Report: xAI distilled Anthropic's Claude to train its coding models

According to The Information, xAI spent months distilling Anthropic's Claude—using outputs from the rival model to train xAI's own coding systems—and reportedly kept doing so through personal accounts even after Anthropic revoked xAI's official API access in January 2026. Distillation, training a model on another model's outputs, sits in a legal and ethical gray zone and typically violates the source provider's terms of service.

The report lands amid a broader Business Insider thesis that frontier-model, agent and 'vibe-coding' companies are racing to become full-stack AI players as ever-rising valuations demand new revenue streams, blurring the lines between competitors and pushing them into each other's markets. xAI has been aggressively building coding capabilities, recently shipping Grok worktree support and a V9-Medium core model.

Competitively, the allegation is awkward for xAI given Anthropic's simultaneous positioning as the safety-first, governance-minded lab now calling for an industry pause. If accurate, it underscores how dependent fast-moving labs are on rivals' outputs to bootstrap capability—and how hard such usage is to police once access is revoked. Anthropic has not detailed enforcement steps publicly. Readers should watch for any formal response or legal action from Anthropic, and whether other labs face similar distillation scrutiny.