A China-linked hacking group known as Vapor Panda has been observed using an artificial intelligence (AI)-powered penetration testing framework called ARTEX to conduct cyberattacks against more than 200 organizations worldwide.
According to cybersecurity company CrowdStrike, the campaign began on October 17, 2026, and targeted government agencies, telecommunications providers, defense companies, semiconductor manufacturers, and financial institutions across several regions.
The targets include foreign ministries in Latin America, Europe, Asia, and the Middle East, as well as 19 US-based organizations operating in the pharmaceutical, semiconductor, defense, nuclear, and financial sectors. Telecommunications providers in the US, Belgium, and the Philippines were also targeted.
ARTEX is a Go-based framework that uses AI agents powered by DeepSeek and an unidentified commercial large language model (LLM). The framework deploys multiple AI agents that work together, with some responsible for planning attacks while others perform technical tasks such as network scanning, host reconnaissance, and lateral movement.
A human operator supervises the activity and directs the agents to perform tasks such as creating web shells, extracting administrator passwords, and identifying endpoint security systems. This approach allows the attackers to conduct multiple operations simultaneously with less manual involvement.
CrowdStrike reported that, in one intrusion alone, ARTEX's planning agents consumed approximately 1.4 billion input tokens.
The attackers have used several techniques to gain access to targeted networks, including exploiting a weakness in Security Assertion Markup Language (SAML) authentication and an XML External Entity (XXE) vulnerability to access cloud metadata services. The threat actors also abused container orchestration services, extracted Local Administrator Password Solution (LAPS) credentials, and used Active Directory delegation techniques to obtain higher privileges.
In nine confirmed cases of post-compromise activity, the attackers deployed the BLOODALCHEMY backdoor derived from the ShadowPad backdoor. Additional activities included password spraying, credential scanning, network reconnaissance, and attempts to move through compromised environments.
CrowdStrike said the AI agents could use several attack vectors at the same time and switch methods when an approach failed.
Vapor Panda, previously tracked as VexatiousTemper, was first identified in February 2021 and has been linked to the broader China-associated Earth Estries threat cluster.
In its August report, CrowdStrike detailed four previously undocumented China-linked state-sponsored threat groups tracked as Hollow Panda, Overcast Panda, Titan Panda, and Veiled Panda, focused on intelligence gathering and intellectual property theft.