Elon Musk criticises Anthropic over allegations about AI model training practices

Elon Musk is criticising Anthropic after it accused Chinese AI firms, including DeepSeek, of scraping millions of answers from Claude to train their own AI. Musk has been quite active on the microblogging platform, X (formerly Twitter), firing at Dario Amodei-led company, saying that it is stealing data to train its Claude AI model.

Anthropic has shared a detailed blog post accusing three Chinese AI firms, DeepSeek, Moonshot AI, and MiniMax for training their AI models on Claude’s responses. The company claims that these Chinese firms have created over 24,000 fraudulent accounts and have generated over 16 million exchanges with Claude, in efforts to refine their own models.

Soon after Anthropic's claims, Musk shared a post on X, on February 23, saying, “Anthropic is guilty of stealing training data at a massive scale and has had to pay multi-billion dollar settlements for their theft. This is just a fact.” 

Anthropic is guilty of stealing training data at massive scale and has had to pay multi-billion dollar settlements for their theft. This is just a fact. https://t.co/EEtdsJQ1Op

To support claims, Musk shared screenshots of X Community Notes that say Anthropic previously settled a $1.5 billion lawsuit tied to Claude and alleged the firm training the model stole data. 

The note said, “Anthropic settled a $1.5B lawsuit for pirating 7M+ books from shadow libraries to train Claude and faces a $3B suit for torrenting 20,000+ songs. They built their models on stolen creative work, then complain when others extract their outputs.”

In addition to directly accusing Anthropic, Musk has been resharing multiple posts on X that report similar allegations, including sharing posts of people urging users not to trust the company. The remarks appear to be intentional, implying that Anthropic should look into its own practices before accusing others of misusing its data.

However, distillation attacks have become a common concern for AI companies. This practice is being used to train a less capable model on the outputs or capabilities of a stronger model. Several tech giants, including OpenAI and Google, have reported instances of suspected misuse. In addition, companies are also strengthening safeguards to prevent unauthorised extraction of AI models.