ARTEX AI Hacking Tool: What the South Korean Bank Breaches Reveal About Exposed Systems, What Is Confirmed and What to Check This Week

Attackers used an open-source AI hacking tool called ARTEX in a string of break-ins at South Korean banks and financial companies in late September and early October 2026. That is according to reports from CrowdStrike and the threat-hunting firm Hunt.io. It matters well beyond Korea because the attacker combined free software with commercial AI models to probe ordinary internet-facing systems: the same kind of portals, partner lookup services and staff apps most companies run today.

What happened with ARTEX and the South Korean banks

CrowdStrike published its findings on October 7, 2026. It says it found infrastructure linked to a targeted campaign against South Korean financial organizations that "resulted in exfiltrated data." The campaign ran from late September to early October 2026. CrowdStrike gives two examples. At one bank, the attacker reportedly broke into a loan progress inquiry service used by financial brokers. At another, the attacker got into an employee mobile work-support system.

In early November, Hunt.io published a closer look at the attacker's own servers. Hunt.io says seven South Korean financial institutions were breached through "internet-facing peripheral systems." It puts the toll at approximately 66,000 to 67,000 individuals and around 2,200 corporations, and says personal, employment, income and loan-related information was exposed.

What is ARTEX?

CrowdStrike describes ARTEX as "a recently released open-source agentic penetration testing (pentesting) tool developed in China." Penetration testing tools are built for authorized security testing. An agentic tool hands much of the work to an AI model: scanning, deciding what to try next and running commands.

In this campaign, CrowdStrike says ARTEX ran mainly on the DeepSeek v4.1-flash model. The attacker also used GLM-5.3 and Grok 4.6 in additional Claude Code sessions. Hunt.io adds that the setup ran Claude Code but sent its requests through proxies to DeepSeek, not to Anthropic. Hunt.io, citing the Korean security firm Genians, says the original ARTEX GitHub repository and Docker image were taken down on October 11, 2026. Taking a project off GitHub does not, by itself, remove copies people have already downloaded.

What is confirmed and what is still claimed

Coverage of this story mixes firm findings with estimates. Here is how the primary reports line up.

CrowdStrike bases its assessment on the attacker's use of a Chinese-developed tool and Chinese-language prompts. It also says the attacker asked Claude where Korean breach data is typically sold. Some things are still not publicly confirmed: who the attacker is, how many organizations and people were hit, and exactly what was stolen.

What Hunt.io found on the attacker's servers

The attacker left open directories on their own servers, and much of this came to light that way. Hunt.io reports:

• A main server at 38.244.50[.]120 that hosted the ARTEX console, and a jump server in Hong Kong at 23.248.226[.]250 with an open directory of 42 files.
• Root SSH credentials for that jump server, stored in a file, along with AI service API keys and proxy service logins.
• A staged JNDI/LDAP remote code execution setup, prepared on October 5, 2026.
• Builds tied to two Linux privilege escalation flaws, CVE-2022-2588 and CVE-2025-5054, plus a script that suggests which Linux exploits to try.
• Search results showing what appear to be ARTEX servers in many countries, not only the ones used in this campaign.

Hunt.io also says that no personal records or credentials from the affected companies appear in its report.

Why AI hacking tools matter for a business network

None of the systems named here are exotic. A broker lookup service and a staff mobile app belong in the same category as a client portal, a remote access gateway or a web server admin panel. Tools like ARTEX change the cost. An AI agent can keep scanning, testing logins and trying known exploits without a skilled person steering every step. That makes forgotten servers, unpatched web apps and reused passwords cheaper to find and abuse.

For a company network or website, "nobody would bother with us" is no longer a defense. If something can be reached from the internet, assume automated tools will find it and test it.

What to do this week

These steps follow Hunt.io's own defensive advice, ordered for a typical business.

To check for the published indicators, search firewall and web server logs for connections to 38.244.50[.]120 and 23.248.226[.]250. The full indicator lists are in the CrowdStrike and Hunt.io reports.

For help cutting down what you expose, see our security hardening and self-hosting infrastructure services.

ARTEX AI hacking tool questions people are asking

Is ARTEX malware?

No. CrowdStrike describes it as an open-source AI penetration testing tool. Like many testing tools, attackers can use it too.

Were the South Korean banks hacked by AI?

An attacker ran the campaign but, according to CrowdStrike, used ARTEX and several AI models to do much of the work.

Who is behind the South Korean bank breaches?

No named group has been blamed. CrowdStrike assesses with moderate confidence that the actor is likely a Chinese speaker and financially motivated.

How many people were affected by the South Korean bank hack?

Hunt.io reports about 66,000 to 67,000 individuals and around 2,200 corporations across seven institutions. CrowdStrike says the number of affected organizations is unconfirmed.

Does ARTEX affect businesses outside South Korea?

The reported victims are in South Korea. Hunt.io's searches found what appear to be ARTEX servers in many countries, so the tool is in wider use.

Can a firewall block ARTEX?

Blocking the published IP addresses helps a little, but attackers change servers. Exposing less, patching and strong logins matter more.

The bigger lesson

AI did not create new ways in here. It made the old ones faster and cheaper to use: exposed services, missed patches and passwords stored in files. The fix is the same discipline as before, done consistently.

Want to know what your business exposes? Start with security hardening or request a quote.

Get security alerts like this by email, with the free checklist, at security alerts.

Sources: CrowdStrike; Hunt.io.