Vapor Panda and ARTEX: 140 targets compromised in 11 days as agentic cyberattacks now operate at state scale

By CyberScoop · October 30, 2026.

A Chinese state-backed group that CrowdStrike calls Vapor Panda has used ARTEX (an automated agentic AI penetration-testing framework) to compromise some 140 sensitive targets in 11 days. According to CrowdStrike's threat report published Thursday, the campaign has targeted more than 200 organizations in at least 14 countries since October 17 and remains active.

Targets and confirmed victims include foreign ministries, the presidency and the prime minister's office of a European country, and defense and intelligence agencies. They also include 19 U.S. organizations in the pharmaceutical, semiconductor, defense, nuclear and financial sectors, as well as telecom operators in the United States, Belgium and the Philippines. CrowdStrike told CyberScoop that about 70% of the targets were confirmed as compromised or show signs of intrusion with moderate confidence.

The key is speed. According to CrowdStrike, Vapor Panda combined ARTEX with DeepSeek and sent requests as often as every 10 seconds, exploring multiple attack paths at once. Adam Meyers, the firm's senior vice president of counter adversary operations, explains that the manual activity they have tracked for 15 years ran at around 40 commands per hour. With ARTEX, he says, it could exceed 1,500.

The growth is also striking. CrowdStrike counted some 45 attacks by agentic adversaries in September and 126 in October, nearly triple, and the number keeps rising. The campaign is separate from the one Microsoft attributed to Storm-0940 this month, which also used ARTEX against ConnectWise customers.

There are caveats. The data come from a single security company with a commercial interest in the subject, and the attribution to a state is CrowdStrike's assessment, not an independently verified fact. Nor is it known how many of the compromises represent deep access and how many are merely a foot in the door.

Our take: this story confirms a thesis we have been advancing. Agentic AI does not invent new attacks; it compresses time and makes scale cheaper. Meyers puts it this way: there is nothing new, it is just happening at a speed and scale we are not used to. For a state attacker, spreading the cost across 200 targets makes mass exploitation almost free.

The corollary is uncomfortable for defenders. A team that triages alerts in hours is competing against an adversary that can iterate every few seconds. The metrics that matter become time to detection and response, along with the cross-domain visibility into endpoints, identities and domains that CrowdStrike calls for. The short-term losers are organizations with slow patching and poorly governed credentials; the winners, those that automate defense too.

The DeepSeek detail deserves attention. It is a Chinese open-weight model, and its use here is a reminder that open capability is a double-edged sword: it democratizes AI, but it also lowers the offensive barrier. That is not an argument for curbing openness, but for investing in evidence-based governance and defense.

In the long run, the same automation that today accelerates intrusion can accelerate patching, formal verification and secure-by-design software. But the transition will be hard: for a while the attacker holds a speed advantage, and that should be said without drama or complacency. As Meyers puts it, the situation is not hopeless, but it demands action.