Blog

Cybersecurity & AI

ARTEX: When AI Gives Every Hacker a Tireless Team

Digitrans · · 9 min read

An illustrated office and workshop connected to cloud servers, with attack and defence paths.

AI-generated editorial illustration.

In this article

The European Commission has been hit by an AI-assisted cyberattack. Banks in South Korea have lost customer data. And a compromised software supplier became a route into organisations using its services. The details differ, but one name keeps appearing: ARTEX, a tool that uses AI agents to carry out security testing and has also been adopted by attackers.

For a small business in Luxembourg, this is not just another story about a distant institution. An old remote-access portal, a forgotten server or a supplier with broad access can become an entry point. What changes with AI is how quickly and cheaply attackers can look for those openings, investigate what they find and move on to the next step.

The encouraging part is that defenders can use AI too. The practical question is how to put it to work before someone else finds the weak point.

Three incidents, and what we actually know

In November 2026, the European Commission confirmed that attackers had used AI tools in a cyber campaign detected in October. It said several applications were temporarily taken offline as a precaution and that most had since been restored. At the time of its statement, the Commission said it had not detected evidence of a breach affecting staff as a whole. Associated Press reported the confirmation and the limits of what officials had disclosed.

Reports citing an EU official link the campaign to ARTEX, although the Commission did not publicly name the tool. Separately, CERT-EU warned about ARTEX and described its use by threat actors. It is important to keep that distinction: the AI-assisted attack is confirmed, while the precise tool attribution rests on reporting from officials familiar with the investigation.

In South Korea, the evidence points more directly to ARTEX. CrowdStrike investigated attacker-controlled infrastructure associated with a campaign against financial institutions and found ARTEX configuration files and AI session records. The FBI's November alert associates the South Korean campaign with the theft of personal information relating to approximately 67,000 people. CrowdStrike described, among other targets, a loan-status service used by brokers and an employee mobile-work system: ordinary business services that had become valuable openings.

The third case should interest any company that relies on managed IT. Microsoft reported that a China-aligned group used ARTEX during a campaign involving ConnectWise Cloud Backup. After gaining access to the supplier's environment, the attackers attempted to sign in to more than 2,250 downstream customer environments. That number describes attempted access, not 2,250 confirmed compromises. Microsoft says most targeting did not succeed, but the attackers obtained privileged cloud access at a number of organisations, in some cases within three hours of initial targeting.

Together, these incidents show three different problems: disruption to important services, exposure of personal data, and the risk that a trusted supplier's access can become an attack route. None requires an SME to be personally famous or strategically important.

What is ARTEX, and how does it work?

ARTEX is an AI-assisted penetration-testing framework. Penetration testing is a legitimate way to look for security weaknesses with the owner's permission. The same kind of software can also be misused against systems whose owners never agreed to a test.

Think of ARTEX as a coordinator for security tools, with an AI model helping decide what to do next. An operator supplies an objective and a target. The framework can discover exposed services, examine the results, try further checks and combine what it learns into a picture of possible routes through a system. Depending on its configuration and access, it can carry an operation further. The FBI describes activity ranging from reconnaissance to data theft.

The AI model is not a magic key. An exposed service still needs a weakness, stolen credential, excessive permission or another viable route. Nor does every attack run without human help. What makes an agentic framework consequential is the combination: it can connect existing tools, interpret their output, adapt its next step and repeat that work across many targets at a pace a small team may struggle to match.

Closing the original public repository did not make the problem disappear. The FBI says copies, variants and a commercial version remain available. ARTEX is one example of a wider shift in how offensive security work can be automated.

Why smaller businesses are exposed

Large organisations can have the money and specialists to monitor their systems around the clock. Many SMEs cannot. The same person may manage laptops, Microsoft 365, the website, a line-of-business application and the relationship with an outside IT provider, while also doing an entirely different day job.

Over time, a company's digital estate grows in ways that are easy to miss: a test website left online after a project, an old VPN, a remote maintenance account, a cloud application granted broad permissions, or a server that nobody wants to update because it runs an important application. Each may have a perfectly ordinary explanation. Together they create a larger surface to defend than the business realises.

AI does not invent all of these weaknesses. It helps an attacker search for them at scale and investigate promising leads faster. That makes the economic calculation more worrying: even a small target can be worth pursuing when much of the initial work can be automated.

There is also the supplier question. Your cloud provider, accountant, website agency or IT support company may have legitimate access to your systems. Ask what access they need, how it is protected, who reviews it and what happens if the supplier suffers a breach. The ConnectWise case is a reminder to investigate those relationships, not a reason to assume every supplier is unsafe.

Start with the doors you can actually close

You do not need an AI security platform to make the first improvements. Start with a short inventory and turn it into an action list that has an owner and a deadline.

  • Find what is exposed. List public websites, remote-access services, cloud accounts and servers, including old or outsourced systems. Check with your IT provider what is actually reachable from the internet.
  • Protect the keys. Require multifactor authentication, particularly for email, administrators and remote access. Remove dormant accounts, review supplier access and reduce overly broad application permissions.
  • Fix the highest-risk openings first. Prioritise actively exploited flaws, internet-facing systems and critical accounts. Where an immediate update is not possible, restrict access or take the service offline until a safe fix can be applied.
  • Make recovery possible. Keep protected backups, including copies an attacker cannot casually change, and test that you can restore a useful system and its data. A backup report is not the same as a successful restore.
  • Decide who acts when an alert arrives. Agree who receives it, who can disable access and how to reach your provider outside office hours. Keep the contact details somewhere available even if email is unavailable.

For a Luxembourg SME, national support is also available: CIRCL provides incident response and practical security information, and Cybersecurity Luxembourg brings together local resources. If you suspect an active incident, contact your IT provider or incident-response team promptly, preserve relevant evidence and avoid making uncontrolled changes that could destroy it.

Where AI helps defenders

The defensive challenge is not merely to produce more alerts. It is to work out which ones matter, understand how a weakness could affect the business and get an effective fix in place before an attacker acts.

AI can help a security team bring together an inventory, configuration and vulnerability findings; investigate suspicious activity; examine code or a proposed change; and prepare a prioritised explanation of what needs attention. For custom software, it can assist with developing a fix and tests. For a commercial product, the right action may be to apply the vendor's update or use a temporary mitigation, rather than asking an AI model to rewrite the product.

A suggested fix is not a deployed, verified fix. Findings need to be checked against the real environment. Patches need compatibility and security testing, an appropriate change window and a way to roll back. Afterward, a retest should confirm that the exposure is actually closed. Depending on the agreed workflow, some safe actions can be automated; changes with significant operational consequences need the right controls and oversight.

How Digitrans uses AI for security testing

At Digitrans, we use OpenAI's Daybreak for parts of our authorised security work, alongside our own security tools and local AI models. OpenAI describes Daybreak as a set of capabilities for approved defenders, including investigating and validating vulnerabilities and developing and testing fixes. In practice, that can help us move from a suspicious finding to a clearer assessment and a proposed remediation.

We also develop and use offensive testing tools in-house, including workflows that connect security tools to local models whose refusal behaviour has been reduced through abliteration. This matters because a defensive assessment sometimes requires investigating the same techniques an attacker might use. Our article on abliterated models and security testing explains the distinction and the approach in more detail.

The two approaches are complementary. Daybreak supports defensive analysis and remediation; our own tooling and local models help us examine offensive possibilities in an agreed test. Neither makes every finding accurate, eliminates the need for security expertise or promises automatic protection against ARTEX.

For a business engagement, we first agree on the systems and permissions in scope, the permitted testing, sensitive data handling and how findings will be reported. We then identify and validate relevant weaknesses, explain the business risk, help plan or implement agreed fixes, and retest. If a supplier operates a system, their involvement or authorisation may also be needed. The goal is not a frightening report full of theoretical issues. It is a shorter list of real problems, with a practical route to closing them.

The question to ask on Monday morning

If someone started probing your business today, would you know which systems they could reach? Would you know who has privileged access? And if an alert arrived tonight, would anyone know what to do with it?

ARTEX has made those questions more urgent. It has not made the answers impossibly complicated. Start by finding the exposed doors, securing the keys and making sure someone can act. Then use AI-assisted testing to find the gaps that an inventory alone will miss.

Talk to Digitrans about a cybersecurity assessment.

Frequently asked questions

Is ARTEX malware?

ARTEX is a penetration-testing framework that coordinates security tools using AI. It can be used legitimately with permission, but attackers have also used it in unauthorised intrusions.

Was ARTEX used against the European Commission?

The Commission confirmed an AI-assisted attack. Media reporting citing an EU official connected the campaign to ARTEX, but the Commission did not publicly name the tool in its statement.

Can a small company be targeted by AI-assisted attacks?

Yes. Automation can make it economical to look for exposed systems and weaknesses across many organisations. A small company can also be affected through a compromised supplier.

Can AI automatically fix every vulnerability?

No. AI can help investigate and develop fixes, but its output needs validation. The right response may be a vendor patch, a configuration change, restricting access or replacing an unsupported system, followed by a retest.

Does Digitrans provide cybersecurity services in Luxembourg?

Yes. Digitrans offers security assessments and related assistance to businesses in Luxembourg, using AI-supported analysis and authorised security testing as appropriate to the agreed scope.