Skip to main content

European Commission Reportedly Hit by Attackers Using the ARTEX AI Framework, With HR Systems Offline for Days

The Commission confirms a detected cyber campaign. Politico's sources say the attackers used ARTEX, the AI hacking framework the FBI warned about on 13 November. What the FBI says it goes after, and what to check this week.

Spacious office floor with rows of empty cubicles and computer monitors.
KBY News Desk8 min read

In this article

Share

The European Commission has spent more than ten days recovering from a cyberattack on its internal systems, and Politico’s sources say the people behind it used ARTEX, a free AI-driven penetration testing framework that the FBI warned about last week. The Commission has confirmed a “detected cyber campaign” and that individual systems were targeted. It has not confirmed the tool, and it has not said whether anyone’s personal data was taken. The Register, which followed Politico’s report on Thursday 19 November, says its own questions to the Commission about that are unanswered.

Two staff applications went offline. Sysper, where Commission employees see payslips, submit timesheets and request leave, and MyPMO, which handles medical reimbursements, were taken down while officials investigated. The Register reports both came back earlier this week, although Sysper failed again on Wednesday, and that the Commission’s translation services were also affected. According to Politico, as relayed by The Register, salaries were still paid on time.

#What the Commission has said, and what it has not

The Commission’s spokesperson, Thomas Regnier, told Politico: “The European Commission has taken the necessary steps in response to a detected cyber campaign. Individual systems were targeted and, in line with our standard practice, we have informed those concerned.” That wording is careful. It confirms a campaign, confirms systems were targeted and says some people have been notified. It does not say how the attackers got in, what they reached or whether the two HR applications were taken offline as a precaution or because they were compromised.

The ARTEX detail comes from Politico’s sources, not from the Commission. The Register reports that the Commission’s defenders detected attackers using the framework, that the Commission’s Corporate Management Board met in an emergency session earlier this month, and that the commissioner for technology, Henna Virkkunen, briefed her colleagues on Wednesday. We have not been able to read Politico’s article ourselves, so everything about its reporting here is as The Register describes it. Nobody has said who is responsible, and we are not going to guess.

#Why ARTEX matters more than the name of one victim

The FBI published a FLASH alert on ARTEX on Friday 13 November. It describes an open-source, agent-based framework written in Go that turns a goal written in plain language into commands run by AI agents, with little human input after the objective is set. ARTEX does not carry its own model. It connects to outside providers, and the FBI lists Anthropic’s Claude and OpenAI’s GPT among them, along with DeepSeek, GLM, Grok and Kimi. Its analysis of attacker-run instances found Chinese models most common, with DeepSeek’s Flash family dominant.

The FBI says criminals have used it since at least August, against organisations in financial services, government, technology, defence, energy, healthcare, manufacturing, telecoms and legal work. Some intrusions went from first targeting to privileged access in under three hours. It names two incidents: attacks on South Korean financial institutions that exposed personal information on about 67,000 people, and a supply chain compromise involving ConnectWise Cloud Backup that led to attempted sign-ins to more than 2,250 downstream tenants.

The original developer took the GitHub repository down on 8 October after reports of abuse. The FBI’s point is that this changed nothing in practice: copies live on as community forks, archives and a commercial edition called ARTEX PRO. As of early November, researchers had found more than 1,600 ARTEX consoles exposed to the internet, and the FBI says the true number is likely much higher. CERT-EU, the response team for EU institutions, had already warned on 5 November that it had seen what it believed were ARTEX-assisted attacks against bodies it looks after, according to The Register.

The Register also quoted Snehal Antani, chief executive of Horizon3.ai, who said what once took a skilled operator days or weeks is now happening in hours. His firm sells security testing, so treat the framing as an industry view. The FBI’s own numbers say much the same thing.

#Where the framework goes looking

The most useful part of the FBI document for an IT team is its description of where ARTEX finds a way in. It says the framework is good at the peripheral systems that get less attention than flagship applications: partner and supplier portals, mobile backends, employee applications, sales-support systems, legacy services, third-party integrations and applications built by individual business units. A staff self-service tool for payslips and leave is exactly that kind of system. We think that is the main reason this story is worth reading even if you are nowhere near Brussels. Nobody has shown that Sysper was breached through a particular flaw, and we are not claiming it was. The point is that the category of system the Commission lost for days is the category the FBI says these agents are best at finding.

The FBI’s list of weaknesses its agents have exploited is blunt. It includes unsigned SAML assertions, anonymous file upload, server-side request forgery, hardcoded administrator passwords sitting in JavaScript bundles, web APIs with no authentication, direct object reference flaws, SQL injection, weak administrator credentials, and stolen VPN configurations. None of this is new. What changes is that one person can point a fleet of agents at hundreds of targets and have them try every item on the list in parallel, without tiring and without needing to know what a SAML assertion is.

Once inside, the FBI describes agents that read LAPS passwords, enumerated more than 110,000 Active Directory objects, harvested service tickets through Kerberoasting, stole the domain’s krbtgt key with DCSync, and forged an administrator token using authentication keys that had never been rotated. One agent inventoried which Windows machines had no CrowdStrike Falcon sensor, which is a neat way to find where nobody is watching. The same document records lateral movement attempts towards domain controllers, pass-the-hash among them, that endpoint protection blocked. That is worth remembering: the tooling is fast, not magic.

#Checks to run this week

We would start from the FBI’s own immediate actions and put the Commission’s experience on top of them. This is our reading, not an instruction from the Commission or the FBI.

  1. List everything with a login page that is reachable from the internet, including the staff and supplier tools nobody owns. The FBI asks for a full inventory of internet-facing applications, peripheral ones included. If you cannot produce that list in an afternoon, that is the finding.
  2. Search front-end code, build output and configuration for embedded secrets. A hardcoded administrator password in a script bundle is on the FBI’s list of things agents found. Rotate anything you find.
  3. Check that SAML and token validation reject unsigned or altered assertions on every application that accepts single sign-on.
  4. Rate-limit and alert on employee and partner portals. The FBI suggests alerting on sequential lookups and on unusually high numbers of records read in one session, because agents enumerate quickly and noisily.
  5. Compare your endpoint sensor list with your device list. A machine without an agent is somewhere an automated intruder can work unobserved.
  6. Look at who can read LAPS passwords, which service accounts can be Kerberoasted, and when your domain’s authentication keys were last rotated.
  7. Watch outbound traffic for connections to AI provider APIs from servers that have no reason to make them. The FBI lists calls to api.anthropic.com, api.deepseek.com, open.bigmodel.cn and api.x.ai as something to investigate alongside other suspicious activity, and says to vet any indicator before blocking.

Our own earlier story on Amazon’s report on AI agents hunting exposed AI tools for API keys covers the other side of this: the same agents use the keys they steal to pay for the next round of attacks.

#Our view

The easy reading is that this is a government embarrassed by a new tool. We think the more useful reading is dull. The Commission detected the campaign, convened its senior board and took systems offline, and The Register still counts more than ten days of disruption. That is what a competent response to a fast attack looks like when the defender’s systems are interconnected and the attacker never sleeps. The lesson for everyone else is that detection is no longer the scarce thing. Being able to cut off a peripheral application quickly, and having a plan for the people who rely on it, is.

We expect more of these. A free tool with over 1,600 exposed consoles and several commercial and community variants will not go away because its original author deleted a repository. What would prove us wrong is a clear pattern of intrusions that ARTEX did not shorten, for example if the Commission’s investigation finds the attackers took days, not hours, to reach anything of value.

#If you run a small business

A firm with ten or two hundred staff is not a target in the way the Commission is, but the tool is cheap and aimed at whatever it can find, so the risk is any login page reachable from the internet: a customer portal, a booking system, a supplier upload page, a forgotten test site. Ask whoever looks after your IT, or your provider, one question in writing: which of our systems can be reached from the internet, and when was each one last checked? For the wider basics, our explanation of Cyber Essentials covers the controls a small organisation can put in place without a security team.

#What we do not know

Whether any personal data was taken. Whether ARTEX is the tool the Commission’s own analysts name, or only what Politico’s sources say. How the attackers got in. Who they are. Whether the repeat failure of Sysper on Wednesday was related. We will update this story if the Commission, CERT-EU or the FBI adds detail.

Evidence trail

Sources and verification

Primary documentation and external technical references used in this article.

  1. 01The Register, 19 November 2026: European Commission scrambles to contain AI-assisted cyberattacktheregister.com
  2. 02FBI FLASH-20261113-001, 13 November 2026: Threat Overview of ARTEX Autonomous Penetration Testing Framework (PDF)ic3.gov
  3. 03Amazon Threat Intelligence: threat actors running cybercrime campaigns with ARTEX AI agentsaws.amazon.com

About this page

Credited to

Dates

Published

Corrections

No corrections are listed for this page. Some review sweeps corrected pages without naming each one; those are in the corrections log.

Reader Interaction

Comments

Add a thoughtful note on European Commission Reportedly Hit by Attackers Using the ARTEX AI Framework, With HR Systems Offline for Days. Comments are checked for spam and held for moderation before appearing.

Loading comments...

More news

All news

Was this useful?

Reports go to the site owner and are used to correct pages. Corrections are listed at /corrections.

Get new fixes by email.

New runbooks, configuration traps and fixes from KBY Technologies, about twice a week.