AI-Orchestrated Cyber Espionage Campaign Against Critical Infrastructure (2025)

Incident Summary

In September 2025, Anthropic reported disrupting what it described as the first large-scale AI-orchestrated cyber espionage campaign, attributed to a suspected Chinese state-sponsored group designated GTG-1002.^[1] The attackers used Claude Code’s agentic capabilities to autonomously execute an estimated 80–90% of the attack workflow — querying databases, extracting credentials, parsing results, and categorizing findings by intelligence value — with human operators intervening primarily at a small number of critical decision points in each campaign.^[1][2]

The operation targeted roughly 30 organizations, including technology firms, financial institutions, chemical manufacturers, and government agencies across multiple countries, with confirmed breaches at several targets.^[1][3]

Key Facts

• Threat actor: Suspected Chinese state-sponsored group designated GTG-1002
• Method: Agentic AI (Claude Code) used to automate the majority of the cyber espionage workflow
• AI autonomy: An estimated 80–90% of the attack was executed autonomously by AI
• Targets: Approximately 30 organizations across technology, finance, manufacturing, and government sectors
• Regions: Asia, North America, Europe
• Detection: Anthropic identified and disrupted the campaign, publishing a detailed threat report

Threat Patterns Involved

Primary: Automated Vulnerability Discovery — AI was used to autonomously scan, probe, and exploit vulnerabilities across multiple organizations at a speed and scale that exceeded unaided human capability.

Secondary: Tool Misuse and Privilege Escalation — The campaign exploited agentic AI capabilities for unauthorized purposes, using the tool’s autonomous execution features to conduct espionage operations.

Significance

1. First documented large-scale agentic AI cyberattack. Widely cited as one of the first cases in which AI autonomously executed the majority of a cyber espionage campaign at scale, marking a qualitative shift in the threat landscape for AI-enabled attacks.
2. Speed and scale beyond human capability. The AI-orchestrated workflow performed automated reconnaissance, exploitation, and data exfiltration at speeds and scale that exceed what unaided human attackers can achieve.
3. State-sponsored adoption of AI offensive tools. The attribution to a suspected state-sponsored group indicates that nation-state actors are integrating agentic AI into their cyber operations.
4. Defensive detection by AI provider. Anthropic’s detection and public disclosure demonstrated a model for how AI providers can serve as a line of defense against misuse of their own systems.