FBI Links Open-Source AI Tool Artex to Real-World Breaches in Under Three Hours
The FBI on Friday issued a FLASH advisory warning that Artex, an open-source AI-powered penetration testing framework, has been weaponized in multiple campaigns that compromised a U.S. IT management giant and South Korean financial institutions, with some intrusions reaching privileged access in under three hours [3]. The advisory, dated November 13, 2026 [6], warns that the framework "enables threat actors with minimal technical expertise to conduct complex, multi-target operations simultaneously, fundamentally accelerating attack velocity and expanding operational scale" [2]. Researchers at Aviatrix characterize Artex as a "paradigm shift" in attacker capability [1].
What Artex Is
Artex is a Go-based autonomous penetration testing framework with a Next.js frontend, originally released on GitHub in July or August 2026, according to different reports [2][3], by a developer using the handle "Autumn-27" [2][3]. The developer was later identified as Chinese cybersecurity engineer Li Puhua [2]. Rather than embedding its own model, Artex plugs into external large language models — including Anthropic's Claude, OpenAI's GPT, DeepSeek, GLM, Grok, and Kimi — to plan and execute reconnaissance, vulnerability discovery, exploitation, privilege escalation, and data exfiltration with little human intervention [3][7]. The framework translates high-level tactical goals into machine commands executed by specialized AI agents [2].
ConnectWise Supply-Chain Compromise
The most extensive confirmed campaign exploited ConnectWise's Cloud Backup application identities (formerly SkyKick). Microsoft attributes this to Storm-0940, a China-state-aligned actor active since 2021 [7]. After compromising the application identities, Storm-0940 used Artex to attempt authentication against more than 2,250 downstream Microsoft 365 tenants [7]. Huntress reported at least 162 tenants compromised [3]. In multiple environments, Artex-driven agents progressed beyond authentication to Microsoft Graph reconnaissance, application credential modifications, and privilege changes that could enable persistent access [7]. Microsoft observed "multiple instances where the time from initial targeting to privileged access was less than 3 hours" [7]. ConnectWise did not respond to requests for comment on the FBI's claims [5].
South Korean Financial Sector Attacks
A separate campaign targeted South Korean banks, including Shinhan Bank, KB Kookmin Bank, and Hana Bank [3]. CrowdStrike found the attacker's exposed infrastructure — an Artex instance at IP 38.244.50.120 and a Hong Kong-based controller — containing Claude Code session histories, Artex configuration files, and Chinese-language prompts directing the LLM to conduct penetration testing [8]. The attacker used DeepSeek v4.1-flash as the primary LLM backend, supplemented by GLM-5.3 (Zhipu AI) and Grok 4.6, accessed via a likely API reseller domain (xcai.pro) [8]. Personal data of approximately 67,000 people was exposed, according to the FBI [3], while TechCrunch reported at least 68,000 [5]. South Korean police subsequently arrested seven suspects linked to intrusions against more than 90 organizations, with roughly 8.2 million records stolen from hospitals, educational institutions, and businesses [3].
Other Campaigns
CrowdStrike also reported that a Chinese cyberespionage group tracked as Vapor Panda used Artex in automated attacks against more than 200 organizations across 14 countries, compromising at least nine [3]. Barracuda disclosed an Artex-linked breach where the tool found an exposed credential belonging to a former third-party vendor, accessing credential stores used to collect data from customer systems [5]. Barracuda's CISO Arve Kjoelen wrote: "Defenders who rely on spotting malicious activity and then responding to it will often arrive after the attacker has finished" [5]. Amazon separately found "disparate Chinese-nexus threat groups" using Artex for large-scale credential theft to access AI models, and for real-time deepfake injection to bypass cryptocurrency exchange KYC checks [2].
Proliferation and Developer Response
Li Puhua removed the GitHub repository on October 8 after abuse reports surfaced [3], denying responsibility for crimes committed with the tool [2]. The FBI says Artex continues to spread through community forks, archived copies, localized variations, and a commercial edition called Artex Pro [2][3]. Censys identified over 1,600 publicly exposed Artex consoles as of early November, with total deployments likely higher [2][3]. MalwareMustDie released an open-source detection signature pack on November 13 [3].
What the Evidence Shows
The FBI advisory, Microsoft's threat intelligence, and CrowdStrike's infrastructure analysis converge on three points: Artex is operational in real intrusions; it can compress the full intrusion lifecycle to hours in some cases; and it lowers the expertise barrier for multi-target campaigns. The "under three hours" figure comes from Microsoft's telemetry on Storm-0940's ConnectWise campaign, not a universal average [7]. The FBI's own advisory says "some intrusions progressing from initial targeting to privileged access in under three hours" [3]. Artex remains widely accessible through forks and archives despite the original repository's removal.
Sources
- FBI ARTEX AI Hacking Tool Warning 2026 (aviatrix.ai)
- FBI flash alert says Artex tool automating cyberattacks at ' ... (therecord.media)
- FBI warns ARTEX AI hacking tool is powering global ... (bleepingcomputer.com)
- FBI Warns AI Hackers Need Three Hours. Can Firms Stop ... (pymnts.com)
- FBI says Artex AI open-source hacking tool used to breach networks in 'under three hours' | TechCrunch (techcrunch.com)
- https://www.ic3.gov/CSA/2026/261113.pdf (ic3.gov)
- AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise | Microsoft Security Blog (microsoft.com)
- Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance (crowdstrike.com)