- xAI spent months training its coding models on outputs from Anthropic’s Claude, The Information reports.
- After Anthropic revoked official access in January, xAI engineers continued via personal accounts and the intermediary service Blackbox AI.
- xAI’s pretraining team reportedly shrank to fewer than five people, with four Grok code leads departing.
- Musk is now renting his accumulated compute to Anthropic (via SpaceX) and Google rather than training his own models.
What Happened
Elon Musk’s xAI spent months distilling Anthropic’s Claude to train its own coding models, The Information reports via The Decoder. xAI trained its coding model directly on Claude‘s outputs — a practice known as distillation.
After Anthropic revoked official access in January 2026, xAI engineers reportedly kept going through personal accounts and the intermediary service Blackbox AI.
Why It Matters
Training one lab’s model on another’s outputs sits in a contested legal and ethical gray zone. Elon Musk previously admitted in court that xAI “partially” used OpenAI models to train Grok, calling it industry standard — but doing so after a provider revokes access raises sharper questions.
It also reframes xAI’s competitive position against the labs behind Claude Opus 4.8 and the broader frontier race: a company leaning on a rival’s outputs is, by definition, trailing it.
Technical Details
The report describes internal turmoil. xAI’s pretraining team reportedly shrank to fewer than five people, and four Grok code leads left within months, along with several co-founders. One employee accidentally deleted critical training data, costing two to three weeks of work.
Meanwhile, the compute Musk amassed for xAI is being leased out — to Anthropic via SpaceX and to Google — rather than used to train xAI’s own models, described as a stopgap.
Who’s Affected
Anthropic is directly affected, having moved to cut off a competitor it says misused access. Developers choosing between Grok and rivals are affected too, since the report raises questions about how much of Grok’s coding ability is independently developed. The leasing arrangement connects to SpaceX’s broader role as an AI-infrastructure provider ahead of its IPO.
What’s Next
The claims come from The Information’s reporting and have not been confirmed by xAI or Anthropic on the record. Whether Anthropic pursues further action, and whether xAI returns to training its own frontier models rather than renting compute, are the open questions.
