Chinese State-Sponsored Group Vapor Panda Leverages AI Framework ARTEX for Global Attacks
Chinese state-sponsored threat actors, identified as Vapor Panda, are using an AI-powered Go framework named ARTEX to conduct simultaneous attacks against over 200 high-value organizations worldwide.
Chinese state-sponsored threat actors, tracked as Vapor Panda, are employing a sophisticated AI-powered penetration testing framework called ARTEX to conduct large-scale, simultaneous attacks against over 200 high-value organizations globally. CrowdStrike, which attributes the activity to a China-based adversary also known as VexatiousTemper and linked to Trend Micro's Earth Estries cluster, highlighted that the use of agentic tooling like ARTEX enables attackers to pursue a wider target set with greater speed than manual operations would allow.
The campaign, which began on October 17, 2026, has targeted a diverse range of entities, including foreign ministries in Latin America, Europe, and Asia, as well as semiconductor, defense, and think tank organizations in the U.S., and telecommunications providers across multiple regions. Post-exploitation activity has been confirmed at nine organizations, where attackers deployed BLOODALCHEMY, a derivative of the ShadowPad malware, followed by credential spraying, network scanning, and reconnaissance. At least 12 additional organizations experienced password-spraying and credential-scanning activity.
The core of this operation lies in ARTEX's integration with DeepSeek and an undisclosed commercial large language model (LLM). This combination powers a system of planner and worker agents that autonomously coordinate to achieve objectives, moving from initial access to host reconnaissance and lateral movement. A human operator oversees the process, guiding the agents to perform tasks such as writing web shells, extracting domain administrator passwords, and enumerating endpoint sensor coverage. CrowdStrike noted that in a single intrusion, Vapor Panda's planning agents consumed approximately 1.4 billion input tokens, indicating the immense scale of the automated activity.
ARTEX facilitates concurrent engagements, allowing a single operator to manage multiple intrusions simultaneously, a significant departure from campaigns focused on individual breaches. The human operator intervenes to steer the agents and bypass endpoint defenses, but the bulk of the operational execution is automated. This approach significantly lowers the technical barrier for conducting complex intrusion campaigns, as observed by Zscaler ThreatLabz, which noted ARTEX's broad adoption among threat actors.
CrowdStrike detailed several attack paths observed during these engagements. These include chaining an unsigned SAML assertion with an XML external entity (XXE) flaw to query the cloud Instance Metadata Service, leading to the acquisition of administrative commands and privilege escalation. Other methods involved deploying web shells, abusing unauthenticated orchestration services for root access in container environments, reading Local Administrator Password Solution (LAPS) passwords, and leveraging resource-based constrained delegation. In one instance, attackers used unrotated authentication keys from 2022 to forge administrator tokens and read mailboxes, subsequently using recovered VPN configurations to access internal networks.
The breadth of attack vectors employed by the agents, spanning web applications, cloud services, container environments, Active Directory, and mail/VPN infrastructure, demonstrates a highly adaptive and resilient operation. When one attack path failed, the agents shifted to alternative approaches, enabling a single operator to cover more ground than would be feasible manually. This adaptability is a hallmark of advanced persistent threats leveraging AI capabilities.
The development and widespread adoption of ARTEX have prompted its developer, Li Puhua (known as "Autumn-27"), to move the framework to a closed-source model. This move underscores the perceived value and effectiveness of such AI-driven tools in modern cyber warfare. The disclosure of Vapor Panda's activities also coincides with the emergence of four other previously undocumented China-linked state-sponsored threat groups: Hollow Panda, Overcast Panda, Titan Panda, and Veiled Panda, indicating a broader trend of sophisticated, state-backed cyber operations originating from China.