{"id":2177,"date":"2026-11-02T07:18:41","date_gmt":"2026-11-02T00:18:41","guid":{"rendered":"https:\/\/vncybers.vn\/ho-so-vapor-panda-ai-agent-artex-gian-diep-toan-cau\/"},"modified":"2026-11-02T07:18:41","modified_gmt":"2026-11-02T00:18:41","slug":"vapor-panda-profile-artex-ai-agent-global-espionage","status":"publish","type":"post","link":"https:\/\/vncybers.vn\/en\/ho-so-vapor-panda-ai-agent-artex-gian-diep-toan-cau\/","title":{"rendered":"VAPOR PANDA Profile: ARTEX AI Agents Scale Global Espionage"},"content":{"rendered":"

VAPOR PANDA<\/strong>, m\u1ed9t nh\u00f3m \u0111e d\u1ecda c\u00f3 li\u00ean h\u1ec7 v\u1edbi Trung Qu\u1ed1c theo c\u00e1ch \u0111\u1eb7t t\u00ean c\u1ee7a CrowdStrike, \u0111\u00e3 s\u1eed d\u1ee5ng khung AI agent ARTEX \u0111\u1ec3 \u0111i\u1ec1u ph\u1ed1i nhi\u1ec1u cu\u1ed9c x\u00e2m nh\u1eadp song song. Trong 11 ng\u00e0y k\u1ec3 t\u1eeb 17\/10\/2026, chi\u1ebfn d\u1ecbch nh\u1eafm t\u1edbi h\u01a1n 200 t\u1ed5 ch\u1ee9c t\u1ea1i \u00edt nh\u1ea5t 14 qu\u1ed1c gia, cho th\u1ea5y AI \u0111ang l\u00e0m thay \u0111\u1ed5i t\u1ed1c \u0111\u1ed9 v\u00e0 quy m\u00f4 c\u1ee7a ho\u1ea1t \u0111\u1ed9ng gi\u00e1n \u0111i\u1ec7p m\u1ea1ng.<\/p>\n

\"Data
Real-world data center infrastructure. Photo: Taylor Vick\/Unsplash.<\/figcaption><\/figure>\n

Who is VAPOR PANDA?<\/h2>\n

CrowdStrike uses the name VAPOR PANDA to track a China-nexus targeted-intrusion actor. In a report published on October 29, its OverWatch threat hunters said the group's ARTEX controller exposed tools, target lists, and transcripts between a human operator and agents. The data allowed researchers to link the activity to VAPOR PANDA and observe how one operator supervised several attacks at once.<\/p>\n

CrowdStrike stressed that this campaign is separate from ARTEX activity that industry sources associated with Storm-0940. The distinction matters because several groups can use the same tool; the presence of ARTEX alone is not enough to attribute every incident to one actor.<\/p>\n

An 11-Day Campaign Targeting More Than 200 Organizations<\/h2>\n

From October 17 through October 27, 2026, one ARTEX controller directed exploitation against more than 200 organizations in at least 14 countries. Targets ranged from foreign ministries across Latin America, Europe, Asia, and the Middle East to semiconductor, defense, pharmaceutical, financial, nuclear, think tank, and telecommunications organizations.<\/p>\n

CrowdStrike confirmed post-exploitation activity at nine organizations, where the attackers deployed the BLOODALCHEMY backdoor, scanned networks, sprayed passwords, and conducted reconnaissance. It also observed credential scanning or password spraying against at least 12 more organizations. CyberScoop cited CrowdStrike as saying that about 70% of targets were compromised or showed evidence indicating a successful intrusion with moderate confidence; this is the company's assessment, not a victim count publicly confirmed by each organization.<\/p>\n

ARTEX Turns One Operator into Multiple Attack Teams<\/h2>\n

ARTEX coordinates a main agent with planner and worker agents that share a graph of facts, findings, and goals. VAPOR PANDA primarily ran the system with DeepSeek and also used a commercial model that CrowdStrike could not identify. In one intrusion, the planning agents alone consumed roughly 1.4 billion input tokens.<\/p>\n

Recovered transcripts show that humans remained in command. The operator issued instructions in Chinese, telling agents to write web shells, extract domain-administrator passwords, and inventory systems without security sensors. When one attack path failed, agents switched to alternatives across web applications, cloud services, containers, Active Directory, email, and VPN infrastructure.<\/p>\n

Technical Turning Points in the Intrusion Chain<\/h2>\n

At one U.S. pharmaceutical company, an agent combined an unsigned SAML assertion with an XML External Entity flaw to query a cloud metadata service. The resulting role could issue administrative commands to thousands of machines and eventually led to an enterprise administrator account.<\/p>\n

At another pharmaceutical target, an agent installed web shells, abused an unauthenticated orchestration service to gain root in a production container environment, read LAPS passwords, and abused resource-based constrained delegation to impersonate a privileged account. At a Greek government entity, the group used a 2022 directory export and authentication keys that had apparently not been rotated, then used VPN configurations recovered from email to reach the internal network.<\/p>\n

BLOODALCHEMY and Defense Evasion<\/h2>\n

After gaining a foothold, VAPOR PANDA deployed BLOODALCHEMY as a backdoor for persistent access and follow-on activity. The group paired automation with familiar tradecraft, including web shells, credential theft, internal scanning, and lateral movement.<\/p>\n

A notable feature was that endpoint-sensor evasion was built directly into the agent workflow. Instructions to find systems without Falcon sensors show that the attackers automated not only exploitation but also the selection of less-visible routes. According to CyberScoop, activity could exceed 1,500 commands per hour, compared with an upper range of about 40 commands per hour commonly seen in earlier hands-on-keyboard operations.<\/p>\n

Impact: AI Accelerates Established Techniques<\/h2>\n

The campaign did not introduce an entirely new category of attack. The shift lies in testing many paths simultaneously, iterating quickly, and scaling across hundreds of targets at low operating cost. This compresses the time between discovering an attack surface and gaining control, while increasing pressure on defenders whose alert-investigation workflows remain sequential.<\/p>\n

The consequences extend beyond technology. Diplomatic, defense, financial, and telecommunications organizations hold strategically valuable data; a successful intrusion can affect international relations, intellectual property, supply chains, and public trust. The use of commercial models and open-source components also raises questions about controlling abuse without restricting legitimate research.<\/p>\n

What Should Organizations Do?<\/h2>\n
    \n
  • Reduce exposed administrative surfaces, patch internet-facing applications, and audit orchestration services or APIs that lack authentication.<\/li>\n
  • Rotate old keys, certificates, service passwords, and secrets; do not assume that old backups or directory exports are harmless.<\/li>\n
  • Enable phishing-resistant MFA, reduce privileges, and monitor password spraying, LAPS access, SAML forgery, and cloud metadata access.<\/li>\n
  • Ensure consistent endpoint and log coverage across servers, containers, identities, email, VPNs, and cloud infrastructure, and hunt for monitoring gaps.<\/li>\n
  • Build machine-speed response through detection rules, controlled automated isolation, and exercises covering several concurrent attack paths.<\/li>\n<\/ul>\n

    Lessons from the VAPOR PANDA Profile<\/h2>\n

    VAPOR PANDA demonstrates that agentic AI is becoming an orchestration layer for intrusions, not merely a coding assistant. Humans still choose targets and make sensitive decisions, but machines can perform reconnaissance and experimentation at a volume far beyond a manual team. Effective defense therefore cannot depend on one product: organizations need identity governance, patching, segmentation, broad visibility, and automated response within a unified architecture.<\/p>\n

    VNCyberS<\/strong> compiled from CrowdStrike, CyberScoop, and The Hacker News<\/em><\/p>","protected":false},"excerpt":{"rendered":"

    VAPOR PANDA, m\u1ed9t nh\u00f3m \u0111e d\u1ecda c\u00f3 li\u00ean h\u1ec7 v\u1edbi Trung Qu\u1ed1c theo c\u00e1ch \u0111\u1eb7t t\u00ean c\u1ee7a CrowdStrike, \u0111\u00e3 s\u1eed d\u1ee5ng khung AI agent ARTEX \u0111\u1ec3 \u0111i\u1ec1u ph\u1ed1i nhi\u1ec1u cu\u1ed9c x\u00e2m nh\u1eadp song song. Trong 11 ng\u00e0y k\u1ec3 t\u1eeb 17\/10\/2026, chi\u1ebfn d\u1ecbch nh\u1eafm t\u1edbi h\u01a1n 200 t\u1ed5 ch\u1ee9c t\u1ea1i \u00edt nh\u1ea5t 14 qu\u1ed1c gia, cho […]<\/p>\n","protected":false},"author":2,"featured_media":2176,"comment_status":"","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"inline_featured_image":false,"footnotes":""},"categories":[70],"tags":[308,46,1055,265,1057,766,1056,939,1054],"class_list":["post-2177","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ho-so","tag-ai-agent","tag-an-ninh-mang","tag-artex","tag-bao-mat-danh-tinh","tag-bloodalchemy","tag-crowdstrike","tag-deepseek","tag-gian-diep-mang-trung-quoc","tag-vapor-panda"],"_links":{"self":[{"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/posts\/2177","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/comments?post=2177"}],"version-history":[{"count":0,"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/posts\/2177\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/media\/2176"}],"wp:attachment":[{"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/media?parent=2177"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/categories?post=2177"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/vncybers.vn\/en\/wp-json\/wp\/v2\/tags?post=2177"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}