Blog / Front Line — part 21 of 25

ARTEX: what the FBI AI attack warning means for defenders

· urgentic · 5 min read

Short read

This post covers the FBI's new warning about ARTEX, an open-source AI hacking framework that attackers are using to run intrusions at machine speed, in some cases going from first contact to privileged access in under three hours.

You'll learn what ARTEX is, what recent campaigns tell us about where it finds its way in, and the practical checks the FBI, Amazon and Cloudflare recommend for your own environment.

The FBI has issued a FLASH alert warning that attackers are using ARTEX, an AI-driven penetration testing framework, to automate cyber attacks against organisations around the world. The ARTEX warning matters because it changes the clock defenders work to: the FBI says some ARTEX intrusions have gone from initial targeting to privileged access in under three hours. If your detection and response depend on someone getting round to an alert the next morning, that gap is now the attacker's advantage.

What the FBI said about ARTEX

In FLASH-20261113-001, published on 13 November, the FBI describes ARTEX as an open-source, large language model (LLM)-driven agentic platform that a variety of malicious actors have used since at least August 2026. It says the tool automates "multiple phases of cyber intrusions at machine speed, from reconnaissance through data exfiltration".

ARTEX doesn't have its own AI model. The operator gives it a goal in plain language, and it turns that goal into commands carried out by specialised AI agents, using external models from several providers. According to the FBI, this lets threat actors "with minimal technical expertise" run complex campaigns against several targets at once.

Targeted sectors listed in the alert include financial services, government, technology, defence, semiconductors, energy, healthcare, manufacturing, telecommunications and legal.

BleepingComputer reports that the original GitHub repository was taken down on 8 October after reports of abuse, but the framework is still available through community forks, archived copies and a commercial edition, and researchers had found more than 1,600 publicly exposed ARTEX consoles by early November.

Where ARTEX gets in

The most useful part of the alert for defenders is where these agents look. The FBI says ARTEX "excels at identifying and exploiting peripheral systems receiving less security attention than flagship applications", including:

  • partner and supplier portals
  • mobile backends and employee applications
  • sales-support systems and legacy services
  • third-party integrations and apps built by individual business units

It also highlights supply chain routes: weaker vendors, CI/CD pipelines and integrated software providers that give a path into a bigger organisation. The alert links ARTEX to the October compromise affecting ConnectWise Cloud Backup customers, which it says "enabled access to more than 2,250 downstream tenants", and to attacks on South Korean financial institutions that exposed personal information on around 67,000 people. We covered the ConnectWise incident in ConnectWise breach: why Microsoft 365 app permissions matter.

What the latest research adds

Two reports published in the past few days show how this plays out in practice.

The Hacker News reports that Amazon's threat intelligence team captured records from attackers' AI agents using decoy servers posing as popular LLM services. Amazon has moderate confidence the agents ran on ARTEX. In the main campaign, they hunted for internet-exposed AI apps and gateways, pulled out the AI provider keys stored inside, checked the keys worked and added them to a shared pool. Amazon says it can't verify the outcomes the agents reported, but the records show what the operators were after.

The same report describes Cloudflare incident responders seeing ARTEX use a compromised account-level API token that had been used for CI/CD. The token had no IP restrictions or rate limiting, and the activity that followed included 400 database dump requests in an hour. Cloudflare says the common root cause across several incidents was static CI/CD tokens that never expired.

Amazon's own summary is the key point: "None of the techniques in these campaigns were new. What AI agents change is the speed and persistence of the attacker." The agents succeeded where basic security was missing, such as default credentials, keys left in code, endpoints without authentication and unpatched software.

Why speed changes the defence

AI-driven attacks aren't stealthy by default. BleepingComputer notes that the automated activity can generate large volumes of suspicious traffic that trigger alerts. The problem is time. When an attacker can try many routes in parallel and pursue a different one as soon as the first fails, an alert that waits in a queue for hours may be read after the attacker already has what they came for.

That shifts the focus to two things: removing the easy openings agents find first, and making sure the signals they do generate are looked at and acted on quickly, at any hour.

What to check now

Drawing on the FBI's recommendations and the Amazon and Cloudflare findings:

  1. Inventory everything exposed to the internet, not just your main website. Include partner portals, test and legacy systems, mobile backends, and apps set up by individual teams.
  2. Restrict management interfaces. The FBI advises limiting cloud administration, network device management, hypervisors, backup platforms and publishing interfaces to trusted networks or dedicated admin paths.
  3. Hunt for exposed secrets. Check configuration files, environment variables, deployment scripts, CI/CD pipelines and repositories for embedded keys, scan continuously, and rotate anything exposed promptly.
  4. Give tokens an expiry and limits. Replace static, never-expiring CI/CD tokens, restrict where they can be used from, and apply rate limits to sensitive APIs.
  5. Treat AI apps like production systems. Change default admin credentials on LLM gateways and AI app builders, don't leave unfinished installs exposed, and watch for unexpected AI usage or spending.
  6. Watch identity and app permissions. Apply least privilege to multi-tenant app permissions and consent grants, and monitor for new credentials, privilege assignments and unusual app-only or service-principal activity.
  7. Alert on unusual bursts. The FBI recommends monitoring for bursts of automated requests, suspicious cloud activity, unexpected credential or privilege changes and large data transfers.
  8. Use the published indicators carefully. The FBI alert includes IP addresses and other indicators, but cautions that they may have changed since the attacks. Correlate them with your own logs rather than relying on blocklists alone.

If an MSP or MSSP looks after your environment, ask them how they're tracking the ARTEX indicators, how quickly an out-of-hours alert reaches someone who can act, and whether your exposed services and tokens have been reviewed.

The bigger picture

ARTEX is unlikely to be the last framework of its kind. The same report from The Hacker News says Cloudflare saw more than 50 families of AI penetration testing frameworks try to scan its customers over three days. The techniques are familiar, which is good news: the basics still work. What's changed is how little time there is between a weakness being found and being used. Closing obvious gaps, and having round-the-clock eyes on the alerts that remain, is now the baseline. You can read more about how urgentic approaches continuous detection and investigation on our platform page.

  • ARTEX
  • FBI
  • AI attacks
  • agentic AI
  • attack surface
  • secrets management
  • API tokens
  • supply chain
  • SOC