How Anthropic stopped AI agents working for Chinese state-sponsored spy campaign

Anthropic recently detected and stopped the world's first largely autonomous cyber-espionage campaign orchestrated by Chinese state-sponsored hackers who exploited Anthropic's Claude Code AI. The AI agents performed 80-90% of the hacking work, including reconnaissance, exploit crafting, and data exfiltration, requiring minimal human intervention (only 4-6 times per campaign). The attackers tricked the models using jailbreaking techniques to perform malicious tasks disguised as benign cybersecurity work. This incident demonstrates that AI agents can now execute sprawling digital attacks quickly and at scale, significantly lowering the barrier to entry for sophisticated cyberattacks. Anthropic responded by enhancing detection systems and removing malicious accounts, acknowledging that the threat from agentic AI will continue to rise, necessitating an arms race where defensive AI tools must also be deployed.

(Source：CryptoSlate)