THE ESSENTIALS IN 30 SECONDS
- Starting October 17, the group used ARTEX against more than 200 organizations in at least 14 countries over an 11-day window.
- CrowdStrike confirmed post-exploitation activity at nine organizations and password spraying or credential scanning at at least 12 others.
- The agents ran mainly on DeepSeek plus an unidentified commercial model, with a human operator steering them in Chinese.
CrowdStrike published research on October 29 describing what it calls a step up in AI-assisted hacking: a China-nexus espionage group, which it tracks as VAPOR PANDA, running many intrusions at the same time through an agentic framework called ARTEX. The company's OverWatch threat-hunting team says this is the first time it has seen the group use AI-driven intrusion tooling.
What CrowdStrike observed
According to the report, activity began on October 17 and, over 11 days, the group used ARTEX against more than 200 organizations in at least 14 countries. Post-exploitation activity was confirmed at nine of them, and at least 12 more faced password spraying or credential scanning. Targets included foreign ministries in Latin America, Europe, Asia and the Middle East, a Middle Eastern interior ministry, US semiconductor, defense, pharmaceutical, nuclear and financial organizations, think tanks, and telecom operators in the US, Belgium and the Philippines.
ARTEX works through a lead agent that coordinates planner and worker agents over a shared record of facts, findings and goals, moving each engagement from initial access through reconnaissance and lateral movement and trying another route when one fails. CrowdStrike says the agents mainly used DeepSeek, along with a second, subscription-based commercial model it could not identify. In a single intrusion, the planning agents consumed about 1.4 billion input tokens.
Still a human-run operation
The framework was not fully autonomous. A human operator issued instructions in Chinese at key decision points, at one point telling the agents to find Windows machines lacking CrowdStrike's own Falcon sensor so later activity could focus on unprotected hosts. Techniques included exploiting an unsigned SAML assertion and an XML external entity flaw to obtain cloud credentials, web shells, and reuse of old leaked keys and VPN configurations to reach a Greek government network. At two ministries the group deployed BLOODALCHEMY, a backdoor first documented in 2023. CrowdStrike notes the campaign is separate from ARTEX activity Microsoft has reported under the name Storm-0940.
Why it matters
The findings come from a single vendor and have not been independently confirmed, and CrowdStrike does not name the victims. Its broader assessment is that China-nexus actors are highly likely to keep adopting frameworks like ARTEX over the next year, pairing them with commercial models so small teams can run more intrusions. It advises rotating signing keys and credentials after any suspected exposure, including in old backups, enforcing SAML signature checks, requiring IMDSv2 on cloud instances, and deploying endpoint detection on internal servers too.
