Chinese hackers used ARTEX, an open-source, multi-agent artificial intelligence penetration-testing framework, in a cyberespionage campaign targeting more than 200 organizations across 14 countries. The operation began on October 17 and compromised at least nine organizations, while attackers also conducted password-spraying and credential-scanning operations against at least 12 other high-value targets.
The targets included government ministries, telecommunications providers, pharmaceutical companies, semiconductor firms, defense organizations, nuclear-power entities and financial institutions. Confirmed victims include telecommunications providers in the United States and Belgium, government agencies in Greece, the Middle East and Latin America, and pharmaceutical companies in the United States.
AI agents coordinated multiple intrusions
CrowdStrike researchers are tracking the attackers as Vapor Panda. They said the campaign shows that the use of Chinese-aligned AI agents is expanding beyond individual intrusions to support multiple simultaneous attacks directed by human operators.
In the campaign, Vapor Panda used DeepSeek and a second unidentified commercial language model to coordinate planning and worker agents across several intrusions. The operators issued instructions in Chinese and intervened when critical decisions were required, according to CrowdStrike.
The AI agents conducted reconnaissance, exploited vulnerabilities, deployed web shells, captured credentials and pursued alternative attack paths when necessary. The operator specifically instructed an agent to skip Windows systems protected by Falcon, CrowdStrike’s endpoint protection platform, and concentrate on unprotected hosts.
Researchers believe the AI agent consumed more than 1.4 billion input tokens during a single breach.
Attackers exploited an old authentication key
In an attack against a Greek government agency, an AI agent reportedly used an authentication key from a 2022 mail-server directory export to forge an administrator token and read mailboxes.
CrowdStrike also found that Vapor Panda deployed BLOODALCHEMY, a backdoor active since 2023, against government agencies in Latin America and the Middle East.
ARTEX AI was originally an open-source project
ARTEX AI is a penetration-testing suite developed by Chinese developers and initially released as an open-source project. Its creators later took control of the project after it was used in real-world attacks, including an intrusion against a South Korean financial institution.
DIVD attack shows similar capabilities
The Dutch Vulnerability Disclosure Institute (DIVD) published details of a cyberattack against its organization in September 2026 and attributed the activity to an AI agent. According to subsequent updates, the tool used against the network was the autonomous ARTEX AI.
In a LinkedIn post, DIVD explained that the attack obtained root privileges within 12 minutes of initial access by chaining the Zammad zero-day with a second zero-day for local privilege escalation.
The agent simultaneously read compromised mailboxes, scanned internal networks, extracted login tokens and attempted to reset passwords. DIVD recorded more than 450,000 requests and 2,500 scripts associated with the attack, describing the activity as “very noisy and troubling.”
Source: www.bleepingcomputer.com