Hackers using an open-source AI hacking tool called ARTEX mostly run it on low-cost Chinese AI models. The finding comes as the FBI warns in a FLASH advisory that the tool is being used to automate cyber intrusions.
ARTEX does not come with its own AI model. Instead, it plugs into outside large language model providers, including Anthropic’s Claude, OpenAI’s GPT, DeepSeek, GLM, Grok and Kimi.
A review of attacker-controlled setups shows Chinese models are the most widely used. The DeepSeek Flash family leads because it pairs solid capability with low cost and shows no visible guardrails when a request is framed as penetration testing.
How the Tool Works
ARTEX is a multi-agent penetration testing platform written in the Go programming language. Users describe a goal in plain language, and specialized AI agents turn it into machine commands.
Little human input is needed beyond the first instruction. The agents handle steps from scouting a target and gaining access to raising privileges and stealing data.
Some intrusions went from first targeting to privileged access in less than three hours. The tool creates noisy traffic that may set off alerts, but the speed of the attacks leaves defenders little time to respond.
A developer known as “Autumn-27” released ARTEX on GitHub in July 2026. It gained an estimated 1,500 to 2,000 stars within about ten weeks.
The developer said the tool was meant for authorized security testing. The repository came down in early October after reports of abuse, but ARTEX remains available through forks, archived copies, localized versions and a commercial offering.
Researchers found more than 1,600 publicly exposed ARTEX console instances in early November. The real number of deployments is likely higher once private and firewall-protected systems are counted.
Attacks Across Many Industries
Hackers have used ARTEX since at least August against targets in financial services, government, technology, defense and semiconductors. Energy, healthcare, manufacturing, telecommunications and legal firms have also been targeted.
Campaigns against South Korean financial institutions led to the theft of personal data on about 67,000 people. A separate supply chain compromise hit ConnectWise Cloud Backup customers and opened access to more than 2,250 downstream tenants.
The tool is good at finding less-watched systems such as supplier portals, mobile backends, legacy services and third-party integrations. Attackers can break into a weaker vendor and then move into larger enterprise networks.
An agency official said ARTEX does not appear linked to other recent cyber problems. These include issues with the National Security Agency’s internal network, banking outages tied to JPMorgan Chase and a Microsoft Outlook outage.
Organizations are urged to list every internet-facing application and limit sensitive services to trusted networks. The guidance also calls for stronger identity controls, rate limits on business APIs and scans for exposed secrets.
Firms that ever installed the affected ConnectWise applications should review every Microsoft 365 tenant involved, searching by application ID. Reviews of related activity should go back to at least Oct. 1, 2026.
Affected ConnectWise connections should stay disabled until the vendor issues formal restoration instructions. The agency said on X that it released the advisory because criminals and state actors are using AI to run familiar techniques at machine speed as the year ends.