Anthropic has accused Chinese AI labs, including DeepSeek, Moonshot AI, and MiniMax, of creating 24,000 fraudulent accounts and executing 16 million prompts to enhance their own models through what they describe as industrial-scale distillation attacks. This method, while legitimate in principle, involves smaller AIs learning from the high-quality outputs of larger AIs. To mask their activities and evade detection, these labs allegedly constructed proxy networks known as hydra clusters, which rotate IP addresses and user profiles to blend nefarious requests with normal traffic. Such practices are being closely monitored by leading AI labs to safeguard against abuse of advanced AI models.
Anthropic accuses Chinese AI labs of large-scale distillation attacks
