In a federal court appearance this week, Elon Musk seemed to acknowledge that xAI used OpenAI's models in training its own, while simultaneously arguing the practice is routine across the AI industry, a position that cuts directly against the legal case he has been building against OpenAI.
There are moments in litigation where a witness says something that stops the room. Elon Musk provided one of those moments this week when, during cross-examination by an OpenAI attorney in federal court, he appeared to indicate that xAI may have used OpenAI's models to train its own systems. His defense of that position, that it is standard practice among AI labs to train on competitors' outputs, is either a candid admission about how the industry actually operates or a significant strategic miscalculation in the middle of a lawsuit he initiated. Possibly both.
The irony is difficult to overstate. Musk has been pursuing legal action against OpenAI on grounds that include allegations the company betrayed its founding mission and operated in ways that harmed the broader AI development ecosystem. He positioned himself, at least publicly, as the principled actor in this dispute, the co-founder who left when OpenAI abandoned its nonprofit commitments and who is now seeking accountability. Testifying under oath that xAI may have done the thing OpenAI is presumably also accused of, and then framing it as acceptable industry behavior, introduces a complexity into that narrative that OpenAI's legal team will not leave unexplored.
The Technical Reality Behind Model Distillation
What Musk described under oath has a name in machine learning circles: model distillation. The technique involves using outputs from a more capable model to train a smaller or less mature system, effectively transferring knowledge without direct access to the original training data or architecture. It is a well-documented approach that has been discussed openly in academic literature for years, and there is credible evidence that multiple companies across the AI landscape engage in some form of it.
However, there is a meaningful distinction between acknowledging that something happens and admitting under oath that your own company does it while suing a competitor over practices you consider improper. The legal threshold here matters. Musk's attorneys now face the task of reconciling their client's sworn testimony with the broader claims in his complaint, which paints OpenAI's behavior as uniquely harmful rather than part of an industry-wide pattern that xAI itself participates in.
Implications for the Competitive Landscape
Musk's courtroom admission also pulls back the curtain on a fiercely competitive race where the lines between inspiration, imitation, and ingestion of rival outputs remain blurry. Grok, xAI's flagship model integrated into the X platform, was developed on an aggressive timeline that raised eyebrows among AI researchers who questioned how quickly the system achieved parity with established players in certain benchmarks. The testimony suggests at least one possible explanation for that speed.
This is not just about two companies. Every major AI lab, from Google DeepMind to Anthropic to Meta's FAIR division, operates in an environment where competitor models are accessible through APIs and consumer-facing applications. The question of whether training on those outputs constitutes fair use, intellectual property theft, or something in between remains unsettled law. Musk's sworn statement could become a reference point in future cases that attempt to establish boundaries around what is permissible when building AI systems.
Legal Strategy Meets Public Relations
From a litigation standpoint, OpenAI's attorneys scored a significant win by eliciting this admission. It allows them to frame Musk not as a wronged co-founder seeking justice but as a competitor who plays by the same rules he is challenging in court. That narrative undermines the moral authority that has been central to Musk's public positioning since he filed suit earlier this year. Judges and juries tend to be unsympathetic to plaintiffs whose own conduct mirrors what they are complaining about.
There is also a public relations dimension that cannot be ignored. Musk has cultivated a reputation as an outsider who challenges entrenched interests, whether that is the auto industry with Tesla, aerospace incumbents with SpaceX, or social media gatekeepers with X. Having it stated on the record that his AI company may have relied on the outputs of a rival he has publicly castigated introduces friction into that brand identity in a way that extends well beyond the courtroom.
What This Means for AI Governance Going Forward
The broader industry should be paying close attention. If training on competitor model outputs is truly as widespread as Musk suggested, then the entire sector is operating in a gray zone that regulators have not yet caught up to. The European Union's AI Act, which is being implemented in phases, does not directly address this practice. Neither do the various executive orders and legislative proposals circulating in Washington. At some point, policymakers will have to decide whether model distillation from commercial competitors is acceptable, and when it crosses into anti-competitive behavior or intellectual property infringement.
For startups building foundation models, Musk's testimony creates an uncomfortable precedent. Many early-stage companies lack the resources to train models from scratch and rely heavily on synthetic data generated by larger, more capable systems. If courts eventually rule that this practice violates the rights of the original model creators, an entire tier of the AI ecosystem could face existential legal risk. Conversely, if it is deemed permissible, then the competitive moats that companies like OpenAI and Google have built around their most advanced models may be thinner than their valuations suggest.
The Road Ahead for Musk's Case
The lawsuit is far from resolved, and Musk's legal team will have opportunities to argue that the issues at stake in his complaint, primarily the alleged breach of fiduciary duty and the abandonment of OpenAI's nonprofit charter, are separate from whatever xAI may or may not have done with competitor outputs. That is a plausible legal argument. But plausible legal arguments do not always survive contact with the court of public opinion, and Musk has always operated in a space where perception matters as much as legal technicality.
What is clear is that this deposition moment has shifted the dynamics of the case. OpenAI now has a sworn admission to work with, and the discovery process may yield additional details about the extent to which xAI used rival models during its development. For a lawsuit that was already being closely watched as a referendum on the direction of the AI industry, the stakes just got considerably higher.
Also read: Z.ai fixes GLM-5 inference bugs and gives the fix back to the open source community • AI has quietly transformed from a software story into a capital intensive industry and most investors have not updated their models • Big Tech is spending $725 billion on AI infrastructure and the market is starting to ask when that bill comes due
