Vapor Panda Uses AI Framework ARTEX to Attack 200+ Organizations
Chinese state-sponsored group Vapor Panda has leveraged the AI-powered ARTEX penetration framework to compromise over 200 organizations globally since October 2026, automating large-scale intrusions.

Chinese state-sponsored hackers from the group Vapor Panda have used an AI-powered penetration testing framework called ARTEX to target more than 200 organizations worldwide. The campaign began on October 17, 2026, and represents a significant escalation in automated cyber operations.
ARTEX is a Go-based framework that uses large language models, including DeepSeek and a subscription-based commercial model, to coordinate an army of planner and worker agents. These AI agents work together to move from initial access to reconnaissance and lateral movement with minimal human oversight. In one analyzed intrusion, the planning agents alone consumed approximately 1.4 billion input tokens, highlighting the massive scale of the automated activity. Human operators primarily intervene to steer the agents toward objectives and to help bypass endpoint defenses.
Attack Techniques and Tactics
The group combines AI automation with established attack methods. Analysis of attack transcripts shows ARTEX was used to chain an unsigned SAML assertion with an XML external entity flaw to query cloud metadata services. The hackers also deployed web shells, abused an unauthenticated orchestration service for root access in container environments, harvested LAPS passwords, and leveraged resource-based constrained delegation to impersonate privileged accounts.
In one instance, agents used authentication keys that had not been rotated since 2022 to forge an administrator token and read mailboxes. They also retrieved VPN configurations from a user's sent mail and used a VPN client key from a May 2024 email leak to gain internal network access. The broad adoption of ARTEX among threat actors has prompted its developer, Li Puhua, also known as Autumn-27, to take the framework closed source.
Global Impact and Targets
Over 200 organizations across multiple continents and sectors have been compromised. Post-exploitation activity, including the deployment of the BLOODALCHEMY backdoor, has been confirmed at nine entities. Vapor Panda has also conducted password-spraying attacks against at least 12 additional organizations, including U.S. think tanks and European government bodies.
Notable confirmed targets include:
Another group, tracked as Storm-0940, has used ARTEX since September 2026 to attack hundreds of organizations, including automakers, financial institutions, government agencies, and semiconductor manufacturers.
Emerging Threat Landscape
The disclosure of the ARTEX campaign coincides with the identification of four previously undocumented China-linked state-sponsored groups: Hollow Panda, Overcast Panda, Titan Panda, and Veiled Panda. These groups target global sectors for intelligence gathering and intellectual property theft.
Overcast Panda specializes in close-access intrusions, deploying a proprietary backdoor called FlowCloud onto devices via physical access when targets are in China. FlowCloud uses DLL sideloading, evades detection, and communicates through legitimate cloud infrastructure to provide persistent remote access for keylogging, screen capture, file collection, and credential harvesting.
Titan Panda has been active since at least 2021, targeting legal firms. Since April 2026, the group has exploited a hard-coded Internet Information Services machine key vulnerability in Intapp Time Server to achieve remote code execution. This led to the deployment of an in-memory framework called IceGiant for reconnaissance, persistence, lateral movement, and data exfiltration. It is suspected that Titan Panda obtained the default machine keys through the compromise of a legal entity running the vulnerable software.
Security researchers continue to monitor the evolution of the ARTEX framework as state-sponsored threat actors increasingly adopt AI tools for scalable intrusion campaigns.





