Back to Newsroom

Threat intel6 min read

FBI warns attackers are using the ARTEX AI framework to break in within hours

An AI-driven pentest framework released in July 2026 is now used by criminal and state-aligned groups. It needs no zero-day: it works through exposed portals, admin interfaces and leaked credentials, sometimes reaching privileged access in under three hours.

4Tify Team

The FBI has issued a FLASH alert (FLASH-20261113-001, dated 13 November 2026) about ARTEX, a freely published penetration-testing framework that hands an AI agent a set of hacking tools and lets it work through a target on its own. Criminal groups and state-aligned actors have used it against organisations worldwide since at least August 2026. In some intrusions the attackers went from first contact to privileged access in under three hours.

None of this depends on a zero-day. ARTEX goes after the ordinary gaps most organisations already have: forgotten portals, exposed admin interfaces, leaked secrets and permissive cloud credentials.

What ARTEX is

ARTEX was posted on GitHub in July 2026 as a tool for authorised security testing. It is written in Go and has no AI model of its own. The operator plugs in a commercial or open-weight language model (the FBI lists Claude, GPT, DeepSeek, GLM, Grok and Kimi), describes a goal in plain language, and a group of specialised agents turns that goal into commands. Each result feeds the next decision, so the tool keeps following leads without waiting for a person.

The developer took the repository down on 8 October 2026 after reports of abuse. That did little. Forks, archived copies, localised versions and a paid edition called ARTEX PRO are still in circulation, and in early November more than 1,600 ARTEX consoles were reachable on the public internet.

What attackers have done with it

The FBI alert lists victims in finance, government, technology, defence, semiconductors, energy, healthcare, manufacturing, telecoms and the legal sector. Two campaigns stand out:

  • An attack on South Korean financial institutions in which personal data of about 67,000 people was stolen.
  • A supply-chain compromise of ConnectWise Cloud Backup that led to sign-in attempts against more than 2,250 downstream Microsoft 365 tenants.

A financially motivated cluster tracked as Storm-3358, based in Cambodia and focused on banks and other financial firms in Southeast Asia, started using ARTEX in October. It scans internet-facing systems, exploits weaknesses such as anonymous file upload and server-side request forgery, and installs remote access trojans (WISP, vshell and AdaptixC2) to stay inside.

The work itself looks familiar: Nmap, curl, wget, Python HTTP clients and a headless browser. What changes is the patience. An agent will spend hours probing a partner portal, a mobile backend or a legacy API that a human attacker would have skipped as not worth the effort. The FBI singles out exactly these peripheral systems as ARTEX's favourite ground.

The tool is a risk to its own users too

Some third-party ARTEX Docker images were backdoored. The images tomicdd/artex and congdonglinux/artex carry a hidden routine in the login handler that gives anyone who knows a secret value a root shell inside the container, with no login needed. There is no evidence that the original ARTEX code contained a backdoor. The problem sits in prebuilt copies, which is why reading the code would not reveal it.

There is also the question of scope. WorkOS said an ARTEX agent run by a researcher during benchmark tests left its test environment and reached one of the company's internal sales tools, then pulled customer records. That tool had an API open to the internet that accepted a predictable development credential, and nothing limited bulk extraction. How the agent got out of its environment has not been disclosed.

What to do now

  1. Find everything you expose. List every internet-facing application, including supplier and partner portals, mobile backends, employee tools and old services that no one owns any more.
  2. Pull admin surfaces off the internet. Cloud consoles, deployment endpoints (such as Azure Kudu/SCM), hypervisors, backup platforms and device management should only answer from trusted networks or through an identity-aware proxy or VPN.
  3. Limit what one stolen credential can do. Apply least privilege to app permissions and consent grants, and alert when credentials or privileged roles are added to applications.
  4. Rate-limit sensitive APIs. Watch for sequential lookups and unusually high record counts per session on portals, and keep stack traces and detailed errors away from unauthenticated users.
  5. Hunt for leaked secrets. Scan repositories, container images and CI/CD pipelines, and rotate anything you find.
  6. Watch for the noise. ARTEX traffic is fast and loud: bursts of requests, retries with small variations, unexpected calls to AI provider APIs from servers that have no reason to make them. Default user-agent strings can help spot it, but an operator can change them in seconds.
  7. If you run ARTEX yourself, use only images you built from reviewed code, keep it on a disposable machine with short-lived credentials, and enforce scope at the network level with an egress allow-list. A prompt is not a boundary. CISA's guidance on adopting agentic AI covers the same controls.

If you find signs of ARTEX activity, keep the logs and report it to the FBI or your national CERT.

Most of the doors ARTEX opens are ones an organisation did not know were there. This is the problem our Attack Surface Management service is built for: it maps the subdomains, portals and admin interfaces you expose and flags new ones as they appear, so you see them before an agent does.

NEWSLETTER

The week’s news,
every Friday.

A short email with what we published that week. You can unsubscribe from any of them.

FRIDAY DIGEST · SAMPLE4TIFY NEWSROOM

This week in security

  • 01Attackers exploit critical Cisco Expressway flaw to gain root on clustered nodes
  • 02SonicWall ends Gen 6 firewall patches, even for critical flaws
  • 03WhatsApp fixes three flaws, one possibly used to back up scam messages

+ 6 MORE IN THIS ISSUE · UNSUBSCRIBE IN ONE CLICK

ONE OFFER

More than one product?
Get one offer.

Tell us what you need to protect and your budget, monthly or yearly. We combine the products into one offer and one contract, sized to that budget, and we stay with you after it is signed.

Ask for an offer Just tell us how many assets you have and what you want to spend.
FBI FLASH: ARTEX AI pentest framework abused by attackers | 4Tify