Musk Testifies xAI Trained Grok on OpenAI Models — Bombshell Trial Admission

AI brain with overlapping logos suggesting copy/inheritance

Elon Musk testified Thursday in the OpenAI trial that xAI trained its Grok chatbot using OpenAI's GPT-4 outputs. The admission, given under direct examination by Musk's own counsel, came in response to a question about how Grok's early architecture was developed. The answer was unexpected — Musk's legal team had not signaled this would surface — and reverberated immediately through the courtroom.

The specific testimony, per court reporters: Musk said xAI used "OpenAI's API and GPT-4 outputs as part of bootstrapping training data" during Grok's initial development, but "transitioned to fully independent training infrastructure" by mid-2024. The phrasing matters legally — "training data" is the trigger for OpenAI's terms of service, which expressly prohibit using GPT outputs to train competing models.

Why this admission matters

Three legal vectors:

OpenAI's TOS violation. The standard OpenAI API agreement explicitly bars using outputs to train competing AI models. If Musk's testimony is taken at face value, xAI breached that agreement directly — and xAI has revenue, so the damages calculation is non-trivial.

The hypocrisy framing. Musk has spent two years publicly accusing OpenAI of breaching foundational principles. The admission that xAI used OpenAI's outputs to bootstrap a competitor cuts directly into the moral framework of his lawsuit. The jury heard him say it on the same day he was cross-examined about his own contradictory tweets.

Counterclaim potential. OpenAI's legal team had not previously raised a counterclaim. The admission gives them grounds to file one. Sources close to OpenAI's litigation strategy told reporters the team is "actively evaluating" filing within 30 days — which would significantly complicate Musk's primary case.

The strategic miscalculation

Musk's counsel apparently asked the question to set up a contrast — to argue that xAI's commercial position validates the for-profit pivot Musk is suing OpenAI over. The strategy presumably was: "Here's how successful xAI became under a for-profit model that OpenAI also adopted." The problem is that the answer also includes the admission that xAI's launch leveraged OpenAI's outputs in violation of their API terms.

Trial-strategy observers called it one of the most damaging unforced errors of the 2026 legal year. The admission goes into the trial record permanently, regardless of how the rest of the case resolves. It will be referenced in any future xAI versus OpenAI litigation.

What xAI has said publicly

xAI's official position has historically been that Grok was trained "from scratch" on web-scale data plus X (Twitter) data. The new admission suggests that public framing was incomplete at best. xAI hasn't issued a statement responding to today's testimony but the question is now in the court record and reportable.

From an enterprise-AI sales perspective, this matters too. xAI has been pitching Grok to Fortune 500 customers on the basis of clean-room independence from OpenAI. That pitch is harder to make credible with Musk's own testimony in the record.

My Take

This is the bigger legal story of the day, more than the tweet contradictions cross-examination. The tweet contradictions damage Musk's credibility as a witness. This admission damages xAI as a company and creates a credible counterclaim that could materially affect the outcome. If OpenAI files the counterclaim within 30 days, the case complexity multiplies and settlement becomes more attractive for both sides — though on terms much less favorable to Musk than what he's been pushing for. The deeper question is what happens to xAI commercially. Customers buying Grok for differentiation reasons (we're not OpenAI) now have to grapple with the fact that Grok was bootstrapped from OpenAI in violation of TOS. That's a procurement-team conversation that doesn't go well. I'd bet xAI's enterprise pipeline takes a measurable hit through Q3.

FAQ

Was this a slip-up or a planned admission? Likely a slip-up. Musk's counsel showed visible surprise; sources said the legal team didn't expect Musk to volunteer the OpenAI-bootstrap detail.

Can OpenAI still file a counterclaim mid-trial? Yes — counterclaims arising during trial are allowed if they don't fundamentally change the case scope. OpenAI could file as a separate action or as an amended counterclaim within procedural deadlines.

Does this affect xAI's Series E? xAI's recent fundraise closed in March 2026. New investors may push for legal-protection language; the existing round terms are presumably less protective than they would be today.

The Bottom Line

Musk's own counsel asked a question that produced a major admission against xAI: Grok was trained partly on OpenAI's API outputs, in apparent violation of OpenAI's TOS. The opening for OpenAI to file a counterclaim is real. Trial dynamics shifted noticeably.

Related Articles

Sources