News

European Commission Confirms AI-Assisted Cyberattack: HR and Medical Systems Down for Over Ten Days

The European Commission confirmed on November 19, 2026 that its internal systems were breached in a cyberattack carried out with the help of artificial intelligence tools. The first signs appeared on October 22 and the incident was classed as an attack on October 26. The HR platform Sysper, the medical reimbursement system MyPMO and translation services were down for more than ten days, while salaries were paid on schedule. According to Politico, the EU's CERT-EU identified the open-source AI penetration-testing framework ARTEX; the Commission has not named the tool and has not confirmed the attacker's identity or whether data was copied.

One-line takeaway

The attack ran for ten days and took out the two systems that least can afford to stop: payroll and reimbursement. What is confirmed is that AI tools were used. What is not confirmed is who, how they got in, and what they took.

Background ​

On November 19, 2026, European Commission spokesperson Thomas Regnier confirmed at a briefing that the institution's internal systems had been hit by a cyberattack conducted with the help of artificial intelligence tools. The first signs appeared on October 22; the incident was formally classified as an attack on October 26.

Regnier's language was measured. The extent of the damage is hard to assess, but there is no evidence of a large-scale leak of staff personal data, and no ransom demand has been received. The European Data Protection Supervisor has been notified, affected staff have been informed, and the investigation continues.

Timeline and affected systems ​

DateEvent
October 22First anomaly detected
October 26Classified as a cyberattack
November 5CERT-EU warns that attackers are using the tool
November 13An FBI notice, as reported, says the tool has been in use since at least August
November 18Sysper comes back, then fails again
November 19Commission goes public; Sysper restored

Three systems were affected:

  • Sysper: the Commission's HR platform, used by staff to check payslips, file leave requests and log working hours;
  • MyPMO: the medical expense reimbursement system;
  • Translation services: briefly interrupted.

The outage lasted more than ten days. According to press reports, staff were temporarily unable to enter payroll and leave data, and senior officials met to discuss the situation. Salaries themselves were paid on time.

Known and unknown ​

ItemStatus
Use of AI toolsConfirmed; the spokesperson called it "an AI generated cyber campaign"
Specific toolNot named by the Commission; Politico reports ARTEX
DurationMore than ten days
Data breachNo evidence of large-scale staff data theft; investigation open
Ransom demandThe Commission says it is not aware of one
Attacker identityUnconfirmed
Entry vectorNot disclosed
Whether data was copiedUnconfirmed

The tool and the warnings ​

According to Politico, CERT-EU, the EU institutions' cyber emergency response team, identified ARTEX in the attack: an open-source penetration-testing framework that is publicly available and uses AI to identify vulnerabilities in IT systems. The Commission itself has not named the tool, so this detail belongs to press reporting.

The order of the warnings deserves attention. CERT-EU issued an alert on November 5, and an FBI notice dated November 13 said the tool had been in use since at least August. In other words, attackers' use of such tools was not a sudden development, and the warning preceded the Commission's public confirmation by nearly two weeks.

Security outlets differ on how to characterise ARTEX. This article makes no judgement about the tool's attribution or country of origin and records only three verifiable traits: open source, publicly available, AI-driven.

The third disclosure this year ​

According to a tally by The Record, this is the third cyber incident the Commission has disclosed this year, after January and March, with no official link drawn between the three.

For an institution administering the business of 27 member states, three disclosures is itself a signal: a rising disclosure rate may indicate better detection, or a widening attack surface. Both can be true.

What the incident shows ​

First, AI's role on the attack side has become concrete. It is not "AI writes a virus automatically" but AI accelerating vulnerability identification and path planning, compressing a reconnaissance phase that once took weeks of manual work. That maps precisely onto the hardest part for defenders: broad asset surfaces, complex configurations and lagging vulnerability discovery.

Second, availability risk outweighs confidentiality risk in critical systems. What went down were process systems for payroll and medical claims; the short-term effect is administrative standstill, and the long-term effect depends on whether data was taken. The Commission says there is no evidence of a large-scale leak, but the investigation is open, so that is not yet a conclusion.

Third, a gap remains between warning and response. CERT-EU alerted on November 5; the Commission went public on November 19, while systems kept failing intermittently. That is not a criticism but a record of a fact: institutional response takes time, and attackers move faster.

Three defensible inferences ​

The following three points are drawn from public facts, not from official conclusions.

First, availability outranks secrecy. The systems that failed handle payroll queries, leave registration and medical reimbursement, processes that cannot simply stop. For an institution, restoring service and establishing whether data leaked are parallel tracks, and public communication usually has to address the former first.

Second, a window exists between warning and disclosure. CERT-EU warned on November 5 and the Commission confirmed on November 19. During that window the work is triage, isolation and assessment rather than immediate disclosure, an order most regulatory frameworks permit. The corollary is that external observers only ever see the post-hoc version.

Third, the tooling barrier is falling. ARTEX is described as publicly available, AI-driven and open source. Penetration-testing tools are legitimate products; the issue is that the same capability serves both self-assessment and offence. Defence and offence got cheaper at the same time, which is what distinguishes this round of AI security debate from earlier ones.

Why "AI generated" matters as a phrase ​

The spokesperson's wording - "an AI generated cyber campaign" - shifts the discussion from who did it to how it was done.

Broken down, an attack of this kind has roughly three stages: AI scanning assets and configurations, AI matching known vulnerabilities against exploit conditions, and AI generating the action sequence. None requires deep manual penetration expertise; the tooling has packaged that part.

Defenders' countermeasures are not new: asset inventories, patch timeliness, least privilege, segmentation. What changed is tempo. When reconnaissance shrinks from weeks to days, the window left to defenders shrinks proportionally. That is the second piece of "AI acceleration" news this week, alongside the data centre power figures.

Boundaries ​

  • The confirmation date, spokesperson remarks, affected systems, salary payment and investigation status come from the Commission's public statements and reporting by Politico, Euronews, The Record and TechRadar;
  • ARTEX is a Politico attribution; the Commission has not formally named the tool, and this article does not judge its origin or country;
  • "The attack lasted more than ten days" and "Sysper failed again on November 18 and recovered on November 19" are press accounts quoting EU officials;
  • "Third this year" comes from The Record's tally; officials have not linked the three incidents;
  • Whether data leaked, the attacker's identity and the entry path are all unconfirmed, and this article does not speculate;
  • This article does not assign responsibility to any country, institution or individual.

Sources ​