TL;DR
- AI attack frameworks like ARTEX make attackers faster and let them run many attacks in parallel. What an attacker has to do to break in hasn’t changed.
- Detections we already have in place are catching activity from these frameworks in customer environments.
- We’ve added targeted detections, run two emerging threat hunts, and stood up a dedicated research group to stay ahead of what comes next.
ARTEX lets attackers chain a full intrusion together and run it at machine speed. The steps are the same ones attackers have always needed: sign in, move around, run code, and find your data. Expel has spent 10 years building detection and response around exactly those steps.
What’s going on
ARTEX was released publicly in late July 2026. It uses large language models (LLMs) to chain together the stages of an attack (scanning, credential abuse, and post-compromise activity) and run them at machine speed across cloud and on-premises environments. Work that used to take an attacker days can now take hours.
Attackers started using it shortly after release. The author pulled the public repository, but by then people had copied it many times. New frameworks followed—some built on ARTEX and others written from scratch—and several are in active use today.
Why our existing detections still hold up
These frameworks automate attacks and speed them up. They don’t invent new kinds of attacks. Every attack, whether a person or a machine runs it, still has to authenticate, cross boundaries, manipulate processes, and touch data. Those are the behaviors our detections are built to find.
That’s why we don’t rely on indicators of compromise (IOCs) alone. An AI framework can generate new IP addresses, file hashes, and domains faster than anyone can block them. It can’t skip the behaviors an attack requires. So far, the evidence backs this up: detections we already had in place are catching the attacks these frameworks run.
What we’re doing now
New targeted detections
On top of our existing library, we’ve added detections aimed specifically at AI attack frameworks:
- Framework fingerprints. Detects the user-agent strings ARTEX and related frameworks use. We update it as new versions appear.
- Campaign-specific activity. Detects indicators tied to known campaigns, including Storm-0940 targeting Microsoft 365 through a compromised ConnectWise (SkyKick) app, attacks on South Korean financial networks, and other actors using AI-augmented tooling.
- Web shells. Detects web shell variants attributed to ARTEX and similar frameworks.
Most of what these frameworks do is already covered, so new detections address coverage gaps..
Emerging threat hunts
We’ve run two hunts across customer environments:
- ARTEX agentic AI attacks. This hunt is focused on the Storm-0940 campaign and the ConnectWise (SkyKick) Microsoft 365 app compromise.
- ARTEX framework abuse across multiple actors. A broader hunt covering several campaigns and rebranded attack consoles.
Both hunts confirmed our detections are catching this activity, and a detection built from the second hunt has since caught a live incident. Most ARTEX-driven attacks so far are individual and targeted, and existing detections catch them.
Threat intelligence
Our threat intelligence team tracks these campaigns and how the frameworks evolve. They pull from three sources:
- Attacks hitting our own customers
- External feeds focused on AI attack tooling
- Industry information-sharing groups we belong to
We were tracking this before ARTEX appeared. AI-driven attack tooling was an expected step in how attackers work, and it has shaped our detection and automation strategy for some time.
How we run the SOC
- Mapped to industry frameworks. Our detections map to MITRE ATT&CK and MITRE ATLAS, the industry frameworks for traditional and AI-specific attack techniques. That shows us our coverage at a glance and tells us when attack patterns shift.
- AI working alongside our analysts. We use AI agents inside our SOC to triage alerts, pull together context, and run response playbooks. Our human analysts still make the critical judgments. That’s how we keep pace with automated attacks.
- Continuous checking. We review detection performance daily and examine trends weekly and monthly. We tune detections as needed so our telemetry and detections keep working as threats change.
What we’re doing next
We’ve formed an applied threat research working group with experts from across Expel. Its job is to stay ahead of ARTEX and the frameworks that follow it. They are:
- Closing gaps now by testing our detections against ARTEX-style attacks and fixing anything they miss.
- Expanding visibility by identifying any new telemetry or capabilities we need to see these attacks earlier.
- Preparing for the next framework by getting ready for attack kits that haven’t been released yet, so we aren’t starting from zero when they are.
- Running targeted research scoping focused projects that strengthen our defenses against this class of attack.
- Keeping customers informed by sharing what we learn and what we change, with you and across Expel.
What you can do
Three things make the biggest difference against these frameworks, and against threats in general:
- Turn on auto remediation. Automated attacks need an automated response. In past incidents, auto-remediation has significantly reduced the damage attackers could do.
- Keep your coverage current and complete. We can only detect what we can see. As attacks get more automated, signals from more of your environment let us catch them earlier and connect activity across systems.
- Apply least privilege everywhere you can. Limiting what each account can access slows an attack more than almost any other control. In many incidents, we only learn afterward that accounts had far more access than they needed.
Your Expel team can help with any of these. Reach out to your customer success manager to get started.
