Anthropic says it has foiled the first-ever AI-orchestrated cyber attack, originating from China — company alleges attack was run by Chinese state-sponsored group

Anthropic claims to have thwarted the first AI-orchestrated cyber attack, allegedly orchestrated by a Chinese state-sponsored group using Anthropic's agentic coding tool, Claude. The attack targeted 30 institutions across various sectors, with the AI tool running independently for around 80-90% of the operation. Despite built-in safeguards, the attackers circumvented them by deceiving Claude into believing it was conducting legitimate cybersecurity tasks. Anthropic's team detected the attack, banned illegal accounts, and shared their findings to help the industry develop defenses against such AI-driven threats.