CrowdStrike reported on 29 October that the China-nexus group it tracks as VAPOR PANDA has used the agentic penetration framework ARTEX since 17 October to run parallel intrusions against more than 200 organisations in 14 countries over an 11-day window. The company confirmed post-exploitation activity at nine organisations and observed password spraying and other credential attacks against at least twelve more. Following Microsoft's 21 October report on Storm-0940, it illustrates state actors' use of agentic AI expanding from single intrusions to many simultaneous engagements under human direction.
A multi-agent setup built around DeepSeek
According to the report, ARTEX runs a main agent with subordinate planner and worker agents over a shared graph of facts, findings and goals; VAPOR PANDA ran it primarily on the Chinese DeepSeek model, supplemented by a second commercial LLM that CrowdStrike could not identify. In a single intrusion the planning agents consumed on the order of 1.4 billion input tokens, and the operator's Chinese instructions were tagged "【人类指令】" (human instruction). Targets spanned foreign ministries, semiconductors, defence, think tanks, telecommunications, pharmaceuticals, nuclear and finance, and CrowdStrike describes the group, active since 2021, as escalating "from single intrusions to multiple simultaneous engagements directed by a human operator".
Tempo and scale figures given to the press
CrowdStrike told CyberScoop that roughly 70 per cent of targets were confirmed compromised or showed moderate-confidence evidence of intrusion, that requests came as often as every ten seconds, and that hands-on-keyboard activity rose from a historical 40 or so commands an hour to more than 1,500. Senior vice-president Adam Meyers called it "the first time that we're seeing an agentic adversary nation-state threat actor at this scale", with about 45 agentic attacks observed in September and 126 so far in October. The blog's nine confirmed compromises and the press figure of about 70 per cent rest on different confidence thresholds, and actual harm to the 19 US organisations in pharmaceuticals, semiconductors, defence, nuclear and finance has not been disclosed.
Detection markers and defensive measures
CrowdStrike lists "artx" or "artex" in command lines, "art18-" comments on administration commands and the directory C:\ProgramData\artx5v\ as ARTEX markers, and recommends rotating signing and service keys, enforcing SAML signature validation, requiring IMDSv2 and patching internet-facing appliances. The jump in intrusion tempo is common to both Microsoft's and CrowdStrike's October reporting on ARTEX, and no government has yet confirmed the attribution.
Sources
Primary sources
- 1
VAPOR PANDA Uses AI-Driven ARTEX to Scale a Global Espionage Campaign
Source checked 1 Nov 2026
CrowdStrike · External site
Reporting
- 2
China-backed threat group used agentic AI pentesting tool to compromise 140 sensitive targets in 11 days
Source checked 1 Nov 2026
CyberScoop · External site
From developments to evidence and analysis
Background
No explicit relationship is recorded.
Evidence
No explicit relationship is recorded.
Follow-ups / later versions
No explicit relationship is recorded.
Related questions
Cite this article
AI & Security Observatory（2026-10-29）「CrowdStrike、中国系集団がARTEXで14か国200超の組織へ同時侵入と報告 侵害確認は9組織」『AI & Security Observatory』、第1.0版、最終更新2026-11-01。https://ai-security-observatory.org/ja/brief/brief-20261101-06/
AI & Security Observatory (2026-10-29). CrowdStrike reports a China-nexus group ran the agentic ARTEX framework against 200-plus organisations in 14 countries, with nine confirmed compromises . AI & Security Observatory, version 1.0. Updated 2026-11-01. https://ai-security-observatory.org/en/brief/brief-20261101-06/
Version 1.0 · Publication date 29 Oct 2026 · Updated 1 Nov 2026
Revision history
- Earlier revisions have not been recorded.