
Since late August, companies, government agencies and banks around the world have been targeted by cybercriminals using a new wave of hacking tools powered by artificial intelligence. These tools, many of which mimic legitimate security testing techniques, use advanced language models to autonomously scan potential targets and attempt to penetrate poorly protected networks. They then execute attack plans drawn up by human operators.
Artex is the most prominent tool in this wave, according to alerts from security companies and the FBI, which warned last week about its use in cyberattacks around the world. Chinese hackers and other groups have used the hacking tools to target organizations in various sectors, from finance to defense, according to reports from security companies and the FBI.
Artificial intelligence is lowering the threshold for attackers, experts say, while the new tools are turning hacking into a “one-click” attack. According to alerts from about half a dozen security companies, the tools do not introduce new ways of attacking websites or servers and are not sophisticated. They enable criminals to launch attacks at a speed that would be impossible for humans, while hackers do not need to be skilled to use them: attacks that once took days are completed within a few hours.
Dozens of such AI penetration frameworks, also known as “harnesses,” are increasingly being linked to real attacks. The best known of these tools is Artex. The open-source framework, written in the Go programming language, was first published on GitHub in July by a developer with the pseudonym Autumn-27 and has since spread across several hacker communities.
Artex has been used in all phases of an attack, from initial network reconnaissance and gaining access to moving within systems and preparing data for exfiltration. Jacob Larsen, an offensive security researcher and founder of the company Cypen, describes Artex as a comprehensive product.
According to him, earlier versions of AI-agent testing tools only performed target discovery and analysis, while some others only testing, whereas Artex is “a true one-click weapon.” Although it is called an artificial intelligence tool, Artex is not itself such a product.
It is essentially a guide that directs external artificial intelligence models toward a range of existing penetration testing tools and processes. Users can launch operations with simple instructions in natural language, such as a request to launch a penetration attack against a target to gain access to its administrator panel, database or server.
Then, several artificial intelligence agents implement these instructions: one divides the plans into tasks, others carry them out, and the system presents the findings. Security safeguards in the underlying models, especially those from major labs, can stop hacking activity when they detect it.
Increasingly, however, free Chinese models and uncensored ones are becoming better at penetrating systems. Guillaume Valadon, a researcher at the security company GitGuardian, describes Artex as fast and efficient and notes that the simplicity of the tools lowers the threshold for inexperienced hackers, to the point that, according to him, even his grandmother could use it if she had a laptop.
Numerous reports from cybersecurity companies link the use of Artex to attempted and successful attacks, often by Chinese-speaking cybercriminals seeking profit. Among them are cited the theft of personal data from thousands of financial institutions in South Korea, possible cryptocurrency thefts and the search for exposed login credentials on GitHub.
The FBI warned last week that various malicious cyber actors have used Artex for campaigns against organizations around the world, in sectors including financial services, government, technology, defense, semiconductors, energy, healthcare, manufacturing, telecommunications and the legal industry.
In October, China-linked hackers known as Vapor Panda used Artex to target more than 200 organizations, including research institutes, semiconductor and defense firms in the United States, as well as foreign ministries in Europe and elsewhere, according to the security company CrowdStrike.
Microsoft discovered Chinese hackers, tracked as Storm-0940, who used Artex to attempt to enter Microsoft 365 systems after a supply chain attack; the hackers needed less than three hours to gain privileged access to systems. CJ Moses, chief information security officer at Amazon Integrated Security, wrote in a company blog post that agentic frameworks have moved from experimental use to sustained offensive operations with multiple targets, operating with minimal human intervention.
On October 8, the developer Autumn-27 removed Artex from GitHub, stating in a comment that it is “in no way connected to these malicious attacks” and that the tools were developed for “educational and research purposes.” An email sent to an address linked to the developer received no response.
Despite the withdrawal, several versions of the framework have been copied and continue to be developed and used, along with other similar tools. Valadon says GitGuardian has documented 45 different artificial intelligence attack frameworks so far, while among those most mentioned by researchers are CyberstrikeAI, Hexstrike, CAIRN and PentAGI, which operate in similar ways, with some minor technical differences.
One piece of good news for defenders is that these tools are currently easy to detect. Silas Cutler, a principal security researcher at the company Censys, notes that the tools are not yet stealthy and that many attacks create a lot of noise in systems, making thousands of attempts in a short time.
The agentic coding platform Dosu announced that Artex was used to send 80,000 requests to its systems in just 90 minutes. According to researchers, basic security measures, such as updating software, avoiding default credentials or exposed secrets and monitoring unusual network activity, help protect against many of these approaches.
Determining the exact number of criminals using Artex and other frameworks is not easy, but thousands of instances of these configured systems are exposed on the internet, according to analyses by Censys and the Shadowserver Foundation. Some of these instances have been left insecure and completely open on the internet, allowing researchers to see how operators use them.
Last week, two exposed Artex servers appeared to show native Chinese speakers targeting other Chinese entities; in one case, the target was a university, attacked with a password-spraying technique, with the appearance of a student trying to change grades.
Cutler estimates that part of this activity is certainly organized and sophisticated crime, while in parallel another part belongs to the realm of juvenile crime.
