Hackers hit EU Commission with AI agent tool, knocking HR systems offline for days
Unknown attackers have gone after the European Commission using ARTEX, a freely available AI tool built for security testing that has slipped out of its developer's control. HR and reimbursement systems were down for more than ten days, and it's still unclear whether staff data was stolen.

What happened
The European Commission has been hit by a major cyberattack. Brussels-based Politico broke the story on November 19, and Reuters, Numerama and The Next Web, among others, followed up. According to these reports, CERT-EU, the cybersecurity team for the EU institutions, found in early November that international attackers were using a tool called ARTEX against the Commission. ARTEX is a freely available, AI-powered program originally designed to find vulnerabilities in IT systems. As a precaution, the Commission shut down several internal systems, which disrupted its internal administration for more than ten days. Who is behind the attack, and exactly how ARTEX was used, is still unknown.
What ARTEX is and why it's dangerous
ARTEX started life as an open-source tool for penetration testing. In a pen test, security professionals attack a company's own systems on its behalf to find weaknesses before real criminals do. What's new about ARTEX is that it hands the work to AI agents: programs built on large language models that carry out multi-step tasks on their own. According to Numerama, the agents are given a target and a mission. A lead agent then breaks the job into subtasks, which other agents work through tirelessly. The FBI describes ARTEX as an "autonomous penetration testing framework based on large language models" that can automate multiple phases of an intrusion.
As misuse of the tool grew, its developer deleted the GitHub repository on October 8. That did little good. According to Numerama, copies and a commercial version had long since spread among cybercriminals. Numerama calls it a tool that has escaped its own creator.
Several agencies have issued warnings. On November 5, CERT-EU published a technical advisory saying attackers were increasingly using ARTEX to automate entire campaigns, and that because the tool is publicly available, attacks could become more frequent, faster and larger in scale. CERT-EU says it has seen similar AI-driven activity against other organizations it is responsible for. On November 13, the FBI followed with a FLASH alert, an urgent notice to companies and government agencies. It says hackers have been using ARTEX since at least August against financial services firms, government bodies, technology and defense companies, chipmakers, energy providers, healthcare, manufacturing, telecoms and law firms.
Payslips, leave, medical bills: what went down
Staff felt the impact most in their day-to-day work. Among the systems taken offline was Sysper, the Commission's HR platform, which handles payslips, leave requests and time tracking. MyPMO, the system for reimbursing medical expenses, was also offline. Both platforms were back up early this week, but Sysper briefly went down again on Wednesday, November 18. Translation services were also disrupted, a real problem for an institution with 24 official languages. One official told Politico he had "sworn quite a lot."
Salaries were still paid. But some employees couldn't log their hours or request leave. Three staff members complained to Politico about poor communication: the IT department only referred to a general problem with Sysper and never mentioned a cyberattack.
Unclear whether data was stolen
The Commission believes it has contained most of the attack, but it apparently still doesn't know exactly what happened. An EU official told Politico on condition of anonymity:
"The systems may be coming back online. The answers are taking a bit longer."
Investigators are still checking whether personnel data was taken and, if so, how much. According to the official, that uncertainty is exactly why staff have been told so little. The Next Web reports that a large amount of personal data could be affected. Commission spokesperson Thomas Regnier would only confirm that "certain systems were targeted," that the necessary measures had been taken, and that those affected had been informed in line with standard procedures. At the time of the reports, there was no detailed public statement.
Internally, the Commission is taking the incident very seriously. An internal email seen by Politico shows that the Corporate Management Board, the institution's top crisis body, met for an emergency session earlier this month. The Directorate-General for Digital Services introduced tighter security controls. Tech Commissioner Henna Virkkunen briefed the other commissioners on Wednesday at the College's weekly meeting.
Not a superweapon, but an assembly line
Earlier research shows how much AI is really involved in attacks like these. According to Numerama, Amazon lured ARTEX-driven agents into honeypots, decoy systems deliberately set up to attract and observe attackers, which let the company follow their operations in real time. Amazon found no new attack techniques. The agents exploited very basic weaknesses: an admin password that had never been changed, an installation that was never completed, or software missing current updates.
So the problem isn't that AI is inventing entirely new ways to break in. It simply works through well-known vulnerabilities faster, more persistently and at far greater scale than human attackers could. Dig.watch also stresses that it hasn't yet been publicly established what role ARTEX played in the attack on the Commission, which vulnerabilities were exploited, or how automated the attack was.
Not the first incident this year
According to The Next Web, this is already the Commission's second major cyber incident of 2026. In March, attackers used a stolen API key, a kind of digital access key for software, to get into the Amazon Web Services account behind the Europa.eu website. CERT-EU says around 92 GB of compressed data was stolen, including names, email addresses and email contents. Data from at least 29 other EU institutions and bodies may also have been affected. Over the winter, there was also an attack on the systems the Commission uses to manage staff work phones, in which employee names and phone numbers may have been leaked.
The string of incidents is politically awkward, because the Commission is currently trying to strengthen cybersecurity across the EU. In January, it presented a package of new protective measures. In July, Virkkunen said the Commission itself had become a target because of the geopolitical situation.
Analysis
This case illustrates a trend security experts have been warning about for a long time. AI agent tools change less about how attacks happen and more about how many happen. When a program hunts around the clock for forgotten default passwords and missing updates, even small oversights quickly become a way in. For defenders, that mainly means getting the basics right, consistently: install updates, change credentials, monitor your systems.
ARTEX also revives an old debate about dual-use software, meaning tools that can be used for both defense and attack. Classic pen-testing tools like Metasploit have long faced this problem. But AI agents lower the barrier to entry for attackers considerably. And the fact that deleting the repository couldn't stop the tool from spreading shows how hard it is to pull back code once it's been published.
Many questions remain open: who carried out the attack, how exactly ARTEX was used, and whether data was stolen. The Commission has said very little so far, and its internal communication is drawing criticism from staff. After its third major incident in a year, an institution that champions EU-wide cybersecurity is under pressure to be more transparent.
Sources
- Numerama (FR) numerama.com
- thenextweb.com thenextweb.com
- timesofindia.indiatimes.com timesofindia.indiatimes.com
- dig.watch dig.watch
Ask the AI assistant
Questions about this article, installation or background? Our assistant knows this page and runs on our own machine, not in someone else's cloud.
The Assistant is an AI (Gemma 4) and can make mistakes – please check important details against the source. Questions are used only to generate an answer and are deleted after one hour. Please don't enter any personal data.
Comments
What do you think? Already tried it? Write it here, no sign-up needed.
Comments are public. Links aren't allowed, and we delete abusive posts. You can delete your own comment at any time in this browser. Learn more in the Privacy Policy.