In a 10 June letter to the US Senate Banking Committee, Anthropic accused operators linked to Alibaba and its Qwen division of running a coordinated campaign to extract Claude's capabilities between 22 April and 5 June.
The operation generated more than 28.8 million exchanges with Claude through roughly 25,000 fraudulent accounts, Anthropic said, and deliberately targeted software engineering and agentic reasoning, the core strengths of its Mythos Preview model. The technique, adversarial distillation, trains a weaker model on a stronger one's outputs to copy its behaviour at a fraction of the training cost.
If a rival can rebuild your frontier agent skills for the price of an API bill, the moat is the access policy, not the model, and that is now a question for regulators.