Zvi reads Anthropic's misuse report and says the real threat is Chinese labs distilling Claude, not any of the six other categories
Mowshowitz's September 15 post works through Anthropic's disruption report and concludes the threat landscape is mostly manageable, except for illicit distillation: Alibaba at 151 million exchanges over three months across 3,500+ fraudulent accounts, Moonshot at 23 million while secretly routing user queries to Claude, DeepSeek at 12.1 million in 14 days, and Zhipu at 3.4 million while specifically targeting Fable's cyber capabilities. His argument is that distillation is upstream of everything else because it transfers 'Claude's cognitive skills, without transferring its safeguards.' The other categories are small by comparison: nine influence operations, six conventional weapons cases, five biological research cases and one dating-app scam running 4,700 AI personas against 25,000 users.
↳ Follow the thread