xAI Allegedly Trained Coding Models on Anthropic's Claude Outputs for Months

Elon Musk's xAI trained its coding models on Anthropic's Claude API outputs for months. After Anthropic revoked official access in January 2026, xAI engineers continued accessing Claude through personal accounts and third-party service Blackbox AI. Musk admitted in court that xAI partially used OpenAI models to train Grok, calling the practice of training on competitor outputs an industry standard.

xAI training on Claude outputs

• ▪xAI spent months distilling Anthropic's Claude to train xAI's own coding models.
• ▪Elon Musk called using other AI models to train new models an industry standard.
• ▪xAI trained its coding model directly on Claude's outputs.
• ▪Elon Musk admitted in court that xAI partially used OpenAI models to train Grok.

Access revocation workarounds

• ▪Anthropic revoked xAI's official access to Claude in January 2026.
• ▪xAI engineers continued accessing Claude through personal accounts after Anthropic revoked official access.

Internal xAI troubles

• ▪Four Grok code leads left xAI within months.
• ▪Many xAI co-founders left the company.
• ▪One xAI employee accidentally deleted critical training data, costing two to three weeks of work.
• ▪xAI's pretraining team shrank to under five people.

Compute infrastructure rental

• ▪Elon Musk is renting compute infrastructure to Anthropic via SpaceX instead of training xAI's own models.

Perspective of Elon Musk / xAI

• ▪Elon Musk called using other AI models to train new models an industry standard.
• ▪Elon Musk admitted in court that xAI partially used OpenAI models to train Grok.

1 source