Chinese Hackers Use ARTEX AI in Automated Attacks Against 200 Organizations Worldwide

In a significant and alarming escalation of artificial intelligence-driven cyber warfare, a state-sponsored threat group linked to China has launched a massive, automated offensive against more than 200 high-value organizations across 14 countries. Security researchers tracking the campaign have revealed that the threat actor utilized an advanced, multi-agent AI framework known as ARTEX to orchestrate simultaneous intrusions across critical infrastructure, government agencies, and major industrial sectors.

The widespread cyber espionage campaign marks a profound technological shift in how nation-state actors weaponize emerging artificial intelligence systems. Rather than relying solely on traditional manual hacking techniques or isolated AI-assisted scripts, the attackers deployed sophisticated worker agents capable of performing reconnaissance, discovering and exploiting vulnerabilities, harvesting credentials, and adapting to defensive measures in real time with minimal human intervention.

Scope, Targets, and Geopolitical Reach

Beginning in mid-October, the sweeping cyber campaign targeted entities spanning a diverse range of foundational sectors. According to telemetry and analysis from cybersecurity firm CrowdStrike—which tracks the threat actor under the moniker "Vapor Panda"—the offensive focused heavily on government ministries, telecommunications providers, pharmaceutical firms, semiconductor manufacturers, defense contractors, nuclear facilities, and financial institutions.

Geographically, the fallout from the campaign has spanned at least 14 nations. Confirmed victims and heavily targeted organizations include major telecommunications operators in the United States and Belgium, government entities in Greece, the Middle East, and Latin America, and prominent pharmaceutical enterprises based in the U.S.

Out of the more than 200 organizations targeted globally, Vapor Panda successfully compromised at least nine distinct entities, while carrying out aggressive password-spraying attacks and credential-scanning operations against at least 12 other high-priority targets. Security analysts emphasize that the breadth of the campaign demonstrates an unprecedented level of operational scaling, made possible entirely by the integration of autonomous artificial intelligence frameworks into the hacker playbook.

Anatomy of the AI-Driven Campaign

The cornerstone of Vapor Panda’s recent operational capability is ARTEX, an open-source, multi-agent AI penetration-testing framework originally developed by Chinese software engineers. Although the tool was initially released to the public as an open-source project, its creators quickly pulled it and made it proprietary after realizing its immense potential for malicious exploitation in real-world attacks. Notably, ARTEX was previously implicated in a targeted campaign against financial institutions in South Korea.

In the October campaign, Vapor Panda did not rely on ARTEX alone. The threat actor integrated ARTEX with commercial language models, including DeepSeek and an additional, unidentified commercial large language model. This hybrid AI architecture allowed the operators to coordinate strategic planning and deploy specialized worker agents across multiple intrusions simultaneously.

According to CrowdStrike researchers, human operators issued high-level strategic instructions in Chinese and intervened only at critical decision-making junctions. The AI agents managed the granular execution of the cyberattacks. Their operational duties included:

• Performing comprehensive network and system reconnaissance.
• Identifying and exploiting software vulnerabilities.
• Deploying web shells to maintain persistent access.
• Obtaining and validating administrative credentials.
• Dynamically pivoting and pursuing alternative attack paths when initial vectors were blocked.

In one particularly complex intrusion analyzed by researchers, the automated AI agents consumed an astonishing 1.4 billion input tokens, underscoring the sheer computational volume and continuous autonomous processing required to sustain the multi-threaded attacks.

Evading Defenses and Deploying Custom Tooling

The AI-driven nature of the campaign was matched by calculated tactical decisions designed to maximize stealth and persistence. Notably, CrowdStrike’s telemetry revealed that Vapor Panda explicitly programmed or instructed the AI agents to bypass Windows endpoints protected by Falcon, CrowdStrike’s proprietary endpoint protection platform. Instead, the autonomous agents were directed to focus their efforts exclusively on unprotected hosts and vulnerable network perimeters.

Furthermore, the threat actor coupled its AI capabilities with established custom malware. During operations targeting government entities in Latin America and the Middle East, Vapor Panda deployed BLOODALCHEMY, a persistent backdoor that has been actively utilized in cyberespionage operations since 2023.

In specific high-profile incidents, such as the breach of a Greek government entity, the AI agents demonstrated advanced lateral movement and persistence techniques. The agents successfully forged an administrative token and read sensitive mailboxes by leveraging legacy authentication keys harvested from a 2022 mail-server directory export, proving that the AI could effectively stitch together historical intelligence with real-time exploitation.

The Dutch Institute for Vulnerability Disclosure Incident

The scale of automated AI threats was further illuminated by disclosures from the Dutch Institute for Vulnerability Disclosure (DIVD). The cybersecurity nonprofit shared exhaustive details regarding a sophisticated cyberattack it suffered in September, confirming that the intrusion was executed entirely by an autonomous AI agent leveraging the ARTEX framework.

According to DIVD’s post-incident analysis, the AI-driven attack moved with blinding speed, progressing from initial access to root-level privileges in a mere 12 minutes. The agent achieved this by chaining a Zammad zero-day vulnerability with a secondary zero-day exploit specifically tailored for local privilege escalation.

Once inside the DIVD network, the ARTEX agent operated with relentless multitasking efficiency. It simultaneously read a compromised mailbox, scanned the internal network architecture, extracted login tokens, and attempted automated password resets. DIVD reported recording over 450,000 incoming requests and the execution of 2,500 distinct scripts during the assault. The organization described the activity as "super loud and messy," highlighting a characteristic flaw of early autonomous agents—they often prioritize speed and brute-force execution over stealth, generating massive volumes of operational telemetry.

Broader Implications for Global Cybersecurity

The revelation that state-sponsored actors are successfully weaponizing multi-agent AI frameworks like ARTEX at a global scale marks a critical turning point in cybersecurity history. Security experts warn that this campaign signals the definitive transition of AI from a theoretical assistant used for writing phishing emails or basic code snippets into an operational commander capable of directing complex, multi-stage cyberattacks across borders.

By leveraging frameworks like ARTEX alongside commercial models such as DeepSeek, threat actors can bypass traditional human resource bottlenecks. A single human operator can now manage operations that previously would have required a large, coordinated team of skilled human hackers. This dramatically lowers the barrier to entry for advanced persistent threat (APT) groups while exponentially increasing the speed and frequency of potential attacks.

Defenders now face the daunting challenge of securing networks against adversaries that think, adapt, and execute at machine speed. As automated AI agents become more prevalent in the arsenals of nation-state hackers, the cybersecurity industry must pivot toward equally advanced, AI-driven defensive architectures capable of identifying and neutralizing autonomous threats before they can establish root access and compromise critical global infrastructure.