GTG-1002

Summary

Chinese state-sponsored group (Anthropic's designation, high confidence) behind what Anthropic describes as the first reported largely AI-orchestrated cyber-espionage campaign. In mid-Sep 2025 it used Claude Code with tool integrations as an autonomous agent to carry out most of the intrusion lifecycle — reconnaissance, vulnerability discovery and exploitation, credential harvesting, lateral movement and data analysis — against roughly 30 large technology companies, financial institutions, chemical manufacturers and government agencies, with humans approving key steps. Anthropic validated a handful of successful intrusions, banned the accounts, notified affected parties and authorities, and noted the model sometimes hallucinated findings. The value for defenders is the tempo: request rates no human operator could sustain.

Attribution

Single-vendor assessment; no service named

Assessed by security researchers or MITRE ATT&CK; no government attribution on record.

• Anthropic (Research, 2025-11-13)
  High confidence: a Chinese state-sponsored group

Campaigns 1

Newest first. ASSESSED: tied to the actor with hedging or through an alias. CLAIMED: an allegation no independent source confirms.

MITRE ATT&CK techniques 8

Infrastructure patterns

• Claude Code agents orchestrated through MCP servers for remote command execution (Anthropic, 2025)
• Callback systems validate successful exploitation (Anthropic, 2025)

Reporting and advisories 1

• Anthropic
  Disrupting the first reported AI-orchestrated cyber espionage campaign