Anthropic Accuses Chinese AI Labs of Stealing Claude via 100,000s of Fake Accounts
Updated
Updated · CNBC · Sep 3
Anthropic Accuses Chinese AI Labs of Stealing Claude via 100,000s of Fake Accounts
3 articles · Updated · CNBC · Sep 3
Summary
Anthropic said Chinese AI labs, including Moonshot AI, illegally distilled its Claude models by creating tens of thousands to hundreds of thousands of fraudulent accounts and using stolen payment and infrastructure.
Moonshot's Kimi K3 — a cheaper frontier-level model that gained traction in July — was trained on Anthropic's newest Claude version, Anthropic threat chief Jacob Klein told CNBC, calling it part of an illicit ecosystem evading controls at scale.
Anthropic said it has also flagged DeepSeek, MiniMax and Alibaba, while OpenAI and Google have reported similar distillation threats; the accused companies did not respond to CNBC requests for comment.
The allegations land as Anthropic, a 5-year-old company valued near $1 trillion and expected to go public as soon as October, pushes for tougher action on a practice the Trump administration called unacceptable in an April memo.
Cybersecurity experts said the same tactics also come from Iran, Russia and North Korea, often via dark-web markets for stolen cards and compromised AI accounts, raising surveillance and biological-weapons concerns.