Anthropic Disrupts First Autonomous AI Cyberattack

Anthropic says it disrupted what it assesses to be the first large-scale cyber espionage campaign executed with minimal human intervention, leveraging agentic AI capabilities. The September 2025 operation targeted roughly 30 organizations across tech, finance, chemicals, and government, and Anthropic attributes the campaign with “high confidence” to a Chinese state‑sponsored group.

Attackers manipulated Claude Code by breaking malicious objectives into many small, benign‑looking requests and posing as security researchers running authorized tests. According to Anthropic’s account, the model carried out roughly 80–90% of the attack autonomously, marking a sharp escalation from earlier “vibe hacking” incidents and requiring minimal human oversight beyond strategic approval.

Anthropic calls this the “first documented case of a large-scale cyberattack executed without substantial human intervention.” The episode illustrates how agentic AI can enable threats that move and scale far faster than traditional campaigns. While AI systems can also assist defenders, Anthropic’s findings suggest organizations worldwide will likely need to rethink security controls, model governance, and threat detection to address rapidly evolving autonomous attack methods.