Elon Musk's xAI used a technique called model distillation to train its Grok coding models on outputs from Anthropic's Claude for months, even after Anthropic revoked official access in January 2026. The revelation, reported by The Information, exposes how frontier AI labs are now treating their model outputs as proprietary training data worth protecting, and raises questions about what constitutes fair competition in the AI industry.

What Is Model Distillation and Why Does It Matter?

Model distillation is a technique where engineers train a less capable model on the outputs of a stronger one, essentially teaching it to mimic the frontier model's behavior without needing the same massive training budget or data. Think of it like learning to write by studying the work of a master author, rather than starting from scratch. xAI used this approach to improve Grok's coding abilities by learning from Claude's responses.

The practice sits in a gray legal and ethical zone. It's not theft in the traditional sense because the models themselves aren't copied, but it does transfer capability from a frontier lab that invested billions in training to a competitor that invested a fraction of that cost. Anthropic's terms of service explicitly prohibit using Claude to train competing models, giving the company a contractual claim even where copyright law remains unclear.

How Did xAI Keep Training After Access Was Cut Off?

After Anthropic revoked xAI's official access to Claude in January 2026, the company didn't stop. Instead, xAI engineers routed requests through personal Claude accounts and an intermediary service called Blackbox AI, according to The Information. This workaround allowed the distillation pipeline to continue undetected for months.

Musk has previously acknowledged using competitor models for training. In court, he admitted that xAI "partially" used OpenAI models to train Grok, calling the practice "industry standard." However, the scale and persistence of the Claude distillation effort, combined with the deliberate circumvention of Anthropic's access revocation, suggests a more aggressive approach than Musk's public statements indicate.

What Evidence Shows This Was Widespread?

Anthropic has been fighting distillation attacks for months. Earlier in 2026, the company disclosed that it detected "industrial-scale distillation attacks" targeting Chinese AI labs including DeepSeek, Moonshot AI, and MiniMax. These attacks involved over 24,000 fraudulent accounts that generated more than 16 million exchanges with Claude.

The xAI case is different in a crucial way. It's not a Chinese lab scraping through fraudulent accounts; it's a well-funded American competitor with direct ties to the current administration. Musk's political proximity to President Trump adds complexity: xAI is simultaneously positioning itself as a national champion in AI while using a competitor's models as training data.

How to Evaluate AI Coding Tools Based on Training Data Provenance

• Check the source of training data: Ask vendors whether their models were trained on outputs from competing frontier models, or whether they used only their own proprietary data and public sources. This matters because distilled models may inherit both strengths and limitations from their source models.
• Understand the terms of service: Review whether the tool's training complies with the terms of service of any models it may have learned from. Breach of contract claims can affect long-term viability and support.
• Assess capability gaps: If a coding tool's capabilities were partly built on another model's outputs, the tools are less differentiated than branding suggests. Test whether the tool keeps pace with improvements in the source model or gets left behind over time.

What's the Irony in Musk's Broader AI Strategy?

While xAI was distilling Claude's outputs, Musk's broader compute empire took a different path. The GPUs Musk famously stockpiled are now being rented out to Anthropic via SpaceX's Colossus-1 data center, a facility providing 220,000 GPUs for Claude training. Google is also paying SpaceX $920 million per month for AI compute from the same infrastructure.

This creates a peculiar situation: xAI was training on Claude's outputs while SpaceX was profiting from providing the compute that powers Claude. The arrangement highlights how interconnected the AI industry has become, even as companies compete fiercely over model capabilities and training data.

What Internal Problems Does xAI Face?

Beyond the distillation controversy, xAI's own model development appears troubled. The pretraining team shrank to fewer than five people, and four Grok code leads left within months. These departures are part of a broader wave of co-founder exits tied to safety concerns and frustration over Grok's failure to close the gap with frontier models like Claude and GPT-4. One employee accidentally deleted critical training data, costing two to three weeks of work.

These internal challenges suggest that even with access to Claude's outputs, xAI has struggled to build a competitive coding model. The distillation effort may have been an attempt to compensate for slower progress on original model development.

What Does This Mean for the Future of AI Competition?

The xAI revelation makes explicit what many in the industry have suspected: model distillation is widespread, and the line between learning from public outputs and copying a competitor's capabilities is blurry at best. Frontier labs are now treating their model outputs as proprietary training data worth protecting, signaling that the next front in AI competition won't just be about who has the most GPUs.

Anthropic has already shown willingness to enforce its terms. The company cut off DeepSeek, Moonshot AI, and MiniMax after detecting their distillation operations. Whether it takes the same action against Musk's company, given the political complications and the fact that SpaceX is now a major compute supplier to Anthropic, remains an open question. The answer could reshape how frontier labs protect their intellectual property and whether distillation becomes a standard competitive tactic or a legally and contractually prohibited practice.