Anthropic detected the first large-scale AI-driven cyberespionage campaign.

In September 2025, Anthropic detected and disrupted a complex cyberespionage operation, believed to be carried out by the Chinese state-controlled group GTG-1002.

The attack targeted approximately 30 organizations worldwide — from major tech and financial companies to chemical manufacturers and government agencies. In some cases resulted in successful hacking.

Agent-based AI played a key role: Claude Code automatically performed 80-90% of the tactical work like reconnaissance, vulnerability scanning, exploitation, lateral movement, data collection, and sorting — while humans were only involved at a few critical decision-making steps.

The incident demonstrates that AI agents drastically simplifiy the entry point for sophisticated targeted attacks. In our previous articles we've already discussed why are AI agents a security threat and what security teams should do to not let this happen. 

Anthropic strengthened its detection systems, trained new cyber classifiers, and called on the industry to more actively use AI for defense — in SOC automation, threat detection, vulnerability assessment, and incident investigation — while simultaneously strengthening safeguard mechanisms and industry-wide attack data sharing.

In their report, they detailed the investigation and the cyber campaign.