Twelve days after OpenAI went to lawmakers, Anthropic joined it. In late February 2026 Anthropic accused three Chinese laboratories (DeepSeek, Moonshot AI and MiniMax) of running concerted distillation campaigns against Claude.
The numbers
As reported, the campaigns involved more than 16 million API exchanges conducted through over 24,000 fraudulent accounts, with models queried using specially constructed prompts designed to elicit trainable responses. Anthropic described the activity as industrial in scale and coordinated rather than opportunistic.
These figures are distinct from, and earlier than, the 28.8 million exchanges attributed to Alibaba’s Qwen lab in Anthropic’s June letter to the Senate Banking Committee. The two disclosures concern different parties and different windows, and are frequently conflated in secondary coverage.
The context it landed in
The accusations arrived while Washington was actively debating how strictly to enforce export controls on advanced AI chips. That timing was not incidental. If capability can be transferred by querying an API, the case for hardware restrictions looks weaker on its own and stronger as part of a broader regime, which is precisely the argument the frontier labs were making.
What the disclosure established
The account and query figures are the strongest form of evidence produced in this dispute to date, because they are observations of a company’s own infrastructure rather than inferences about somebody else’s model. They establish that large-scale extraction occurred.
They do not establish what was built from it. That gap between demonstrating extraction and demonstrating a resulting model is the one that reopened in July and has still not been closed.