Artex, an open-source AI hacking framework, drives a wave of automated attacks
Security firms and the FBI link the Go-based tool, which directs AI models to run penetration tests, to attacks on banks, agencies and companies worldwide since late August.
· 2 min read

- Since around late August, criminals have used AI-driven penetration testing frameworks, led by Artex, to attack banks, agencies and companies.
- Artex is not an AI model but a guide that links outside AI models to existing hacking tools, letting unskilled users launch attacks in hours.
- Researchers say the tools are noisy and easy to spot, and that basic security hygiene can block many of the attacks.
A new class of AI-driven hacking frameworks, led by a tool called Artex, is being used by cybercriminals to attack organizations around the world, WIRED reported. Security companies say the activity has been building since around the end of August.
The frameworks, sometimes called harnesses, instruct large language models to scan targets on their own and try to break into poorly secured networks. They then carry out attack plans written by human operators. Many mimic legitimate defensive penetration testing. According to alerts from about half a dozen security firms, they bring no new attack methods and are not sophisticated, but they work at a speed humans cannot match. Attacks that once took days can now be finished in hours.
Artex is written in Go and was first posted on GitHub in July by a pseudonymous developer using the handle Autumn-27. It covers every stage of an intrusion, from initial network analysis and gaining access to moving inside systems and preparing data for export. Jacob Larsen, founder of security firm Cypen, called it a "point-and-shoot weapon" because earlier tools usually handled only reconnaissance or only testing.
Artex is not an AI product itself. It connects frontier and open-source models through APIs to existing penetration testing tools. Operators give plain-language prompts, after which one AI agent splits the plan into tasks, others carry them out, and the system reports its findings. Guardrails in models from leading labs can block hacking attempts, but free Chinese models and abliterated models are getting better at breaking into systems, according to the report.
The FBI warned last week that a range of malicious actors have used Artex against financial services, government, technology, defense, semiconductor, energy, health care, manufacturing, telecommunications and legal organizations. CrowdStrike said China-linked hackers known as Vapor Panda used it in October against more than 200 organizations, including think tanks and defense and semiconductor firms. Microsoft said it saw a group it tracks as Storm-0940 use Artex to try to sign in to Microsoft 365 setups, gaining privileged access in under three hours. Other reports tie it to the theft of personal data from thousands of South Korean financial institutions and to possible cryptocurrency theft.
On October 8, the Autumn-27 developer removed Artex from GitHub, saying they were in no way associated with the malicious attacks and that the tool was made for educational and research purposes. WIRED said an email to an address linked to the developer got no reply. Copies of the framework are still being developed and used. GitGuardian has documented 45 AI attack frameworks, including CyberstrikeAI, Hexstrike, CAIRN and PentAGI.
Researchers note that the tools are currently easy to detect. Silas Cutler of Censys said they are not stealthy, and they often generate thousands of probes in quick succession. Dosu said Artex sent 80,000 requests to its systems in 90 minutes. Patching software, avoiding default credentials and exposed secrets, and watching for unusual network activity can counter many of these approaches.
Censys and the Shadowserver Foundation found thousands of exposed instances of these systems, though the number of users is hard to establish. Cutler said exposed servers suggested both organized criminal activity and what looked like juvenile crime, including a password-spraying attempt on a university.