CrowdStrike said VAPOR PANDA used the open-source ARTEX agentic penetration-testing framework with DeepSeek and an unidentified commercial language model against more than 200 organizations in at least 14 countries. The activity began October 17, 2026, and involved foreign ministries, telecommunications providers, and U.S. organizations in the semiconductor, defense and think-tank sectors, among others. CrowdStrike confirmed post-exploitation activity at nine organizations; the broader figure describes targeting rather than 200 confirmed breaches. Human operators directed the operation while ARTEX agents performed reconnaissance, exploitation, credential access, persistence and follow-on intrusion tasks. Separately, DIVD said its own September breach used ARTEX, reached root in 12 minutes and generated more than 450,000 requests and 2,500 scripts, but it did not attribute that incident to VAPOR PANDA.
Why it matters
The campaign demonstrates that agentic tooling can coordinate intrusion work across many targets while remaining under human supervision. It also shows that defenders must account for rapid, automated reconnaissance and escalation, while distinguishing AI-assisted activity from fully autonomous attacks and separating targeting counts from confirmed compromises.
Key points
- CrowdStrike said one ARTEX controller targeted more than 200 organizations in at least 14 countries beginning October 17, 2026.
- Post-exploitation activity was confirmed at nine organizations; at least 12 additional organizations faced password spraying or credential scanning.
- One engagement used about 1.4 billion input tokens, according to CrowdStrike’s account.
- DIVD said ARTEX reached root 12 minutes after initial access during its separate September 2026 breach.
- DIVD recorded more than 450,000 requests and over 2,500 scripts during that incident.
Based on reporting by The original announcement and BleepingComputer.
Join the conversation! Your thoughts help the community grow.