Claude Jailbroken by Chinese Hackers to Orchestrate First-of-Its-Kind AI Cyberattack

Anthropic has confirmed that Claude was used in a large-scale agentic cyberattack in September, marking the first known incident where an AI system carried out most of an operation with minimal human input. The company says a Chinese state-sponsored group jailbroke Claude by disguising malicious tasks as harmless subtasks. Claude then autonomously scanned networks, wrote exploits, stole data and documented breaches across about 30 global targets.