CrowdStrike says a China-linked group used AI agents against 200 organisations
VAPOR PANDA ran the ARTEX framework mainly on DeepSeek across at least 14 countries in 11 days from 17 October; CrowdStrike confirmed follow-on access at nine organisations.
- Security & misuse
- Notable
CrowdStrike reported that VAPOR PANDA, a China-nexus intrusion group it tracks, had used ARTEX, an agentic penetration-testing framework, against more than 200 organisations in at least 14 countries. CrowdStrike’s OverWatch team saw the activity begin on 17 October 2026 and observed it over 11 days. CrowdStrike says the group has mainly pursued long-term access to government, think-tank, energy and extractive-sector organisations since 2021.
According to CrowdStrike, ARTEX uses a main agent that coordinates subordinate planner and worker agents, which share a graph of facts, findings and goals and carry an engagement from initial access through host reconnaissance to lateral movement. A human operator stayed in the loop and steered the agents at decision points. The group ran the framework mainly on DeepSeek’s models, supplemented by a second, subscription-based commercial model that CrowdStrike said it could not confirm. Planning agents in a single intrusion consumed, in its words, “on the order of 1.4 billion input tokens”.
The targets, CrowdStrike said, included foreign ministries in Latin America, Europe, Asia and the Middle East, US semiconductor, defence and think-tank organisations, and telecoms providers in the US, Belgium and the Philippines. It confirmed post-exploitation activity at nine organisations, where the group deployed the BLOODALCHEMY backdoor, and said password spraying and credential scanning reached at least 12 more. The operator told the agents to look for Windows machines without CrowdStrike’s own Falcon sensor so that follow-on activity could concentrate on unprotected hosts. CrowdStrike linked the ARTEX controller to VAPOR PANDA infrastructure; the controller intermittently exposed tooling, target files and agent transcripts.
Several judgements are hedged: CrowdStrike said the group “likely compromised” a third Latin American ministry, and described the multi-provider toolchain as one that “likely reflects” a deliberate choice. It also said the campaign is separate from the ARTEX activity Microsoft reported under the Storm-0940 designation.
The report is the third CrowdStrike account in October of attackers using agents, after the South Korean bank breaches and its survey of four groups. It differs in the actor: a group CrowdStrike describes as a long-running China-nexus espionage adversary, rather than the unnamed operators in those two reports. Microsoft described the framework spreading further a week later.