More Than 800 Artex AI Hacking Tools Found Exposed Online

Security researchers have discovered more than 800 publicly accessible deployments of Artex AI hacking tools, raising concerns about the exposure of systems designed to identify and exploit digital vulnerabilities. The findings, published by cybersecurity firm Censys, show how poorly secured offensive security infrastructure can create risks beyond the organisations operating it.

Artex is an open-source, autonomous penetration-testing tool that uses AI agents to automate parts of the security testing process. Researchers say its publicly accessible deployments could allow outsiders to reach hacking infrastructure without needing a password. The discovery highlights the security risks associated with leaving powerful AI-assisted tools connected to the public internet.

Censys identifies hundreds of exposed Artex deployments

Censys reported that it identified 1,465 internet-facing hosts running an Artex console as of November 4. However, approximately 650 were concentrated in two blocks of identically configured hosting. After accounting for those clusters, the company estimated that roughly 815 deployments appeared to be distinct.

The distinction matters because the raw host count does not necessarily represent the same number of independent installations. Censys also noted that Artex consoles do not need to be publicly accessible to operate. Some exposed instances may belong to legitimate security professionals conducting authorised penetration tests, and the findings do not establish that every deployment is being used maliciously.

The exposure nevertheless creates opportunities for unauthorised access. Censys also found publicly accessible directories containing files associated with offensive operations, including exploit binaries, stagers and implants. These materials could reveal operational information or give outsiders access to tools intended for controlled security testing.

Artex has been linked to previous cyberattacks

The discovery comes after security companies connected Artex to several cyber incidents. CrowdStrike previously attributed attacks against South Korean financial institutions to a Chinese-speaking threat actor that used an Artex instance. Censys said the same infrastructure had remained publicly exposed for months and contained data extending beyond the reported attacks on South Korean banks.

Artex was also linked to an October intrusion involving IT support company ConnectWise. Microsoft reported that Chinese-linked hackers used the tool during attacks on customer cloud environments, including activity that enabled the theft of data. These incidents demonstrate how autonomous penetration-testing capabilities can be repurposed in malicious operations.

Exposed AI hacking infrastructure creates additional risks

Artex can automate stages of penetration testing, from reconnaissance to identifying weaknesses and carrying out further actions. Microsoft has previously described the system as capable of progressing from reconnaissance to data exfiltration with limited human involvement.

However, the latest Censys findings are primarily about exposed deployments and insecure configurations, not a newly discovered vulnerability in the AI tool itself. The risk comes from making an operational interface and potentially sensitive supporting files accessible to outsiders.

The findings also show why organisations need to secure AI-powered security tools with the same care as other sensitive infrastructure. Restricting access to authorised users, avoiding unnecessary public exposure and checking supporting directories can reduce the risk of unauthorised use.

Artex has since announced a move towards closed-source distribution for paid customers. However, Censys noted that copies of its code remain available on code-sharing platforms. The discovery therefore raises a broader question for cybersecurity teams: as AI tools make offensive capabilities easier to automate, securing the infrastructure that operates them becomes just as important as securing the systems they are designed to test.