Guide · AI security · as of 14 November 2026

The FBI warns of an AI attack agent. It goes for the apps you forgot. Four checks.

On 13 November 2026 the FBI published a public alert about ARTEX. It is an open-source AI agent built for security testing: you describe a goal in plain words and it plans and runs the steps itself. The FBI says criminals have used it since at least August 2026, and that some break-ins went from first contact to administrator access in under three hours. The part that matters for most businesses is where it looks. This guide works from the FBI’s own alert and turns its advice into four checks you can ask your team or your developer to run this week.

What the FBI says

A testing tool, turned around.

ARTEX does not carry its own AI model. It connects to the big model providers, and the FBI lists Claude, GPT and several Chinese models among them, with DeepSeek’s Flash models the most common on the attackers’ own copies. Its author removed the code from GitHub in October 2026. The FBI says copies, local versions and a paid edition are still in circulation, and that researchers counted more than 1,600 ARTEX control panels open on the internet in early November. The alert cites one campaign that took personal data on about 67,000 people from banks in South Korea, and another that went through a cloud backup provider to more than 2,250 of its customers.

An AI agent escaped a sandbox: five checks

Where it looks

Not the front door. The side doors.

The FBI says ARTEX is best at systems that get less attention than the main website: partner and supplier portals, the back end of a mobile app, staff apps, sales tools, old services and third-party connections. The ways in it lists are not new. An administrator password left inside a website’s JavaScript. A web API that never asks who is calling. A customer page where changing the number in the address shows another customer’s record. Weak admin passwords. What is new is that an AI agent tries all of them, on every door, at machine speed.

Four checks from the FBI’s advice

Run these this week.

  • 1. List every app you have on the internet, not just the website: portals, staff apps, mobile back ends, sales tools, test copies. Test: for each one, write down who owns it and when it was last updated. Anything with no owner is the first door to close.
  • 2. Search your code and your live site for secrets. The FBI saw administrator passwords sitting inside website JavaScript. Test: have your developer run a secret scanner over the code, the build output and the deployment settings, and rotate any key it finds.
  • 3. Make every portal check each record and count requests. Test: sign in as one customer, change the record number in the address bar, and confirm you are refused. Then confirm a burst of rapid lookups is slowed down or raises an alert.
  • 4. Keep admin and publishing doors off the open internet, and keep errors plain. Test: from outside the office, try your admin and deployment pages; they should not load. Open a wrong address on your app while signed out; you should see a plain page, not code, database errors or file paths.

What we checked on our own

We read our own JavaScript first.

On 14 November 2026 we downloaded the JavaScript that loads with the front page of four apps and sites we run in production, about 2.4 MB of code, and searched it for API keys, service keys and passwords. We found none. That is a narrow check, and we say so: it covers the code that loads first, not every file an app can fetch later. The point of the FBI’s alert is that an attacker’s agent will run the same search on your apps whether you do or not.

AI transformation in Dubai, one agent at a time

Questions

Questions about ARTEX.

  • What is ARTEX?

    An open-source AI agent made for authorised security testing. You give it a goal in plain language and it plans and runs the steps using an AI model from an outside provider. The FBI’s alert of 13 November 2026 says criminals have turned it into an attack tool.

  • Is ARTEX a virus I can scan for?

    No. It runs on the attacker’s own machine and talks to your systems from outside, the way a person would, only faster. You defend against it by closing the doors it tries, not by scanning your laptops for it.

  • Is this only a problem for big companies?

    No. The FBI says it lowers the skill needed to attack, so small groups can run many campaigns at once. Small firms often have the side apps it looks for, such as an old portal or a staff app, with nobody watching them.

  • Does it affect businesses in the UAE?

    The alert names targets in many countries and sectors, including finance, government, health care and manufacturing. It does not list UAE cases. The doors it tries are the same in Dubai as anywhere else.

  • What should I do first?

    Make the list in check one. You cannot protect an app you have forgotten, and the FBI says forgotten apps are exactly where this agent does best.

Sources

read 14 November 2026

Check it at the source.

  1. FBI FLASH-20261113-001: Threat Overview of ARTEX Autonomous Penetration Testing Framework (the full alert: how it works, what was seen, and the FBI’s recommendations)
  2. BRev Digital: An AI agent escaped a sandbox, five checks (our earlier guide on a test agent that reached a real internal tool)
  3. BRev Digital: Your AI server trusts localhost, four checks (the same lesson from the other side: AI software you host is a door too)

A question?

Not sure which of your apps are exposed?

Send us the list you have, even if it is short. We will help you find the ones missing from it, check each against the FBI’s four points, and tell you plainly what to close first.

Talk to usWhatsApp +971 50 728 5494hello@brevdigital.com