CrowdStrike reported on October 29 that China-linked VAPOR PANDA used the ARTEX agent framework and language models to run multiple espionage engagements in parallel. The company says activity began October 17.
Targets are not confirmed victims
Researchers observed targeting of more than 200 organizations during an 11-day window, spanning government, telecommunications, semiconductor and other sectors. CrowdStrike confirmed post-exploitation activity at nine organizations. That distinction matters: the report does not establish 200 successful breaches.
The researchers identified DeepSeek as the primary language model, supplemented by a second commercial model they could not confirm. Their findings draw on telemetry, exposed tooling and recovered agent transcripts.
Human operators remained involved
CrowdStrike says operators issued instructions and steered agents at decision points. The campaign therefore illustrates AI-assisted scaling of established intrusion methods, rather than evidence that an independent AI initiated the operation.
The report describes BLOODALCHEMY backdoor deployment and efforts to identify endpoints without security sensors. CrowdStrike assesses that this activity is separate from ARTEX operations previously reported under Storm-0940. Attribution and scope are the security company’s findings.
Source
CrowdStrike: VAPOR PANDA and ARTEX campaign (October 29)
Feature image: AI-generated editorial illustration for Guava Insights. Conceptual artwork, not an actual product photograph, interface or facility.


