Chinese Threat Actors Deploy AI-Powered ARTEX Framework Against Global Targets
Vapor Panda utilizes agentic AI to automate large-scale cyber intrusions across 200 organizations.
1AI-Driven Cyber Intrusion Scale
Chinese state-sponsored group Vapor Panda is utilizing the Go-based ARTEX framework to execute cyberattacks. Since October 17, 2026, this AI-powered tool has targeted over 200 organizations globally. The framework enables a small team to manage multiple concurrent engagements, significantly expanding their operational reach beyond what manual penetration testing methods would traditionally support.
2Agentic Orchestration and LLMs
ARTEX integrates DeepSeek and an undisclosed commercial LLM to coordinate planner and worker agents. These agents autonomously perform reconnaissance, vulnerability identification, and exploitation tasks. A human operator oversees the process, directing agents to extract domain administrator passwords and deploy web shells while maintaining a focus on bypassing various endpoint security defenses.
3High-Value Global Target Set
The campaign impacts diverse sectors, including foreign ministries in Latin America, Europe, Asia, and the Middle East. U.S.-based semiconductor firms, defense contractors, and think tanks are also primary targets. Additionally, telecommunications providers across multiple regions have faced credential-spraying and network-scanning activities, highlighting the broad strategic scope of these coordinated state-sponsored intrusion operations.
4Post-Exploitation and Lateral Movement
Confirmed post-exploitation activity at nine organizations involved the deployment of BLOODALCHEMY, a derivative of ShadowPad. Attackers utilized this framework for credential spraying, network scanning, and reconnaissance. Planning agents consumed approximately 1.4 billion input tokens in a single intrusion, demonstrating the massive scale of automated activity required to facilitate lateral movement within compromised networks.
5ARTEX Adoption and Development
The ARTEX framework has gained rapid popularity among threat actors, leading developer Li Puhua to transition the project to a closed-source model. Security researchers note that the tool lowers technical barriers for complex campaigns by orchestrating multiple LLM backends. This shift reflects a broader trend of weaponizing autonomous agents for sophisticated cyber espionage.
6Emergence of New Panda Groups
The disclosure coincides with the identification of four additional China-linked groups: Hollow Panda, Overcast Panda, Titan Panda, and Veiled Panda. These actors focus on intelligence gathering and intellectual property theft. Each group employs distinct methodologies, ranging from proprietary backdoors like FlowCloud to exploiting specific software vulnerabilities in legal billing systems for persistence.
7Sophisticated Backdoor Deployment Tactics
Overcast Panda utilizes the FlowCloud backdoor, deployed via direct physical access when targets travel to China. This tool enables persistent remote access, including keylogging and screen capture. The backdoor components are designed to resist forensic analysis and utilize legitimate cloud infrastructure for command-and-control communications, effectively concealing their presence from standard network monitoring tools.
8Legal Sector Vulnerability Exploitation
Titan Panda has targeted legal billing software by exploiting hard-coded machine keys in Intapp Time Server installations. Since April 2026, these attacks have enabled remote code execution and the deployment of the IceGiant in-memory framework. This allows for long-term data exfiltration and persistence within legal organizations across North America, operational since 2021.