AI news, problem solutions & tool picks
DigestX

AI News

More Than 800 Artex AI Hacking Tools Found Exposed on the Open Web

· 4 min read

A cybersecurity expert inspecting code on multiple monitors in a dark room
Photo via Pexels

Security researchers have found more than 800 copies of an AI hacking tool sitting publicly exposed on the internet, with no password needed to open them. The finding, reported by cybersecurity firm Censys in a blog post on November 6, means anyone with a browser could reach a working AI attack toolkit and point it at a target.

Censys, which scans the internet for exposed or vulnerable systems, counted 1,465 hosts running an exposed Artex console as of November 4. About 650 of those sit in two blocks of identically configured hosting, which leaves roughly 815 distinct-looking deployments reachable from the open web. The console does not require public access at all, which makes the exposure a misconfiguration, not a design choice.

What Artex is

Artex is a Chinese-made, open-source autonomous penetration testing tool, published on GitHub on July 26, 2026, under the AGPL-3.0 licence. It wraps a large language model in a harness that hands it a command shell, a shared notes file for tracking what it finds, and a planner that splits a break-in into tasks for several AI agents working in parallel.

Microsoft has described Artex as a system of AI agents that can conduct cyberattacks “from reconnaissance through data exfiltration with little needed human interaction.” As open-source software, it is free for anyone to download and run on their own servers, which has made it popular with people who cannot access the most advanced cybersecurity AI models from Anthropic or OpenAI.

What the exposed systems contain

The publicly accessible consoles do not just show a login page. They expose attack tooling, including implants — hidden malicious software used to compromise a victim's network and maintain persistent access. Censys also found open directories on 32 Artex hosts, exposing post-exploitation kits, stagers, exploit binaries and operator files to anyone who browsed them.

The exposure is not static. Between November 3 and November 5, ten exposed consoles closed and fifteen new ones appeared on other hosts, which suggests operators are still deploying the tool without fixing the access controls. Censys found the consoles by their page title and the Next.js static assets they serve, and it notes that overall global deployment is likely higher than what it counted.

Of the 1,465 exposed hosts, 155 — about 10.6% — also run at least one other threat that Censys detects, including the remote-control tools Cobalt Strike, Vshell and Adaptix C2, as well as other AI attack frameworks, reconnaissance platforms and tunnelling tools. By country, the United States had the most exposed consoles at 726, followed by Hong Kong with 387, China with 177, Japan with 51, Singapore with 38 and South Korea with 21.

The attacks already linked to Artex

Security companies have tied Artex to several broad-scale cyberattacks since its public release in July. CrowdStrike reported on October 7 that an unattributed actor used Artex against South Korean financial institutions from late September to early October, and assessed with moderate confidence that the actor was a financially motivated Chinese speaker. The attacker had left an open directory hosting the system, and Censys said its data showed that instance had been publicly exposed for months — since August 18 — and appeared to contain attack data extending “well beyond South Korean banks,” including payment and gaming companies.

Artex was also used in an attack on IT support company 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 attributed the compromise to the group it calls Storm-0940 and said the group used Artex for scanning, password spraying and OAuth application abuse across more than 2,250 tenants, stealing data at such speed that defenders had “little time to detect and respond.”

Many of the attacks linked to Artex involved exploiting simple vulnerabilities, such as guessing weak or known passwords and a lack of multi-factor authentication. The Censys findings show the same lesson on the other side: simple misconfigurations can expose both the attacker and their targets.

Not every copy is malicious, and closing the source will not help

Silas Cutler, the principal security researcher at Censys who wrote the blog post, said that not every instance of Artex is operated maliciously and many are used by authorized penetration testers. The problem is that hundreds of these systems are exposed to the web for no reason, since the software does not need to be accessible from the internet at all.

Artex's creator, a Chinese cybersecurity engineer who goes by the alias Autumn, took the project private earlier this fall after South Korean investigators linked it to the bank breaches, and has since said the tool will go closed-source, limiting further releases to paying customers. That change does nothing for the 1,465 consoles Censys found still running and still exposed as of November 4, since those are existing deployments, not new downloads. Dozens of copies of the code remain available on code-sharing sites and can still be used in active attacks.

The finding is a reminder that the most dangerous part of an AI hacking tool is not always the model inside it. Sometimes it is the default port — 8787, in Artex's case — left open to the whole internet, with no password on the other side.