Elon Musk just imploded his own lawsuit against OpenAI, and he did it while under oath. In a federal courtroom in California on Thursday, the billionaire testified that his AI startup xAI used OpenAI's models to train its Grok chatbot through a process called model distillation—where one AI system trains another by processing its outputs. The irony cuts to the core of the case he's spent millions prosecuting: Musk is suing OpenAI for betraying its nonprofit, open-source origins by becoming a closed, profit-first company. His own company, it turns out, built Grok on OpenAI's work.
The tension here is not subtle. Musk's lawsuit argues that OpenAI's pivot toward commercialClosedness violated the spirit of its founding agreement—a mission to develop artificial general intelligence for humanity's benefit, not for private profit. Yet xAI appears to have benefited directly from OpenAI's models, using them as teaching data to improve its own system. If OpenAI's outputs were so tainted by commercialism that they represented a betrayal of public trust, why did xAI's engineers find them useful enough to incorporate into Grok?
Musk's legal team sought to frame distillation as standard industry practice, not a uniquely suspicious behavior. Under questioning, he acknowledged that using competitor models to train one's own is common across the sector. But this defense is a double-edged sword. If distillation is ubiquitous—and Musk himself now confirms it—then the lawsuit's claim that OpenAI uniquely violated some ethical standard becomes harder to sustain. The narrative shifts from OpenAI betraying its mission to OpenAI building models valuable enough that even its critics' companies want to learn from them.
The timing compounds the awkwardness. Musk made headlines weeks ago with a $97.4 billion offer to buy OpenAI's nonprofit assets. That bid was already contentious. Now his own testimony suggests xAI benefited from OpenAI's research while simultaneously attempting to acquire the source of that benefit. OpenAI's lawyers will likely argue this undermines any claim of unique harm—Musk cannot simultaneously exploit OpenAI's work and argue that work damaged him.
The broader context matters too. This case unfolds as frontier AI labs increasingly treat model distillation as a competitive threat, not merely an industry courtesy. Companies like Google and Anthropic have built infrastructure specifically to prevent rivals from using their outputs for training. The debate over whether distillation constitutes fair knowledge transfer or intellectual property theft remains unresolved. Musk, by admitting xAI engaged in it, positions himself somewhere in that gray zone—decrying OpenAI's practices while partaking in them.
What happens next is legally murky. The case was always going to be complicated; AI governance lacks clear precedents for distinguishing legitimate knowledge transfer from exploitative copying. But Musk's testimony introduces a new variable: the plaintiff himself appears to have done what he's accusing the defendant of. Courts will have to decide whether that matters, and whether the lawsuit's central premise—that OpenAI uniquely wronged the public by becoming profitable—survives contact with reality.
For now, the image is striking: a man who funded OpenAI to benefit humanity now suing that same organization while running a competitor built partly on its work. Whether you view this as a tragic betrayal or simply the messy reality of AI competition, the contradiction is difficult to ignore. Musk's lawyers will adjust their strategy. The judge will weigh the evidence. But the testimony cannot be unsaid—xAI trained on OpenAI, and Musk said so himself.