AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Hundreds of Artex AI hacking tools found publicly exposed to the web

Collected Nov 6, 2026

Censys, a cybersecurity firm that continuously scans the internet for exposed systems, reported finding more than 800 distinct Artex deployments that anyone could reach from the web without a password. Silas Cutler, a principal security researcher at Censys, wrote the blog post announcing the finding. Artex is an open-source autonomous penetration testing tool, released publicly in July, that has since been linked to several cyberattacks.

Penetration testing software is normally used by defenders to probe their own systems for weaknesses before attackers find them. Artex automates that work with AI agents. Microsoft has described it as a system of agents capable of carrying out attacks from reconnaissance through data exfiltration with little human involvement. The tool is Chinese-made and free to download and run on your own servers, which made it attractive to people who cannot get access to the most advanced cybersecurity AI models from Anthropic or OpenAI for vulnerability discovery.

Cutler noted that not every exposed Artex instance is run maliciously and that many are used by authorized testers. What remains unexplained is why hundreds of them sit on the open internet at all, since nothing about the software requires internet-facing access. This suggests the exposure is mostly a configuration problem rather than an intentional choice — the same class of mistake as an unauthenticated database or an open admin panel, except here the thing behind the missing login is an attack platform.

The consequences go beyond someone else borrowing compute. Publicly reachable systems also expose the tooling stored on them, including implants — hidden malicious software used to compromise a victim's network and keep persistent access. In practice, an open Artex instance could let an outsider read whatever campaign data lives on that server, or point the tool at targets of their own choosing. Censys' scan approach means these systems are discoverable by anyone running similar internet-wide searches, not just by researchers.

There is already evidence of real-world abuse. Security companies have connected Artex to several broad-scale attacks, including one against the Korean finance industry in September. CrowdStrike attributed those attacks to a Chinese-speaking hacker who used an Artex instance, partly because the attacker left an open directory hosting the system. Censys said its data showed that same instance had been publicly exposed for months and appeared to hold more attack data extending well beyond South Korean banks, including payment and gaming companies.

Artex also figured in an attack on IT support giant ConnectWise earlier in October, which allowed Chinese-linked hackers to spy on thousands of organizations whose data and files are hosted in Microsoft's cloud. Microsoft, which had visibility into the incident, said in a blog post that the hackers used Artex to break into customer cloud environments and steal data fast enough that defenders had little time to detect and respond.

Many of the attacks tied to Artex relied on unglamorous weaknesses: guessed or known weak passwords and a lack of multi-factor authentication. That is a useful corrective to the idea that capable AI attack tooling only matters against sophisticated targets. A password-spray or credential-stuffing campaign against cloud tenants without MFA is cheap to run and hard to distinguish from legitimate logins, and automating reconnaissance and lateral movement shortens the window in which defenders can notice. The Censys finding cuts both ways: the same simple mistakes that let attackers in can leave the attacker's own infrastructure sitting in the open.

Artex has announced it will move to closed-source development, which would effectively limit further releases to paid customers. Even so, dozens of copies of its code remain available on code-sharing sites and can still be used in active attacks. Open-sourcing a dual-use tool is not reversible in any practical sense; once the repository has been cloned, withdrawing the source changes who gets updates, not who can run the software.

Why it matters: Security teams and cloud administrators are the people most directly affected, because exposed Artex instances and the credential weaknesses the tool exploits both sit on infrastructure they control. The practical work is unglamorous: inventory any AI security tooling running on your networks, confirm it is not reachable from the internet without authentication, and close the MFA gaps that made the linked attacks work. For developers building or self-hosting agentic tools, the lesson is that an autonomous system with network access inherits every deployment mistake around it — my inference is that default-deny network exposure and mandatory authentication should be treated as baseline requirements, not optional hardening, for anything that can act on its own.

Read at TechCrunch · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Security researchers found hundreds of servers hosting the penetration testing tool Artex exposed to the internet, allowing anyone access to the hacking tool.