Elon Musk's xAI reportedly trained its coding models on Claude outputs for months before getting cut off

2026-06-06 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Fundamental Awareness, quick

Summary

Elon Musk's xAI reportedly trained its coding models for months by "distilling" outputs from Anthropic's Claude, according to a June 6, 2026 report by The Information. After Anthropic revoked official access in January, xAI engineers continued this practice using personal accounts and the intermediary service Blackbox AI. This follows Musk's prior admission that xAI "partially" used OpenAI models to train Grok, which he termed an "industry standard." Internally, xAI faces significant challenges, including a shrunken pretraining team, the departure of four Grok code leads and multiple co-founders, and an incident where an employee accidentally deleted critical training data, costing two to three weeks of work. Consequently, Musk is now renting out substantial compute resources, acquired for xAI, to Anthropic and Google via SpaceX, describing this as a temporary measure.

Key takeaway

For Directors of AI/ML building competitive models, this report highlights the risks of relying on competitor outputs and the critical need for robust internal operations. Your team's data governance and retention policies must be impeccable to prevent costly setbacks like accidental data deletion. Furthermore, consider the strategic implications of compute acquisition. Ensure your internal projects fully utilize resources before large investments, or plan for alternative monetization strategies.

Key insights

xAI reportedly trained its coding models on competitor outputs, facing internal turmoil and now renting out its acquired compute.

Principles

• Model training can involve distilling competitor outputs.
• Internal operational issues can severely impede AI development.
• Compute resources can be monetized if internal projects stall.

Method

The reported method involved training xAI's coding models directly on outputs generated by Anthropic's Claude, initially via official access, then through personal accounts and Blackbox AI after access was revoked.

In practice

• Investigate competitor model outputs for training data.
• Implement strict data management and backup protocols.
• Diversify compute utilization strategies.

Topics

• xAI
• Anthropic Claude
• Model Training
• Compute Infrastructure
• Data Governance
• AI Ethics

Best for: CTO, VP of Engineering/Data, Executive, Tech Journalist, Investor, Director of AI/ML

Related on AIssential

• Is context the new compute? How the AI race is moving to the data layer — AI competitive advantage now stems from proprietary organizational context and robust data infrastructure, not just model compute.
• Yet another co-founder departs Elon Musk's xAI — xAI faces significant leadership turnover and regulatory scrutiny amidst strategic mergers and financial challenges.
• Top 6 Claude Security Risks to Watch as AI Becomes Your Employees' Operating System — Claude's pervasive enterprise integration creates significant, often unmonitored, security risks across data, access, autonomous operations, and code generation.
• Musk's SpaceX bets $60 billion on Cursor to fix xAI's coding gap — Strategic acquisitions can bridge technology gaps and provide critical resources like compute capacity.
• The $200 Billion Data Debt: Every Major AI Lab Is Running Out of Fuel and Nobody Is Talking About… — The AI industry faces a critical data debt crisis as high-quality human-generated data is exhausted and synthetic data causes model collapse.
• One company reportedly spent $500 million on Claude in one month after failing to cap AI usage — AI cost overruns stem from uncapped usage, misuse, and poor model selection, necessitating internal expertise.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.