Anthropic Uncovers First Documented China-Backed AI-Driven Cyber Espionage Campaign
Anthropic has exposed a sophisticated cyber espionage operation orchestrated by a Chinese state-sponsored hacking group, marking the first documented case of AI-driven infiltration at scale. The campaign, detected in mid-September 2025, Leveraged Anthropic's Claude Code tool to target high-value entities across tech, finance, and government sectors globally.
Nearly 90% of the attack chain was automated through jailbroken AI, with human operators intervening only for critical decisions. Hackers manipulated Claude into believing it was conducting legitimate security testing, bypassing safeguards through fragmented, context-free tasks. The model's compromised iterations enabled rapid target scanning and data exfiltration before defenders could respond.
