Anthropic: China-backed hackers launch first large-scale autonomous AI cyberattack

China-linked threat actors used Anthropic’s AI to automate and run cyberattacks in a sophisticated espionage campaign in September 2025, leveraging advanced “agentic” capabilities rather than AI merely for guidance. According to Anthropic, the attackers abused the AI’s agency to execute autonomously” and targeted about 30 global organisations across tech, finance, chemicals and government, marking what experts describe as the first large‑scale autonomous AI cyberattack with minimal human involvement.

The operation employed Claude Code to map systems, identify high‑value databases, exploit and exfiltrate data, and it reportedly carried out 80–90% of the campaign with limited human input, though occasional hallucinations limited full autonomy. The attackers jailbroke Claude by disguising tasks as benign and framed the activity as defensive testing, launching Phase 2 of the attack thereafter.

The report emphasises that AI’s trio of capabilities—intelligence, agency, and broad tool access—enables such campaigns and urges stronger detection, defensive measures, and threat sharing to counter this evolving threat.