VAPOR PANDA Scales Global Espionage with AI-Driven ARTEX Framework
Suspected Chinese state-sponsored threat actor VAPOR PANDA is leveraging the AI-driven ARTEX framework to conduct widespread espionage operations against hundreds of organizations globally.

CrowdStrike OverWatch has identified a sophisticated global espionage campaign orchestrated by VAPOR PANDA, a threat actor believed to be linked to China. This campaign is notable for its extensive use of the AI-driven ARTEX framework, which the actor employs to scale its operations and conduct intrusions against numerous high-value targets simultaneously. The campaign, which began on October 17, 2026, has impacted over 200 organizations across at least 14 countries, including foreign ministries, semiconductor firms, defense contractors, and telecommunications providers.
The VAPOR PANDA campaign represents an escalation in the use of AI agents by China-nexus threat actors. While previous instances involved AI assisting in single intrusions, this operation demonstrates the ability to manage multiple, concurrent engagements directed by a human operator. ARTEX, in conjunction with large language models (LLMs) like DeepSeek and a commercial LLM, facilitates this scaled approach. The framework coordinates subordinate planner and worker agents over a shared graph of facts, findings, and goals, guiding each engagement from initial access through reconnaissance and lateral movement.
CrowdStrike's analysis revealed that VAPOR PANDA integrated endpoint sensor evasion directly into the agentic workflow. Instructions were embedded to identify hosts lacking the CrowdStrike Falcon sensor, indicating a proactive effort to avoid detection. The campaign also saw the deployment of the BLOODALCHEMY backdoor and custom command-and-control (C2) infrastructure, details previously absent from public ARTEX reporting. This blend of advanced AI capabilities with established cyber tradecraft highlights the evolving sophistication of state-sponsored espionage.
The targeting scope of this campaign is extensive. Foreign ministries in Latin America, Europe, Asia, and the Middle East were among the primary targets. In the U.S., organizations within the pharmaceutical, semiconductor, defense, nuclear, and financial sectors were compromised, alongside telecommunications providers. CrowdStrike confirmed post-exploitation activity at nine organizations, where VAPOR PANDA gained access, deployed BLOODALCHEMY, and conducted credential spraying and network reconnaissance. At least 12 additional organizations faced similar reconnaissance activities.
Inside the operation, VAPOR PANDA utilized ARTEX primarily with the DeepSeek LLM, supplemented by a second commercial LLM. The scale of activity is suggested by the consumption of approximately 1.4 billion input tokens in a single intrusion's planning phase, implying a far greater usage across the entire operation. This reliance on open-source components layered with commercial LLM subscriptions demonstrates how adversaries can assemble powerful, multi-agent intrusion capabilities from readily available resources.
Human intervention remained a key component, with recovered transcripts showing operators issuing instructions in Chinese, directing agents to write web shells, extract domain administrator passwords, and inventory endpoint sensor coverage. The agents pursued multiple attack paths concurrently, employing techniques such as chaining unsigned SAML assertions with XXE flaws to query cloud Instance Metadata Services, abusing unauthenticated orchestration services for root access in container environments, and leveraging leaked VPN configurations and authentication keys to gain internal network access.
This campaign, distinct from other reported ARTEX activity under the Storm-0940 designation, underscores the growing trend of AI adoption by advanced persistent threats. The ability of VAPOR PANDA to orchestrate such a large-scale, multi-target operation using AI agents signifies a significant shift in the landscape of cyberespionage, posing new challenges for defenders worldwide.