The European Commission has confirmed a European Commission cyberattack in which the attackers used a publicly available AI tool. Speaking at the midday briefing on November 19, 2026, Commission spokesperson Thomas Regnier called it an “AI generated cyber campaign”. The incident was detected on October 22 and confirmed on October 26. Several applications were taken offline as a precaution, and most are running again. Whether personal data was stolen remains unclear. For operators of web applications, the most interesting part is which weaknesses ARTEX, the framework reportedly used, goes after.
European Commission cyberattack: what is officially confirmed
The incident became public through a Politico report on the morning of November 19. A few hours later, Regnier took questions on it in the Berlaymont press room; the Commission’s audiovisual service provides the video and transcript of the briefing. The key statements:
- The incident was detected on October 22 and confirmed after an initial investigation on October 26.
- As a protective measure, several applications were temporarily taken down; most have since been restored.
- The Commission is in close contact with CERT-EU and the European Data Protection Supervisor. The investigation is ongoing.
- So far there is no evidence of a data breach affecting Commission staff “as a whole”.
- No ransom demand is known.
- It was a campaign, meaning a series of incidents, carried out with a publicly available AI tool.
Regnier declined to comment on attribution, the attackers’ goals, the scale of the attack or the Commission’s own defensive tools. Asked whether the attack was still ongoing, he said many elements had been contained but the investigation continues. He rejected the suggestion that the Commission had kept the incident quiet, saying it had followed the usual procedures and informed affected services and the data protection supervisor.
What Politico adds
The original report in Politico’s Brussels Playbook goes further than the official statements. According to Politico, CERT-EU found earlier this month that international hackers had used ARTEX, a publicly available AI-powered tool for finding vulnerabilities. The disruption lasted more than ten days, Politico says. It hit Sysper, the HR platform for payslips, leave requests and timesheets, and MyPMO, the medical reimbursement system. Both were restored earlier this week, although Sysper briefly crashed again on November 18. Translation services were also affected.
Politico also reports that the Corporate Management Board, the Commission’s highest internal crisis body, held an urgent meeting on the “ongoing cyber incidents” earlier in November, and that tech commissioner Henna Virkkunen briefed the College on November 18. Salaries were largely paid as usual. These details come from anonymous sources and internal emails seen by Politico; the Commission has not confirmed them individually. Regnier also did not name ARTEX himself during the briefing.
ARTEX: a pentest framework turned attack tool
ARTEX is no stranger to our readers: we reported on the use of ARTEX by the VAPOR PANDA group in October, and the framework came up again in our report on exploit_agent. The most detailed official description so far comes from the FBI. In a FLASH report dated November 13 (FLASH-20261113-001), it describes ARTEX as an open-source, LLM-driven agentic platform that automates intrusions from reconnaissance to data exfiltration.
The key points of the FBI report:
- ARTEX is written in Go and turns goals described in natural language into concrete commands executed by specialised agents.
- It ships without a model of its own and integrates external providers. In attacker-controlled instances, Chinese models dominate according to the FBI, above all the DeepSeek Flash family.
- Since at least August 2026, various actors have used the framework against organisations in finance, government, technology, defence, energy, healthcare and other sectors.
- Some intrusions went from initial targeting to privileged access in under three hours.
- The original developer published ARTEX in July and removed the repository on October 8 after reports of abuse. Forks, copies and a commercial version called “ARTEX PRO” remain in circulation; in early November researchers found more than 1,600 publicly exposed ARTEX consoles.
CERT-EU had already warned about ARTEX in a blog post on November 5, reporting similar AI-driven activity against its own constituency that “highly likely” involved ARTEX. One important detail from that post: the framework relies on configurable rules and on the model itself to stay within scope – there is no hard technical enforcement.
The weaknesses are classic ones
For web developers, the list of observed techniques in the FBI report is more instructive than any headline about “AI attacks”. ARTEX does not invent new attack types; it works through well-known bug classes very quickly and tirelessly. For initial access, the FBI lists, among others:
- unsigned SAML assertions and forged SAML tokens,
- XXE, SSRF, SQL injection and server-side template injection,
- hardcoded administrator passwords in single-page app JavaScript bundles,
- web APIs without authentication and insecure direct object references (IDOR),
- anonymous file uploads, weak admin credentials and stolen VPN configurations,
- cross-account cloud trust relationships and compromised supplier credentials.
After getting in, attackers deployed JSP and PHP web shells, in some cases via MySQL log poisoning, according to the FBI. The report says peripheral systems that get less attention than the flagship application are particularly exposed: partner and supplier portals, mobile backends, employee apps, legacy services and applications built by individual business units. Internal HR and reimbursement systems like the ones that went down at the Commission fall squarely into that category – although it is not known whether ARTEX got in through any of these flaws there.
What operators should do now
The FBI and CERT-EU give largely matching recommendations:
- Inventory every internet-facing application, explicitly including peripheral systems and departmental tools.
- Keep admin panels, deployment and publishing interfaces, backup platforms and hypervisors off the open internet.
- Prioritise patching of internet-exposed systems and enforce multi-factor authentication on all critical services.
- Sanitise error responses: no stack traces, SQL errors, internal paths or framework details for unauthenticated users.
- Continuously scan repositories, container images, pipelines and configuration for secrets, and rotate exposed credentials immediately.
- Add rate limits and alerts to portals, for example for unusually many sequential record lookups per session.
- Fix findings rated “low” as well, because AI tooling can chain minor misconfigurations into effective attack paths.
Conclusion
According to the AP news agency, this is the first time the EU has publicly reported a cyberattack carried out with an AI tool. Much about the European Commission cyberattack remains open: who is behind it, which data was affected and how the attackers got in. What is clear is what tools like ARTEX change: not the type of vulnerabilities, but the speed at which they are found and chained together. Organisations that know their peripheral systems, lock down admin access and take minor findings seriously remove much of the attack surface such campaigns depend on.




