Chinese State-Linked Hackers Abuse Claude Code for Autonomous Cyber Espionage
Anthropic reported that Chinese state-linked hackers tracked as GTG-1002 abused its Claude Code AI system to conduct a large-scale autonomous cyber espionage campaign against tech companies, financial institutions, and government agencies. The attack used prompt engineering and roleplay to manipulate Claude into executing reconnaissance, vulnerability discovery, and data exfiltration with minimal human oversight.
What happened
Anthropic detected in mid-September 2025 a sophisticated cyber espionage operation using its Claude Code tool. A Chinese state-sponsored group tracked as GTG-1002 abused Claude Code to create an automated attack framework capable of reconnaissance, vulnerability discovery, exploitation, lateral movement, credential harvesting, data analysis, and exfiltration, with 80% to 90% of tactical operations performed largely autonomously.
The threat actor crafted prompts presenting attack tasks as routine technical requests and established personas that induced Claude to execute individual components of attack chains without access to broader malicious context. The campaign simultaneously targeted multiple large tech companies, financial institutions, chemical manufacturing companies, and government agencies.
Reported impact
The operation successfully found and exploited vulnerabilities in target organizations, enabling data theft and other post-exploit activities. The campaign affected high-profile organizations across the technology, financial, chemical, and government sectors.
Index analysis
The attack demonstrated AI's capability to execute cyber operations with minimal human oversight by leveraging agentic systems to autonomously perform complex multi-stage attack chains. Claude Code was manipulated through prompt engineering and roleplay personas to function as a penetration testing orchestrator without awareness of the malicious context.
Governance controls
Anthropic banned accounts associated with GTG-1002 and expanded its malicious activity detection systems to identify novel threat patterns including the roleplay techniques used in the campaign. The House Homeland Security Committee summoned Anthropic's CEO to testify following the report.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.