Hackers Use AI-Powered ARTEX Framework to Automate Cyberattacks Without Relying on Zero-Day Exploits

Blog WriterCybersecurity News - Original News Source is cybersecuritynews.com


Attackers are using an AI-driven testing tool called ARTEX to break into company networks. It needs no secret software flaws to do this.

Instead, it hunts for ordinary mistakes such as exposed apps, weak passwords and loose permissions, and it keeps working at them far longer than a human would.

ARTEX is an open-source framework that connects a large language model (LLM) to a set of hacking tools. It was built for authorized security testing, but criminals and state-linked groups now use it.

The FBI warned that hackers have used it since at least August 2026 and, in some cases, reached privileged access in under three hours.

Researchers at TrustedSec noted that ARTEX works like a familiar scanner with an AI brain attached. The framework gives a model tools to inspect a target and act on what it finds. Each result feeds the next decision, so the tool can follow a lead without waiting for a person.

Hackers Use AI-Powered ARTEX Framework

The attack flow is not new. According to TrustedSec, attackers still find assets with subfinder, check web services with httpx, map services with Nmap, run nuclei templates and use ffuf to find hidden content.

What changes is the work after that first scan. The model picks an action, the framework runs it, and the output becomes the next input.

It may adjust a request with curl, write a small script or check an exposed file. A secret it finds becomes a new question: is it valid, and what can it reach?

Microsoft showed the tool paired with Nmap, curl, Python HTTP clients and Playwright. It tracked Storm-3358 using ARTEX to scan internet-facing systems and exploit flaws like anonymous file upload and SSRF.

Mandiant also logged rapid AWS permission probing. Other groups have used it to automate Active Directory attacks and move sideways inside networks.

The real shift is cost. Many firms have custom apps, old APIs and forgotten subdomains that nobody tests closely. Attackers used to skip these because one exploit for a common product hits thousands of targets, while a custom app pays off once.

TrustedSec called limited attacker attention an economic assumption, not a control. An AI agent can spend hours on every odd response for little money.

The tool also has weak points. It can burn many tokens on a false lead, and repeating a wrong guess does not make it right. Broad scanning and fast retries also create noise that defenders can spot.

Backdoored Copies and Sandbox Escapes

ARTEX is risky even for people who run it legitimately. The tool holds model API keys, target data and captured sessions, so it becomes a rich target itself.

TrustedSec described a third-party ARTEX copy that pointed users to backdoored Docker images. The changed login handler allowed unauthenticated command execution inside the container. The finding covers those copies only.

It does not show that the original ARTEX source held a backdoor. The WorkOS case shows the other side of the problem. The company said an ARTEX agent used by a researcher running benchmark tests reached a real internal tool.

Reports say the agent escaped a sandbox and pulled sales and support records. The tool had a public API, a predictable development credential and no rate limits.

WorkOS has not shown a true container flaw. An agent with internet access can leave its intended scope without breaking out of anything.

Defenders can act on several points:

  1. Move one exposed employee tool or management API behind an identity-aware proxy, private access service or VPN, then confirm the original server cannot be reached directly.
  2. Set limits in the WAF and application controls for path enumeration, repeated exploit probes and bulk record retrieval.
  3. Trim the permissions of public web apps so one flaw does not open the wider network.
  4. Test whether the team can link odd API use to an account and revoke access while retrieval is still running.

User-agent blocking helps only against default setups, since an operator can change the string. Anyone testing ARTEX should use a disposable VM, temporary credentials and a lab network whose outbound policy denies unrelated destinations.

Scope rules should be enforced by the network, not by a prompt. Security teams should also watch for frameworks exposed on public servers, because open consoles give attackers a free starting point.

The original developer pulled the public code in October after reports of abuse, but forks keep circulating. Defenders should not wait for a zero-day alert, because these intrusions rely on gaps that already exist.

Indicators of compromise (IoCs):-

Type Indicator Context
Docker image tomicdd/artex Backdoored third-party ARTEX image
Docker image congdonglinux/artex Backdoored third-party ARTEX image
User agent ArtEx/1.0 Literal string in WorkOS indicators
File hashes Not published None listed in the source

Note: IP addresses and domains are intentionally Qdefanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms sQuch as MISP, VirusTotal, or your SIEM.

Give your SOC the alert context on the latest threats to stop incidents quickly -> Integrate Threat Intel from 16K SOCs

The post Hackers Use AI-Powered ARTEX Framework to Automate Cyberattacks Without Relying on Zero-Day Exploits appeared first on Cyber Security News.