China's Vapor Panda Hits Over 200 Organizations Using AI Framework ARTEX

Chinese state-sponsored hackers are using ARTEX, a Go-based, AI-powered penetration testing framework, to attack multiple high-value targets simultaneously. That is the finding of a CrowdStrike analysis published on Thursday.

CrowdStrike attributes the activity to Vapor Panda (formerly tracked as VexatiousTemper). The group was first identified in February 2021 and overlaps with the cluster Trend Micro calls Earth Estries.

Scale of the campaign

According to the researchers, starting October 17, 2026, Vapor Panda turned ARTEX against more than 200 organizations worldwide. Targets included foreign ministries in Latin America, Central America, Europe, Asia, and the Middle East; a Middle Eastern interior ministry; and a European country's foreign ministry, presidency, prime minister's office, and defense, justice, finance, and intelligence bodies. In the U.S., 19 organizations in pharmaceuticals, semiconductors, defense, nuclear, and finance were hit, along with telecom providers in the U.S., Belgium, and the Philippines.

Post-exploitation activity was confirmed at nine organizations: the attackers deployed BLOODALCHEMY, a ShadowPad derivative, then moved on to credential spraying, network scanning, and reconnaissance. At least 12 more organizations, including U.S. think tanks and European government entities, faced password-spraying attempts.

How ARTEX works

The framework pairs DeepSeek with an as-yet-unidentified subscription-based commercial language model. It runs planner and worker agents that coordinate everything from initial access to host reconnaissance and lateral movement. A human operator merely steers them, directing the agents to write web shells, extract domain administrator passwords, or map which endpoints are covered by security sensors.

In a single intrusion, the planning agents consumed around 1.4 billion input tokens. "Agentic tooling like ARTEX enables this scale of expansion at speed, allowing a small team to pursue a wider target set than manual operations would support," CrowdStrike said.

Recovered transcripts show the agents blending AI with proven techniques: chaining an unsigned SAML assertion with an XXE flaw to query a cloud metadata service and ultimately recover an enterprise administrator account; abusing an unauthenticated orchestration service to gain root in a container environment; reading LAPS passwords. In one case, the agent was handed authentication keys unrotated since 2022, forged an admin token, read mailboxes, and then used VPN settings found in sent mail plus a key from a purported May 2024 leak to reach the internal network. When one path failed, the agents switched to another.

ARTEX spread quickly among threat actors, prompting its developer Li Puhua (Autumn-27) to take it closed source. Zscaler ThreatLabz noted earlier that the tool significantly lowers the technical barrier for complex campaigns.

More pandas

CrowdStrike also detailed four previously undocumented China-linked groups: Hollow Panda, Overcast Panda, Titan Panda, and Veiled Panda. Overcast Panda plants its FlowCloud backdoor on victims' devices through physical access while they are in China; the backdoor supports keylogging, screen capture, and file and credential collection. Since April 2026, Titan Panda has been targeting law firms via a hard-coded IIS machine key in Intapp Time Server, deploying the in-memory IceGiant framework; researchers believe the group has been active since at least 2021.

Source: The Hacker News