Enter your email address below and subscribe to our newsletter

xAI allegedly used Claude outputs to train Grok for months

Share your love

  • xAI allegedly ran a months-long distillation project using Anthropic’s Claude outputs to train its own coding models, according to a report citing six sources.the-decoder
  • After Anthropic blocked xAI’s access in January, engineers reportedly used personal accounts and third-party service Blackbox AI to keep querying Claude.the-decoder
  • The revelations complicate a massive compute deal in which Anthropic is paying xAI $1.25 billion monthly to lease its Colossus 1 data center, according to TechCrunch.techcrunch

xAI Allegedly Trained Grok on Claude Outputs After Anthropic Revoked Access

Elon Musk’s xAI spent months running a model distillation project that used outputs from Anthropic’s Claude to train its own coding models — and reportedly continued the practice even after Anthropic cut off the company’s access in January 2026, according to a report citing six people familiar with the projects.the-decoder

Workarounds After the Cutoff

In January 2026, xAI co-founder Tony Wu informed staff internally that Anthropic had formally blocked the company’s enterprise access to Claude, a policy enforcement targeting major competitors. But according to two people with knowledge of the projects, xAI engineers shifted to using personal accounts to continue querying Claude, prompting Anthropic to shut down batches of associated individual accounts.sina

xAI also allegedly routed requests through Blackbox AI, a third-party service, to access Anthropic’s models through encrypted channels. Two industry sources told The Decoder that as recently as mid-May 2026, xAI was still using the platform to call Anthropic models for benchmarking and research work. Blackbox AI did not respond to requests for comment.the-decoder

A Pattern of Distillation

The allegations follow Musk’s own courtroom admission in late April 2026 that xAI had “partly” used OpenAI’s technology in training Grok, a practice he described as common in the industry. Anthropic’s terms of service explicitly prohibit customers from using Claude outputs to develop competing AI products.innobu

The practice mirrors accusations Anthropic itself leveled against Chinese AI labs in February 2026, when the company said DeepSeek, Moonshot AI, and MiniMax had created over 24,000 fake accounts and generated more than 16 million exchanges with Claude in what it called “industrial-scale” distillation campaigns.firstpost

An Awkward Business Relationship

The revelations add tension to what is already one of the AI industry’s most unusual partnerships. In May 2026, Anthropic agreed to lease the entirety of xAI’s Colossus 1 data center — more than 220,000 Nvidia GPUs — for $1.25 billion per month through May 2029, a deal that could generate over $40 billion in revenue for xAI. xAI has since moved its own training operations to the newer Colossus 2 cluster.kucoin

Neither xAI nor Anthropic has publicly commented on the distillation allegations. Musk and xAI have not disclosed the full extent of the company’s use of Anthropic’s technology or the workarounds its employees employed after the access cutoff.sina

Leave a Reply

Your email address will not be published. Required fields are marked *

Stay informed and not overwhelmed, subscribe now!