Chinese state-sponsored hackers have deployed an AI-powered penetration testing framework to strike more than 200 separate organizations around the globe simultaneously, according to a report published Thursday by CrowdStrike. The cybersecurity firm attributes the campaign to a Beijing-linked threat cluster it calls Vapor Panda, which has been active since February 2021. Starting October 17, 2026, the group used a Go-based tool named ARTEX—powered by DeepSeek and an unidentified commercial large language model—to breach high-value targets at a scale previously impossible for manual operations.

The targets span foreign ministries across Latin America, Europe, Asia, and the Middle East; semiconductor, defense, and think tank entities in the United States; and telecommunications companies in multiple regions. Among those hit were the foreign ministries of two Latin American nations, one Central American country, and one Asian state, plus a Middle Eastern interior ministry. A European country saw intrusions across its foreign ministry, presidency, prime minister's office, and defense, justice, finance, and intelligence bodies. Nineteen U.S.-based organizations in pharmaceutical, semiconductor, defense, nuclear, and financial sectors were attacked, alongside telecommunications providers in the United States, Belgium, and the Philippines. Post-breach activity has been verified at nine organizations, where attackers installed BLOODALCHEMY—a variant of the ShadowPad malware—then carried out credential spraying, network scanning, and reconnaissance. Vapor Panda also launched password-spraying and credential-scanning efforts against at least 12 additional organizations, including U.S. think tanks and European government entities.

The operation's core relies on ARTEX's ability to coordinate an army of planner and worker agents that pursue a shared objective—moving from initial access through host reconnaissance to lateral movement—while a human operator supervises and directs the agents to write web shells, extract domain administrator passwords, and map endpoint sensor coverage. CrowdStrike found that "in a single intrusion, Vapor Panda's planning agents consumed on the order of 1.4 billion input tokens, indicating the scale of the automated activity." Recovered transcripts show the agents chaining an unsigned SAML assertion with an XML external entity flaw to query cloud metadata services, then obtaining roles capable of issuing administration commands to thousands of instances before escalating access to recover enterprise administrator accounts. The report states that "agentic tooling like ARTEX enables this scale of expansion at speed, allowing a small team to pursue a wider target set than manual operations would support."

ARTEX represents a fundamental shift in how nation-state threat actors conduct cyber espionage campaigns, according to the report. Unlike earlier uses of AI to assist with individual break-ins, ARTEX runs multiple engagements concurrently, with the human operator stepping in only to steer agents toward objectives and bypass endpoint defenses. The framework pursues multiple attack paths in parallel—spanning web applications, cloud services, container environments, Active Directory, and mail and VPN infrastructure—and shifts to alternative approaches when one path fails. In one intrusion, agents used authentication keys that hadn't been rotated since 2022 to forge an administrator token and read mailboxes, then leveraged VPN configurations from a user's sent mail along with a VPN client key from an alleged May 2024 email leak to reach the internal network. CrowdStrike notes this breadth "likely enables a single operator to apply a wider range of techniques against more targets at once than would be feasible through manual operations." The disclosure arrives as ARTEX has seen rapid adoption among threat actors, prompting its developer Li Puhua to take the tool closed source. The framework "orchestrates multiple LLM backends to autonomously perform reconnaissance, vulnerability identification, and exploitation, significantly lowering the technical barrier for conducting complex intrusion campaigns," according to a separate Zscaler ThreatLabz report. Organizations that maintain long-outdated credentials or rely on static defenses face dramatically elevated risk as automated tools compress the timeline from reconnaissance to compromise. The competitive pressure among adversaries to adopt such capabilities suggests the window for defenders to harden authentication and rotation policies is narrowing rapidly.