CrowdStrike links Vapor Panda to global AI powered intrusion campaign

CrowdStrike has attributed a global intrusion campaign to Vapor Panda, a China-based threat actor, which leverages an AI-enabled Go framework named ARTEX to conduct large‑scale, automated intrusions. The operation began around 17 October 2026 and targeted more than 200 organisations worldwide, spanning foreign ministries in Latin America, Europe, Asia and the Middle East; plus semiconductor, defence and think‑tank groups in the United States and telecommunications providers in several regions.

ARTEX works with DeepSeek and an as‑yet‑unknown subscription‑based commercial large language model (LLM), using a planner and worker agent network to manage multiple engagements in parallel. A human operator remains in the loop to steer the agents, writing web shells, extracting domain administrator passwords and enumerating endpoint sensor coverage. CrowdStrike notes that in a single intrusion the planning agents processed about 1.4 billion input tokens, illustrating the scale of automation involved.

Post‑exploitation activity has been confirmed at nine organisations, with BLOODALCHEMY, a derivative of ShadowPad, deployed alongside credential spraying, network scanning and reconnaissance. The operation has also involved password spraying and credential‑scanning across at least 12 additional targets, including U.S. think tanks and European government entities.

The campaign demonstrates broad abuse of ARTEX to run concurrent engagements across web applications, cloud services, container environments, Active Directory, and mail and VPN infrastructure, with adversaries shifting tactics when paths failed.

CrowdStrike emphasises that the human operator’s involvement helps bypass endpoint protections, while the evidence includes recovered transcripts showing chaining SAML/XXE abuse, unauthenticated orchestration service compromise in containers, and the reuse of long‑standing authentication keys to forge administrator tokens.