Zero Day Room
Live
Defence

Chinese Hackers Launch AI-Powered Attack on 200 Organizations

A China-linked threat actor used the ARTEX AI framework to autonomously target over 200 organizations across 14 countries, compromising at least nine in a global campaign.

A China-linked threat actor used the ARTEX AI framework to autonomously target over 200 organizations across 14...

A massive Cybersecurity AI attack has targeted more than 200 organizations worldwide. The campaign, which spanned 14 countries, was operated by a China-linked threat actor using autonomous artificial intelligence agents.

At least nine organizations were successfully compromised. Targeted sectors included government bodies, telecommunications providers, pharmaceutical firms, semiconductor manufacturers, defense contractors, nuclear energy organizations, and financial institutions. European government bodies and foreign ministries were also in the crosshairs.

Attack Details

The attack involved various tactics, including password spraying and credential scanning. Twelve other high-value targets were subject to these specific activities.

AI agents performed extensive reconnaissance, exploited vulnerabilities, deployed web shells, and obtained credentials. They also pursued alternative attack paths autonomously. In a single intrusion engagement, the AI consumed a staggering 1.4 billion input tokens.

Threat Actor and Tools

The tracked threat actor, known as Vapor Panda, used the ARTEX AI framework to carry out the attack. ARTEX is an open-source, multi-agent AI penetration-testing framework that was later made proprietary. It was developed by Chinese developers, initially by Li Puhua and published on GitHub.

Vapor Panda used the DeepSeek large language model as a backend for the framework. Another unidentified commercial language model was also employed. This marks the first time an autonomous AI agent framework has been observed operating at this scale.

Specific Targets and Incidents

Confirmed victims spanned the globe. Telecommunications providers in the United States and Belgium were hit. Government organizations in Greece, the Middle East, and Latin America were also confirmed victims, as were pharmaceutical firms in the U.S.

A specific incident at the Dutch Institute for Vulnerability Disclosure (DIVD) in September 2026 illustrates the speed of the AI-powered attack. DIVD attributed a cyberattack to an AI agent using ARTEX AI. The attack went from initial access to obtaining root privileges in just 12 minutes. Over 450,000 requests and 2,500 scripts were connected to this intrusion, with ARTEX AI acting autonomously throughout.

Campaign and Operator

The 11-day campaign was operated by a China-linked operator. By September 2026, at least three separate China-linked threat actor clusters were running the ARTEX framework. The campaign resulted in confirmed post-exploitation activity within nine organizations.

The campaign was disclosed by CrowdStrike on October 29.

Topics

#Defence

Related coverage

More from Defence