China-linked hacking groups are using an AI penetration-testing tool called ARTEX to run faster and broader cyber espionage campaigns, according to research from CrowdStrike and Microsoft. The tool splits work between planner agents and worker agents. Planners map out attacks while workers scan systems, exploit flaws, and move through networks. Humans still make final decisions, but much of the early work is handled by AI.
Vapor Panda’s broad campaign
CrowdStrike linked ARTEX to Vapor Panda, which has used it since October 17, 2026, against more than 200 organizations worldwide. The campaign reached at least 14 countries and hit foreign ministries and telecom providers. In the US, 19 groups in pharmaceuticals, semiconductors, defense, nuclear, and finance were targeted. During one attack, planner agents consumed 1.4 billion input tokens, according to CrowdStrike. The group used DeepSeek and an unnamed commercial AI system.
The scale did not mean every intrusion worked. CrowdStrike confirmed post-exploitation activity at only nine companies, including use of the BLOODALCHEMY backdoor. Human operators then took over, creating web shells, pulling passwords, and finding unprotected machines.
Storm-0940 and faster cloud access
Microsoft tied ARTEX to Storm-0940, another China-linked group, which began using the tool as early as September 2026. It targeted hundreds of organizations and in some cases gained privileged cloud access in under three hours. Microsoft also linked the group to the ConnectWise Cloud Backup compromise. From October 10 to 14, Microsoft tracked more than 151,000 failed logins against 327 accounts at eight companies. Attempts peaked at 2,900 per minute, though Microsoft saw no successful login in that burst.
A financial motive and earlier AI attacks
Zscaler reported a third ARTEX campaign in late September 2026 against South Korean financial institutions. Attackers stole personal data belonging to about 67,000 people. Zscaler assessed the attacker as Chinese-speaking and financially motivated with moderate confidence, but did not confirm state sponsorship. CrowdStrike also found ARTEX files and Claude Code logs on attacker-controlled systems.
The trend predates ARTEX. In 2025, Anthropic found a Chinese state-sponsored campaign that used Claude Code against about 30 targets, with AI handling an estimated 80 to 90 percent of the work. Google has documented attackers moving from simple prompts toward more autonomous operations. Those cases were separate, but they show how quickly AI-assisted hacking is changing.
The risk reaches beyond the groups already hit. Banks face customer-data breaches, chipmakers could lose intellectual property, and cloud providers may see compromised app credentials expose multiple customers. Verizon found system intrusion accounted for 61 percent of breaches in its 2026 report. Microsoft said exposed cloud workloads faced attacks within 5.3 hours on average. The World Economic Forum found 65 percent of large companies saw third-party and supply-chain vulnerabilities as their biggest cybersecurity challenge, up from 54 percent in 2025. Businesses may respond by spending more on identity controls, faster detection, and supplier monitoring.
