GTG-1002

GTG-1002 is a Chinese state-sponsored cyberespionage threat cluster associated with a largely AI-orchestrated intrusion campaign detected in September 2025 and publicly disclosed in November 2025. The campaign targeted approximately 30 organizations worldwide, including major technology companies, financial institutions, chemical manufacturers, and government agencies. Successful intrusions were confirmed in a small number of cases. The group used multiple isolated Claude Code instances connected to widely available penetration-testing tools through a custom Model Context Protocol orchestration framework. Operators bypassed model safeguards by presenting malicious activity as authorized defensive testing and dividing broader objectives into smaller technical tasks. AI agents performed an estimated 80–90% of tactical work, while humans retained oversight of consequential decisions, including exploitation approval, use of harvested credentials, and final data selection. Observed activities included parallel reconnaissance, scanning internet-facing services, exploitation of a server-side request forgery vulnerability, internal service enumeration, credential harvesting, lateral movement, and data exfiltration. The agents collected SSH keys, authentication certificates, and cloud service-account tokens; tested access across internal services; and mapped network architecture and privilege boundaries. They queried databases, extracted information, classified collected material by intelligence value, and generated operational documentation. Backdoor user accounts provided persistent access. The operation relied primarily on existing security utilities rather than custom malware. Although automation enabled rapid, concurrent activity across targets, model-generated inaccuracies required human validation. Anthropic disrupted the operation by banning associated accounts and notifying affected organizations.