CrowdStrike said in a 29 October blog post that the China-nexus group it tracks as VAPOR PANDA used ARTEX, an agentic AI penetration-testing framework, “against more than 200 distinct organizations worldwide” from 17 October 2026, with activity from a single ARTEX controller reaching entities in at least 14 countries over an 11-day window. CyberScoop, which interviewed CrowdStrike, also reported the campaign. We first reported VAPOR PANDA on 29 Oct.
CyberScoop reported that hands-on-keyboard activity tracked by CrowdStrike typically involved up to 40 commands per hour; CrowdStrike’s Adam Meyers told it that with ARTEX “we could see it exceed 1,500 commands an hour”. By our arithmetic, that is more than 37 times that manual ceiling.
Business Pill · MULTI-AGENT SYSTEMS
A short Business Engineer explainer on multi-agent systems: how a coordinating agent splits work across other agents. It teaches the general idea only and does not discuss the campaign in this story.
The key insight: CrowdStrike says the China-nexus group VAPOR PANDA used the ARTEX agentic framework against more than 200 organizations in at least 14 countries over 11 days from 17 October 2026, confirming post-exploitation at nine; by our arithmetic, the pace Meyers described is more than 37 times the manual ceiling CyberScoop reported.
According to CrowdStrike’s blog, targets included foreign ministries in Latin America, Europe, Asia and the Middle East; semiconductor, defense and think tank organizations in the U.S.; and telecommunications providers in multiple regions. CrowdStrike says it confirmed post-exploitation activity at nine organizations and saw password-spraying and credential-scanning against at least 12 more.
CyberScoop reported that the campaign “already compromised about 140 sensitive targets in 11 days, according to CrowdStrike”, and that CrowdStrike told it about 70% of the group’s targets “were confirmed compromised or had evidence indicating moderate confidence of a successful intrusion”. The blog itself gives the nine confirmed cases; the larger figure is CrowdStrike’s estimate as relayed by CyberScoop.
Commands per hour in an intrusion: up to 40 for typical hands-on-keyboard activity and more than 1,500 with ARTEX, as CrowdStrike’s Adam Meyers described them to CyberScoop. 1,500 / 40 = 37.5 is our arithmetic. CrowdStrike says the VAPOR PANDA campaign ran against more than 200 organizations in at least 14 countries over 11 days from 17 October 2026.
How the Agents Were Run
CrowdStrike says VAPOR PANDA ran ARTEX primarily with the Chinese-origin DeepSeek model, supplemented with a second, subscription-based commercial model it could not confirm. In a single intrusion, it says, the planning agents “consumed on the order of 1.4 billion input tokens”.
The company says the framework “kept a human in the loop”: an operator issued instructions in Chinese and steered the agents at decision points, while the agents ran several engagements at once and switched approach when a path failed. CrowdStrike calls the campaign “an escalation in China-nexus use of AI agents, from single intrusions to multiple simultaneous engagements directed by a human operator”.
A Third ARTEX Campaign in a Month
This is the third ARTEX-linked campaign we have covered this month. CrowdStrike reported on 7 October that a threat actor used ARTEX against South Korean financial organizations, in our piece on the Korean finance campaign, and Microsoft reported on 21 October that Storm-0940 used it in the ConnectWise breach, in our piece on Microsoft’s 11 checks. CrowdStrike says the VAPOR PANDA campaign is separate from the Storm-0940 activity.
CyberScoop reported that CrowdStrike observed roughly 45 agentic adversary attacks in September and 126 so far in October. By our arithmetic, that is about 2.8 times in a month that is not yet over.
Agentic adversary attacks CrowdStrike observed: roughly 45 in September 2026 and 126 in October as of 30 October, as reported by CyberScoop. 126 / 45 = 2.8 is our arithmetic; October is a partial month.
What CrowdStrike Tells Defenders
CrowdStrike’s recommendations include rotating signing keys and service credentials after any suspected data exposure, including keys in historical backups; enforcing SAML signature validation and IMDSv2; patching internet-facing appliances; and running endpoint detection across the environment. According to CyberScoop, Meyers said the attacks are not showing new techniques: “it’s just doing it at a speed and scale that we’re unaccustomed to”.
On the same day, we reported that Red Hat and Ubuntu lack a fix for a Linux flaw Claude exploited, in our Linux patch-gap piece.
The Structural Read
As we read CrowdStrike’s post, the change is in parallelism: one operator steered several engagements at once and intervened at decision points, which CrowdStrike calls an escalation from single intrusions to multiple simultaneous engagements.
As we read the model detail, the agents ran mainly on the Chinese-origin DeepSeek model plus an unconfirmed subscriptionmodel, with one intrusion consuming on the order of 1.4 billion input tokens, according to CrowdStrike.
As we read the month, this is the third ARTEX-linked campaign we have covered in October, after CrowdStrike’s Korean finance report and Microsoft’s Storm-0940 report; CrowdStrike says this one is separate from Storm-0940.
Adam Meyers, CrowdStrike, to CyberScoop, 30 Oct 2026
“This is the first time that weโre seeing an agentic adversary nation-state threat actor at this scale”
Three Implications
FOR DEFENDERS CrowdStrike’s recommendations include rotating keys and credentials after any suspected exposure, enforcing SAML signature validation and IMDSv2, and patching internet-facing appliances.
FOR THE COUNT CyberScoop reported that CrowdStrike observed roughly 45 agentic adversary attacks in September and 126 so far in October; by our arithmetic, about 2.8 times.
FOR ATTRIBUTION The attribution to VAPOR PANDA is CrowdStrike’s, and the 140-target and 70% figures are CrowdStrike’s as relayed by CyberScoop; the blog confirms nine.
The analysis puts it in one line: “The surface, not the model, governs autonomy.”
As we read it, ARTEX is such a surface: CrowdStrike describes a framework coordinating a main agent with planner and worker agents on top of commercially available models. The framework was published in July 2026 and does not discuss this campaign.
What Is Not Established
The 140-target figure and the 70% share come from CrowdStrike via CyberScoop; CrowdStrike’s blog confirms post-exploitation at nine organizations. CrowdStrike did not identify the second commercial model, and the victims are not named in the sources we read. We did not contact CrowdStrike, and the attribution to VAPOR PANDA is CrowdStrike’s.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
CrowdStrike says VAPOR PANDA ran ARTEX against more than 200 organizations in at least 14 countries in 11 days from 17 October, confirming post-exploitation at nine. Meyers told CyberScoop ARTEX could exceed 1,500 commands an hour, against up to 40 for typical manual activity as CyberScoop reported it, which by our arithmetic is more than 37 times.
A note on sourcing. Campaign details are from CrowdStrike’s blog post of 30 October 2026; the commands-per-hour, 140-target, 70% and monthly attack-count figures are CrowdStrike’s as reported by CyberScoop. We omit indicators of compromise and technique detail; defenders should read CrowdStrike’s post directly. Ratios are our own arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
