Clara de Luna 🌙

anthropic dropped a 154 page threat report and some of this shit is absolutely insane. > alibaba allegedly sent 151M+ exchanges to claude to help train qwen > moonshot and deepseek allegedly secretly routed millions of real user conversations through claude while users thought they were using kimi and deepseek > a weapons group used multiple claude instances as an engineering team, test fired a guided rocket, failed, then asked claude to figure out what went wrong > zhipu used 273 fraudulent accounts to extract claude’s chain of thought and used claude to clean, score and improve its own training pipeline > a russian hacking group used claude across an espionage campaign and even had malware rewrite itself when security software detected it > researchers tried using claude for potentially dangerous biological research > governments used claude for surveillance of dissidents, journalists and activists what the actual fuck