Anthropic Vs Musk: AI Rivalry Escalates over Data Ethics and Model Theft

Fresh tensions have erupted in the artificial intelligence (AI) sector between Anthropic and billionaire Elon Musk over data practices and ethics. The situation began when the Dario Amodei-led company accused three Chinese AI companies of conducting distillation attacks, a method where outputs from Anthropic's flagship Claude model were allegedly extracted via millions of interactions using thousands of fraudulent accounts to train rival systems.

In a blog post, the company said, "We've identified industrial-sclae distoillation attacks on our models by DeepSeek, MoonShot AI and MiniMax. These labs created over 24,000 fraudulent accounts and generated over 16 million exchanges with Claude, extracting its capabilities to train and improve their own models."

Anthropic described the activity as improper and potentially dangerous, claiming it could strip away safeguards designed to prevent misuse.

Musk Hits Back

This disclosure drew immediate fire from Musk, whose own venture, xAI, operates the Grok chatbot in direct competition with Claude. In a post on X that referenced community-sourced notes, Musk accused Anthropic of being a hypocrite.

He said, "Anthropic is guilty of stealing training data at massive scale" and quoted claims that said the company settled a $1.5 billion lawsuit related to the training of Claude AI. The Community Notes thread referenced the past legal settlements tied to claims that Anthropic used copyrighted books and other material to train its models without permission.

The post, amplified by users online, have reignited scrutiny of how proprietary AI models are trained and where the line lies between legitimate industry practice and improper data use.

Calling out Anthropic, Tory Green, Co-founder of AI infrastructure firm IO.Net said, "You trained on the open internet and then call it 'distillation attacks' when others learn from you. Labs that like to preach 'open research' suddenly crying about open access. This is what happens when intelligence sits behind a centralized api with subsidized tokens [sic]."

What Is Distillation?

The distillation technique is a widely used AI technique in which a smaller or less advanced model is trained using the outputs of a more powerful system. Companies commonly apply this method internally to make their own models more efficient and cost-effective.

However, Anthropic argues that the issue arises when rival firms use the same approach to replicate capabilities and potentially bypass years of original research and development.

According to Anthropic, the alleged activity was neither casual nor experimental, but highly coordinated. The company claimed that the Chinese AI labs relied on fake accounts, proxy networks and synchronised usage patterns to access Claude at scale while avoiding detection.

The stated objective, Anthropic said, was to extract high-value capabilities such as advanced reasoning, coding skills and tool use; features considered among the most commercially and strategically important aspects of the model.

Anthropic further detailed that each lab appeared to run its own campaign. Moonshot AI allegedly conducted more than 3.4 million interactions with Claude. MiniMax's activity was even larger, exceeding 13 million exchanges. DeepSeek’s campaign was smaller, with over 150,000 interactions, but, according to Anthropic, included attempts to elicit Claude's step-by-step reasoning processes, which could then be used to train other AI systems.

The company also cautioned that models developed through unauthorised distillation may not retain critical safety guardrails. It warned that such systems could pose national security risks if deployed in cyber operations, surveillance activities or offensive AI applications.