DooDooLamb News
Elon Musk Seemingly Admits xAI Has Used OpenAI's Models to Train Its Own
Brief published May 2, 2026 ยท Original source published April 30, 2026
Original reporting by Maxwell Zeff, Paresh Dave at wired.com.
Automated brief. Verify important details at the original source.
Elon Musk Accidentally Confesses to AI Training's Dirty Little Secret While Under Oath
In a stunning display of courtroom honesty that probably made his lawyers consider early retirement, Elon Musk casually admitted Thursday that his AI company xAI has been using OpenAI's models to train its own systems. This revelation came during federal court testimony where Musk, apparently forgetting that "everyone does it" isn't actually a legal defense, argued that using competitors' AI models is just standard industry practice.
The admission emerged while Musk was defending his various crusades against OpenAI, the company he co-founded before dramatically storming out in 2018 like a billionaire leaving a group chat after an argument about splitting the dinner bill. During his testimony, Musk seemed to suggest that xAI's Grok chatbot (yes, the one named after a word from a 1960s science fiction novel because tech bros never met a literary reference they couldn't awkwardly appropriate) may have been trained using OpenAI's models as part of its development process.
For those blissfully unaware of how AI training works, imagine teaching a parrot to speak by playing it recordings of other, more articulate parrots. That's essentially what's happening here, except the parrots cost hundreds of millions of dollars to create and the recordings are actually mathematical representations of human language compressed into incomprehensible numerical soup. When AI companies "train" their models, they feed them vast amounts of text data so the systems can learn patterns and generate human-like responses. The dirty secret Musk accidentally spilled is that sometimes, instead of starting from scratch with raw internet text, companies take shortcuts by using the outputs of already-trained models as training data. It's like learning to paint by copying masterpieces, except the masterpieces were created by robots and nobody wants to admit they're copying.
The context here is deliciously ironic. Musk has spent the better part of two years publicly feuding with OpenAI and its CEO Sam Altman, accusing them of betraying their original open-source mission by becoming a for-profit company focused on building artificial general intelligence. He's sued them multiple times, withdrawn lawsuits, then filed new ones in a legal dance that resembles a billionaire's version of sliding into someone's DMs just to argue. The whole dispute centers around Musk's claim that OpenAI abandoned its founding principles, which is rich coming from someone whose own AI company apparently can't resist borrowing from the very models he's spent months criticizing in increasingly unhinged late-night Twitter posts.
What makes this particularly entertaining is that Musk's defense essentially boils down to "everyone's doing it, so it's fine." This argument might work when you're a teenager caught vaping behind the gym, but it's slightly less compelling when you're a multi-billionaire who positioned your AI company as a righteous alternative to the corrupt establishment. It's like starting a organic restaurant and then getting caught buying ingredients from McDonald's while insisting that "all chefs do this." Technically true doesn't make it less hypocritical when you've built your entire brand on being different.
The broader industry context makes this admission even more awkward. The AI world is currently embroiled in massive copyright lawsuits over training data, with publishers, artists, and writers arguing that their work was used without permission to create these AI systems. Meanwhile, the AI companies themselves are becoming increasingly secretive about their training processes, with newer models like GPT-4 and Claude offering virtually no details about what data was used to train them. It's an industry built on borrowed content arguing about intellectual property while simultaneously refusing to reveal their own sources, like a group of art forgers forming a union to protect their trade secrets.
The reality check here is that Musk's casual admission reveals something everyone in AI already knows but rarely discusses openly: the entire industry is built on a foundation of borrowed intelligence. Every major AI lab has likely used competitors' models at some point, whether for comparison, benchmarking, or training their own systems. The difference is that most companies have the good sense not to admit it under oath while simultaneously suing those same competitors for ethical violations. This isn't necessarily illegal or even unethical, but it does make Musk's moral crusade against OpenAI look about as principled as a convenience store wine selection.
For developers and founders watching this unfold, the lesson is clear: maybe don't position your company as the ethical alternative to Big Tech while your engineers are quietly copying Big Tech's homework. The AI industry has enough credibility problems without its most visible leaders accidentally confessing to the very practices they publicly condemn. But hey, at least we finally got an honest answer about how the sausage gets made, even if it took a federal courthouse to extract it.