Musk Admits Under Oath: xAI Used OpenAI Models to Train Its Own AI

Musk Admits Under Oath: xAI Used OpenAI Models to Train Grok

In a closely watched legal proceeding, Tesla and xAI founder Elon Musk appeared to confirm under oath what the industry had long speculated — xAI used output data from competitor OpenAI's models during the training of its own large language model, Grok.

The admission quickly sent shockwaves through the AI industry, not only because of the longstanding feud between Musk and OpenAI, but also because the practice touches on one of the most sensitive gray areas in AI: Is it permissible to use a competitor's model outputs to train your own model?

Musk's Defense: 'Standard Industry Practice'

When questioned, Musk did not shy away from the issue. Instead, he characterized it as a "standard practice" among AI labs. He argued that in the current competitive landscape, it is common for labs to reference or even use outputs from rival models to optimize and train their own systems.

This claim is not entirely without basis. In AI research, a technique known as "Model Distillation" is indeed a widely used method — leveraging the outputs of a stronger model to guide the training of a weaker one. Multiple AI companies have previously been reported to engage in similar practices. Google researchers, for instance, have published papers on using distillation techniques to improve the performance of smaller models.

However, a vast gap exists between "technically feasible" and "legally compliant." Most AI companies' terms of service explicitly prohibit users from using model outputs to train competing models. OpenAI's terms of use include similar restrictive clauses, expressly forbidding the use of its API outputs to develop models that compete with OpenAI.

A New Chapter in the Musk-OpenAI Feud

The drama of this revelation is amplified by the complicated history between Musk and OpenAI. Musk was one of OpenAI's co-founders and provided substantial early funding. However, after OpenAI transitioned from a nonprofit to a "capped-profit" company, the relationship deteriorated sharply.

Musk has repeatedly criticized OpenAI for "abandoning its mission," accusing CEO Sam Altman of turning what was meant to be an open-source project benefiting all of humanity into a "profit-driven machine." He even filed a lawsuit against OpenAI, alleging it violated its founding agreement. Meanwhile, Musk founded xAI in 2023 and launched the large language model Grok, directly competing with OpenAI's GPT series.

Now, Musk's admission that xAI used OpenAI's model outputs for training makes this entanglement even more delicate. On one hand, Musk is suing OpenAI in court; on the other, his company has been leveraging OpenAI's technological outputs. This "suing while borrowing" dynamic has placed Musk in a rather awkward position in the court of public opinion.

Industry Impact: The Gray Area of AI Training Data

Musk's testimony has thrust a long-standing but rarely discussed issue in the AI industry into the spotlight.

Blurred intellectual property boundaries. Currently, there is no clear legal framework worldwide regarding the intellectual property ownership of AI model outputs. Do model outputs constitute protected intellectual property? Does using one company's model outputs to train another company's model constitute infringement? These questions remain in a legal gray area.

Industry open secrets come to light. While Musk's characterization of this as "standard practice" is unsettling, it may indeed reflect the reality of the industry. Previously, Chinese AI company DeepSeek faced similar questions about whether it used OpenAI's model outputs for training when it released its model. Such practices may be far more prevalent in the industry than outsiders realize.

Deepening trust crisis. If "mutual borrowing" among AI companies is widespread, user trust in AI service providers will face serious challenges. Could the data and interaction logs generated when enterprise users call models via APIs also be used to train competing products? This concern will directly influence enterprise decisions on adopting AI services.

Legal Consequences and Industry Trajectory

From a legal perspective, Musk's testimony could have multiple ramifications. First, OpenAI could file a lawsuit against xAI for violating its terms of service, or even pursue intellectual property infringement claims. Second, this incident could push regulators to accelerate the development of regulations governing the use of AI model training data.

From an industry development standpoint, the incident could catalyze the following trends:

• Stricter API terms of use, with AI companies potentially strengthening technical monitoring and legal restrictions on how model outputs are used
• Accelerated development of model output "watermarking" technologies to help AI companies track whether their model outputs are being used to train competing products
• Greater urgency for industry self-regulatory standards, with leading AI labs potentially needing to reach some consensus on data sources for model training

Outlook: Transparency Will Be the Next Critical Issue for the AI Industry

Musk's sworn testimony, while potentially exposing xAI and himself to legal risks, has objectively lifted the veil on an important secret of the AI industry. As competition among large AI models intensifies, transparency around training data sources, intellectual property protection of model outputs, and the boundaries of fair competition will all become core issues the industry must confront.

For the entire AI ecosystem, this may represent an opportunity to establish a more regulated and transparent industry order. After all, only within a clear framework of rules can innovation truly develop sustainably.