
European Commission fights an AI-assisted cyberattack as ARTEX spreads: what is confirmed
By Yash Malviya·
Hackers reportedly used the ARTEX AI pentest framework against Commission systems that handle payslips and medical claims. The attack is confirmed, the tool is reported, and whether staff data was stolen is not established.
The European Commission has spent more than 10 days dealing with a cyberattack on internal systems that run staff payslips, timesheets, leave and medical reimbursements, and people familiar with the response told Politico that the intruders used an AI-powered tool called ARTEX. The Register summarized the reporting on November 19, 2026. The facts are thinner than the headline suggests, so here is what is confirmed, what is reported, and what nobody has said.
What is confirmed
The Commission confirms that it detected a cyber campaign. Spokesperson Thomas Regnier told Politico that the Commission "has taken the necessary steps in response to a detected cyber campaign," that individual systems were targeted, and that the Commission informed those concerned, in line with standard practice.
Two systems were taken offline. Sysper, which staff use for payslips, timesheets and vacation requests, was restored earlier in the week and went down again on Wednesday. MyPMO, which handles medical reimbursements, was also taken offline and restored. Translation services were affected as well. Per Politico, some staff could not book time off or submit timesheets, but salaries were paid on time.
What is reported but not confirmed
The ARTEX connection comes from Politico's sources, not from a Commission statement. They say defenders detected the attackers using ARTEX, a publicly available AI-powered penetration testing framework. An emergency meeting of the Commission's Corporate Management Board was held earlier in November, and tech commissioner Henna Virkkunen briefed fellow commissioners on Wednesday.
The Register says the Commission has not answered its questions about whether employee data was stolen or whether systems were shut down only as a precaution. Whether payslip data was accessed "has not been established," in the outlet's words. Until that changes, there is no public evidence of a data breach, only of an outage.
“threat actors have weaponized to automate multiple phases of cyber intrusions at machine speed, from reconnaissance through data exfiltration”
What ARTEX is
The best primary description is the FBI's. In a FLASH alert dated November 13, 2026 (FLASH-20261113-001), the bureau calls ARTEX "an open-source, large language model (LLM)-driven agentic platform" that threat actors have "weaponized to automate multiple phases of cyber intrusions at machine speed, from reconnaissance through data exfiltration." It is written in Go, translates goals stated in natural language into commands run by specialized agents, and contains no model of its own. It plugs into outside providers, including frontier models from Anthropic and OpenAI and open or Chinese-origin models such as DeepSeek, GLM, Grok and Kimi. The FBI says Chinese models are the most common in attacker-run instances, with the DeepSeek Flash family dominating, because of "their balance of capabilities, relatively low cost, and lack of observable guardrails when framed for penetration testing."
One detail will catch readers of this site: the FBI document calls the agents "super intelligence (SI) agents," its shorthand under the administration's rename, even though they are ordinary LLM agents. It is a reminder that the label now covers everyday tooling.
The numbers are specific. The FBI says some intrusions progressed from initial targeting to privileged access in under three hours. It lists a campaign against South Korean financial institutions that led to the exfiltration of personal information of about 67,000 people, and a supply chain compromise affecting ConnectWise Cloud Backup customers with attempted sign-ins to more than 2,250 downstream tenants. As of early November, researchers had found more than 1,600 publicly exposed ARTEX consoles, and the FBI says total deployments are probably far higher.
A tool that could not be recalled
ARTEX appeared on GitHub in July 2026 from a developer using the handle Autumn-27. The developer pulled the repository on October 8 after reports of abuse. By then the code had spread. CERT-EU, which protects EU institutions, warned on November 5 that it had seen ARTEX-assisted attacks against organizations it oversees. The Cloud Security Alliance's November 16 research note counts more than 12,000 GitHub fork events, according to Wiz data, and several rebrands and a commercial edition. Withdrawal, in other words, removed the official download and nothing else. Our tracker of the ARTEX campaigns, forks and exposed consoles keeps the dated list.
What experts say
Snehal Antani, CEO of Horizon3.ai, told The Register that such frameworks let a single actor "stand up fleets of agents" that plan, probe, exploit and move laterally with little human input, and warned that state-backed hackers and ordinary criminals now use the same tools, which could make attribution harder. He is a vendor in the security testing market, so weigh the framing accordingly.
The FBI's own list of what ARTEX is good at is more useful to defenders: peripheral systems that get less attention than flagship applications, such as partner portals, employee applications, legacy services and third-party integrations. An HR and payroll system fits that description.
What it means for the EU's own AI debate
Germany's Marco-Alexander Breit argued this week that Europe must adjust its AI rules to stay competitive, according to Semafor's November 17 report. The Commission episode is an awkward companion to that argument. The institution drafting the rules for frontier AI has just spent more than 10 days with HR systems disrupted by an AI-assisted tool that was free to download. Whether that changes the political mood is speculation, and we do not assert it. It does show that the cyber risk from open agentic tools is a current operating problem for governments, not a hypothetical on a risk register. For how labs are gating their own cyber-capable models, see our tracker of restricted-access cyber AI models.
Our take
This episode shows a structural shift, not a single breach: intrusion skills that once needed a team can now be rented from a repository. For Brussels the open question is narrow and answerable, namely whether any personal data left the building. For everyone else, the FBI's mitigation advice is dull and right: remove default and hardcoded credentials, close unauthenticated admin interfaces, and patch internet-facing software, because an agent that tries thousands of cheap probes will find the one you forgot. We would wait for the Commission's answer on data theft and for CERT-EU's technical write-up before drawing any conclusion about attribution.
Frequently asked questions
Did hackers use AI to attack the European Commission?
Reportedly yes. The Commission confirms a detected cyber campaign. The AI link comes from Politico's sources, relayed by The Register, who say defenders detected attackers using ARTEX, an AI-powered penetration testing framework. The Commission has not confirmed the tool.
Was European Commission staff data stolen?
That has not been established. The Register reported the Commission had not answered whether employee data was stolen or whether systems were shut down only as a precaution. Salaries were paid on time, per Politico.
What is ARTEX?
An open-source, LLM-driven penetration testing framework written in Go. The FBI says threat actors have used it to automate intrusions from reconnaissance through data exfiltration. It has no model of its own and connects to outside AI providers.
How fast can ARTEX break into a network?
The FBI's November 13, 2026 FLASH says some intrusions went from initial targeting to privileged access in under three hours. The Cloud Security Alliance's note, citing Google's threat intelligence group, gives an average of about four hours and one case of 15 minutes.
Why can't ARTEX just be taken down?
Its developer removed the GitHub repository on October 8, 2026, but forks, archived copies, localized versions and a commercial edition called ARTEX PRO remained. The CSA counts more than 12,000 fork events recorded by Wiz.
What should organizations do about ARTEX-style attacks?
The FBI advises organizations with internet-exposed apps, portals and cloud infrastructure to follow its mitigations. The CSA's note stresses removing default credentials, rotating exposed secrets, closing unauthenticated admin interfaces and patching internet-facing software.
Sources
What each one is, and whose it is.
- 1
European Commission scrambles to contain AI-assisted cyberattack, The Register (November 19, 2026)
Press reportIndependent of the vendor - 2
Threat Overview of ARTEX Autonomous Penetration Testing Framework (FLASH-20261113-001), FBI Internet Crime Complaint Center (November 13, 2026)
OtherIndependent of the vendor - 3
ARTEX and Its Forks: Agentic Pentest Frameworks as Attack Infrastructure, Cloud Security Alliance AI Safety Initiative (November 16, 2026)
OtherIndependent of the vendor