Musk Admits Under Oath That xAI Used OpenAI Models for Training

Musk's Sworn Testimony Ignites Controversy

In a closely watched legal proceeding, Tesla and xAI founder Elon Musk appeared to acknowledge a fact that has sent shockwaves through the industry during sworn testimony — xAI had used competitor OpenAI's models to train its own AI systems. When questioned, Musk defended the practice as "standard practice" among AI labs, claiming that companies routinely use competitors' models as references and training aids.

This statement is particularly noteworthy given that Musk has long been one of OpenAI's fiercest critics. He has repeatedly and publicly attacked OpenAI for straying from its nonprofit mission and has even filed lawsuits against the company. The admission that xAI relied on OpenAI's technological output to build its own capabilities undeniably complicates the narrative of this rivalry.

What Is 'Distillation'? The Gray Area of the AI Industry

The "standard practice" Musk referred to is technically known as Model Distillation. In simple terms, it involves using the outputs of a powerful model to train another model, enabling the latter to approach the former's performance while potentially gaining advantages in efficiency or on specific tasks.

This practice is indeed not uncommon in the AI industry. Multiple AI companies have previously been accused of or have admitted to using competitors' model outputs to improve their own products. For instance, Google researchers were found to have used ChatGPT data, and several open-source model projects have sparked controversy for training on outputs from the GPT series of models.

However, "widespread" does not equate to "legitimate and legal." The terms of service of major AI companies like OpenAI typically explicitly prohibit the use of their API outputs for training competitive models. If xAI did indeed use OpenAI's API calls or model outputs to train the Grok series of models, this could constitute a violation of those terms of service and potentially raise deeper intellectual property concerns.

A New Chapter in the Musk-OpenAI Feud

The tensions between Musk and OpenAI have a long history. As a co-founder and early major backer of OpenAI, Musk departed the organization's board in 2018. Since then, he has repeatedly accused OpenAI, under CEO Sam Altman's leadership, of transforming from a nonprofit dedicated to AI safety into a profit-driven commercial entity.

In 2024, Musk formally filed a lawsuit against OpenAI, alleging it had betrayed its founding mission of openness. Simultaneously, he founded xAI and launched the Grok series of large language models, directly competing with OpenAI's GPT series.

Now, the sworn testimony admitting to using a rival's models for training has placed Musk in a rather awkward position. On one hand, he accuses OpenAI of betraying the principles of openness; on the other, his own company has been leveraging OpenAI's technological achievements for its own benefit. Critics point out that this exposes the contradictions in Musk's stance within the AI space.

Industry Impact and Legal Risks

This incident could have far-reaching implications on multiple levels:

Legal implications: OpenAI could use this as grounds to bring claims against xAI for violating its terms of service or even intellectual property infringement. Given that the two parties are already in legal confrontation, this testimony could become a powerful weapon in OpenAI's counterattack.

Industry standards: Musk's "standard practice" claim will force the entire AI industry to confront the ethical and legal boundaries of model distillation. Major AI companies may strengthen monitoring of API usage and further clarify restrictions in their terms of service.

Competitive landscape: If xAI's core capabilities are partly built on a competitor's technological foundation, it will raise questions about the originality and technical independence of the Grok models.

Looking Ahead: AI Competition Rules Need Clarification

Musk's testimony has thrust a long-standing but rarely publicly discussed "unwritten rule" of the AI industry into the spotlight. As competition among large AI models intensifies, defining the boundary between legitimate technical reference and improper model copying is becoming an urgent question that demands answers.

Regardless of the ultimate legal outcome, this incident serves as a reminder to the entire industry: establishing fair and transparent rules of competition is vital to the healthy development of the AI ecosystem while pursuing technological breakthroughs. Going forward, compliance requirements surrounding the sources and methods of AI model training data are expected to become a key focus for regulators.