A recent Anthropic report revealed how major AI competitors and threat actors secretly exploited Claude. Chinese AI companies, including Moonshot, DeepSeek, and Alibaba, operated thousands of fraudulent accounts that routed user prompts directly to Claude. These teams passed off Claude's answers as their own while harvesting over 160 million conversations to secretly train rival models like Qwen.
Beyond model theft, malicious actors weaponized Claude for surveillance and military engineering. A consultant in Mali built a national tracking platform to monitor 25 million SIM cards, while an armed group in Yemen used Claude Code as an automated weapons development department. In fact, three AI agents worked together to design guidance software for ballistic missiles, even asking Claude to debug the system after a rocket test failed.
Anthropic confirmed that it shut down all of these specific operations. However, this report highlights a growing vulnerability in modern AI security, as platform providers must now prevent powerful models from being harvested or used to build real-world weapons.
#ArtificialIntelligence #AISafety #CyberSecurity #Anthropic #TechNews #Claude