The European Commission runs large IT systems and has CERT-EU, the EU institutions' own cyber response team, to call on. Even so, an attack in which the hackers used AI took parts of its staff systems offline for more than ten days. If you run a company of a few hundred people, that is worth a closer look, not because you are a likely target for the same group, but because the tools involved are available to anyone.

What happened

At the Commission's midday press briefing on 19 November 2026, spokesperson Thomas Regnier confirmed that attackers had used AI. He called it "an AI generated cyber campaign" and said the first signs were detected on 22 October.

Reports from The Register, RTÉ, Help Net Security and SecurityWeek between 19 and 20 November agree on the following:

  • Systems affected: Sysper, the Commission's HR platform for payslips and leave, and MyPMO, its medical reimbursement system, were disrupted, and so were translation services. The Commission took some services offline to contain the attack.
  • Duration: the disruption to staff systems lasted more than ten days, according to Politico's reporting. Sysper has since been restored, though reports differ on whether every staff service is fully back.
  • Data: Politico reported that the attack may have exposed large amounts of personal data. Regnier said investigators had found no evidence of a data breach affecting Commission staff as a whole. That question is still open.
  • Tool and attackers: Politico reported that CERT-EU found the attackers had used ARTEX, a publicly available AI-powered penetration testing framework. The Commission did not name the tool. Regnier did not attribute the attack to any known group and said he was not aware of a ransom demand.
  • Notification: Regnier said the European Data Protection Supervisor had been informed.

The investigation is still running, and some details, including the exact timeline and what data was reached, may change.

What it means for leadership teams at mid-size companies

Three points apply well beyond Brussels.

Attack skill is getting cheaper. Penetration testing frameworks exist so security teams can find weaknesses before attackers do. When one with AI built in is publicly available, the same speed is available to people with fewer skills and worse intentions. A mid-size company that assumed it was too small to attract a skilled attacker should now assume the skill can be rented or downloaded.

The window between a weakness and an attack is shrinking. If finding and testing a weakness takes hours instead of weeks, a patch you planned for next month's maintenance window may come too late. The question for your team is not whether they patch, but how fast for systems that face the internet.

Staff systems are targets too. The systems hit here held payslips, leave records and medical claims. Many companies protect customer data carefully and treat HR, payroll and benefits tools as back-office. Those tools hold some of the most sensitive data you have, and when they go down, every employee notices.

There is a fourth, quieter point. Weeks after the first signs, whether staff data was reached is still not settled in public. Uncertainty is normal early in an investigation, but it damages trust if nobody has planned what to tell people and when.

What to do this quarter

None of this needs a new department. A mid-size company can make real progress in a quarter.

  1. List what faces the internet. Know every login page, VPN, remote access tool and SaaS admin console your company exposes, including HR, payroll and benefits systems run by vendors. You cannot defend what is not on the list.
  2. Set a patch deadline for exposed systems. Agree a target in days, not weeks, for critical fixes on internet-facing systems, and check it is being met.
  3. Turn on phishing-resistant sign-in for admin accounts. Use hardware keys or passkeys for anyone who administers HR, finance, email or cloud systems.
  4. Test yourself the way attackers will. Ask your security team or an outside firm whether their testing uses the same kind of automated, AI-assisted tools now available to attackers. If your testing is slower than the attack, it will miss things.
  5. Use AI on defence where it is proven. Log review, alert triage and summarising incidents are reasonable places to start. Keep a person responsible for every decision to block, isolate or restore.
  6. Govern the AI tools your own teams run. If your IT or security staff use AI agents or testing frameworks, write down who may run them, against which systems, and with what credentials. A tool that can find weaknesses is dangerous in the wrong hands, including on your own network.
  7. Write the staff communication plan now. Decide who tells employees what, how quickly, and through which channel if HR or payroll systems are hit, and who handles any report to your data protection regulator.

Where I can help

Most of this sits with your security team, but the AI parts of it, from which AI tools your teams may run to where AI belongs in your defences, are governance questions. I work through them with leadership teams in my Fractional AI Lead and Advisory service. If you are still deciding where AI fits in your company, an AI readiness assessment covers security and data handling alongside use cases. And if your AI risk sits with suppliers rather than attackers, my earlier piece on the Fireworks AI breach and AI vendor risk has a checklist for that side.

Sources