Security Firm Reveals Hackers Abusing AI Penetration Framework ARTEX and DeepSeek in Global Espionage

Cybersecurity firm CrowdStrike reported that Chinese threat group Vapor Panda launched global cyber espionage operations beginning October 17, combining the open-source ARTEX penetration testing framework with DeepSeek and another commercial model to target over 200 organizations across at least 14 countries. Targets included diplomatic bodies in Latin America, Europe, Asia, and the Middle East, along with U.S. semiconductor, defense, pharmaceutical, nuclear, and financial institutions, as well as telecom operators in the U.S., Belgium, and the Philippines. Nine victim organizations were compromised with BloodAlchemy backdoors, password spraying, and network scanning activities. According to CrowdStrike, ARTEX acted as the primary orchestrator, allocating tasks to primary, planning, and worker agents based on reconnaissance data to advance lateral movement simultaneously. The planning agent alone consumed approximately 1.4 billion tokens in a single operation. The threat actors maintained hybrid oversight, manually submitting Simplified Chinese instructions directing agents to deploy backdoors, obtain domain admin credentials, and locate endpoints lacking endpoint security protection to focus attacks. Related incidents involving AI penetration frameworks were also noted in Taiwan, Japan, and South Korea.

Key takeaways

• CrowdStrike revealed Vapor Panda leveraged the ARTEX framework and DeepSeek to target over 200 organizations across 14 nations.
• ARTEX orchestrated multi-agent workflows across reconnaissance and lateral movement, consuming 1.4 billion tokens in planning tasks.
• Attackers maintained hybrid operation, using Simplified Chinese prompts to direct AI agents toward unprotected endpoints and deploy the BloodAlchemy backdoor.