Artex Used Against Over 200 Organizations

According to CrowdStrike, hackers linked to China used Artex in October to target over 200 organizations. This penetration testing framework appears in several campaigns reported by cybersecurity specialists. It automates existing methods without making the attacks discreet.
Dosu reports 80,000 requests in 90 minutes
The agent programming platform Dosu claims to have received 80,000 requests from Artex in just 90 minutes. Silas Cutler, a senior security researcher at Censys, emphasizes that these tools still lack discretion. The attacks sometimes multiply thousands of high-frequency probes, leaving easily detectable activity.
Their operators are also exposed. Censys and the Shadowserver Foundation have identified thousands of instances of these systems accessible on the Internet, although it is unclear how many criminals are using them. Some Artex installations, lacking protection, have allowed researchers to examine the operations being conducted.
Last week, Silas Cutler mentioned two servers that appeared to be used by native Chinese speakers against Chinese entities. One university was notably undergoing a password spraying attack. The researcher suspects, in this case, a student eager to change their academic results, without presenting it as an established identification. He believes that the uses could involve both organized and sophisticated criminal groups as well as juvenile delinquency.
In response to many of these attacks, researchers recommend basic measures: updated software, removal of default passwords, and absence of exposed secrets. They also advise monitoring unusual behaviors on the network.
FBI warns of international targets
Since the end of August, attacks using this category of tools have targeted companies, government agencies, and banks in various countries. Last week, the FBI reported campaigns involving Artex in finance, technology, defense, semiconductors, energy, healthcare, manufacturing, telecommunications, legal activities, and public agencies.
CrowdStrike attributes the October campaign to Vapor Panda, a group linked to China. Among the targets were think tanks, American defense or semiconductor companies, as well as European ministries and others. Microsoft has also observed Chinese hackers, referred to as Storm-0940, attempting to access Microsoft 365 environments with Artex after a supply chain attack. According to Microsoft, the attackers gained privileged access in less than three hours.
Cybersecurity companies also link Artex to successful intrusions and attempts, often carried out by Chinese-speaking criminals seeking profit. They cite theft of personal data from thousands of South Korean financial institutions, a possible cryptocurrency theft, and the search for accessible credentials on GitHub.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
Artex distributes tasks among multiple agents
Artex is not an autonomous AI model. It connects external and open-source models, through programming interfaces, to already available penetration testing tools. These configurations often borrow their methods from legitimate defensive audits: they examine targets and attempt to enter inadequately protected networks, following plans defined by humans.
A natural language command is enough to initiate an operation. Check Point observed a group requesting access to a site's administration, a database, or a server. One agent transforms the request into tasks, others attempt to accomplish them, and then the system presents the results. Artex has been used from the initial examination of a network to the preparation of data for export, including intrusion and exploration of systems.
Jacob Larsen, an offensive security researcher and founder of Cypen, emphasizes this integrated coverage, where previous tools were limited to reconnaissance or testing. However, half a dozen security companies do not attribute any particular sophistication or new attack methods against sites and servers to them. They rather highlight a speed beyond human reach: according to researchers, some operations have reduced from several days to just a few hours. For Guillaume Valadon, a researcher at GitGuardian, Artex stands out for its speed and efficiency; like other specialists, he believes its simplicity makes it accessible to amateurs lacking skills.
Over the past year, security researchers have warned of a possible acceleration in intrusions and thefts by AI agents. According to CJ Moses, Chief Information Security Officer at AWS, these frameworks are no longer limited to experimentation and are conducting prolonged attacks targeting multiple objectives with limited human intervention. The protections integrated into the models, particularly those from major labs, can nonetheless interrupt detected malicious activity. Free and modified Chinese models are also advancing in their ability to penetrate systems.
The withdrawal leaves Artex copies circulating
Written in Go and distributed as open source, Artex was published on GitHub in July by a developer known by the pseudonym Autumn-27. It then spread within certain hacker communities, becoming one of the most notable frameworks in this category. On October 8, its creator removed it from GitHub, denying any connection to the attacks and claiming to have designed it for teaching and research purposes.
However, copies continue to be used, and their development is ongoing. Dozens of AI penetration testing frameworks are increasingly associated with real attacks. Guillaume Valadon notes that GitGuardian has identified 45 AI attack frameworks. Among the names frequently cited by researchers are CyberstrikeAI, Hexstrike, CAIRN, and PentAGI, whose operating principles are similar despite some technical differences.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.