Chinese state-linked hackers have used ARTEX AI, a multi-agent penetration-testing framework, to conduct automated attacks against more than 200 organizations across 14 countries, according to CrowdStrike.
The campaign began on October 17 and targeted high-value sectors including government, telecommunications, pharmaceuticals, semiconductors, defense, nuclear energy, and financial services.
Researchers track the threat actor as Vapor Panda and say the operation marks a significant escalation in the use of AI agents for cyberespionage, moving from isolated experiments to multiple simultaneous intrusions coordinated by human operators.
AI Agents Used Across Multiple Intrusions
CrowdStrike identified at least nine confirmed compromises and additional password-spraying and credential-scanning activity against more than a dozen high-value targets.
Victims included:
- Telecommunications providers in the United States and Belgium.
- Government organizations in Greece, Latin America, and the Middle East.
- Pharmaceutical companies in the United States.
The attackers used ARTEX AI together with DeepSeek and another unidentified commercial language model to coordinate planning agents and worker agents across multiple targets at the same time.
Operators issued instructions in Chinese and intervened mainly when higher-level decisions were required.
ARTEX AI Automated Reconnaissance and Exploitation
During the intrusions, AI agents performed tasks typically handled by human penetration testers or threat actors, including:
- Network and service reconnaissance.
- Vulnerability exploitation.
- Credential discovery.
- Web-shell deployment.
- Lateral movement.
- Alternative attack-path selection after failed attempts.
In one intrusion, CrowdStrike estimated that the AI agents consumed more than 1.4 billion input tokens, illustrating the scale of automated reasoning used during the campaign.
The operators also instructed the framework to avoid Windows endpoints protected by CrowdStrike Falcon, redirecting attacks toward systems without that endpoint protection.
Greek Government Attack Used Stolen Authentication Material
In an intrusion against a Greek government organization, Vapor Panda reportedly used previously exposed authentication material from a 2022 mail-server directory export.
The AI agents leveraged those keys to forge an administrator token and access mailboxes without relying on valid user passwords.
This technique demonstrates how historical credential exposure can remain valuable years later when combined with automated attack tooling.
BLOODALCHEMY Backdoor Deployed
CrowdStrike also observed the deployment of BLOODALCHEMY, a backdoor active since at least 2023.
The malware was used against government organizations in Latin America and the Middle East, providing persistent access after initial compromise.
The combination of AI-driven reconnaissance and conventional malware highlights how attackers are integrating autonomous systems into established cyberespionage workflows rather than replacing traditional tooling entirely.
ARTEX AI Previously Used in Real-World Attacks
ARTEX AI was originally released as an open-source penetration-testing framework developed by Chinese developers.
Its creators later made the project proprietary after the tool appeared in real-world attacks, including operations targeting South Korean financial institutions.
The framework is designed around multiple specialized AI agents that can divide tasks, execute commands, analyze results, and adapt to changing conditions during an intrusion.
DIVD Breach Shows ARTEX AI’s Speed
Additional evidence of ARTEX AI’s offensive capabilities came from the Dutch Institute for Vulnerability Disclosure (DIVD), which attributed a September 2026 breach to the framework.
According to DIVD, the AI agent moved from initial access to root privileges in approximately 12 minutes by chaining a Zammad zero-day vulnerability with a second zero-day for local privilege escalation.
During the incident, ARTEX AI simultaneously:
- Read a compromised mailbox.
- Scanned the internal network.
- Extracted authentication tokens.
- Attempted password resets.
- Executed large numbers of automated requests and scripts.
DIVD recorded more than 450,000 requests and 2,500 scripts, describing the activity as highly aggressive, noisy, and easy to detect.
AI-Driven Cyberattacks Are Scaling Faster
The Vapor Panda campaign illustrates how agentic AI can increase the speed and scale of cyber operations by allowing one operator to manage multiple active intrusions at once.
For defenders, this creates a new challenge: attack volume can increase dramatically even if the attacker’s underlying techniques remain familiar.
Security teams should prioritize strong endpoint protection, rapid patching of internet-facing systems, credential hygiene, MFA, behavioral monitoring, and detection rules capable of identifying high-volume automated reconnaissance and exploitation activity.
Leave a Reply