Musk ranked Anthropic first among AI companies, then named OpenAI, Google, and Chinese open-source models.
• He described xAI as a much more compact company with a few hundred employees.

This part of the testimony is also important. On one hand, Musk essentially acknowledges young xAI's dependence on stronger players at an early stage. On the other hand, it shows how normalized the idea of "learning" from a competitor through their public product has become. For the market, this is another signal: the race is not only in computing and talent, but also in who better limits access to their model's behavior.

What It Means

The Grok story shows that the dispute over distillation has long moved beyond theory and geopolitics. While major laboratories talk about protecting against copying, the mechanics of training remain a gray area where competitors simultaneously fight each other and borrow from each other. For users and business, this means the battle for AI leadership is increasingly shifting from model quality to control over who and how can retrain them.