European Commission AI cyberattack — ARTEX tool named
The European Commission confirmed on November 19, 2026 (Brussels time) that it had been hit by a cyberattack carried out with an artificial intelligence tool. Spokesperson Thomas Regnier said the first signs were detected on October 22 and that several applications were taken offline as a precaution. According to press reports, the HR platform Sysper, the medical reimbursement system MyPMO and translation services were disrupted for more than 10 days. Reports citing Politico say CERT-EU, the incident response team for EU institutions, identified the use of ARTEX, an open-source AI penetration-testing framework, but the Commission has not named the tool. The Commission says it has no evidence of a broad breach of staff data and is not aware of a ransom demand. Who the attackers are, how they got in and whether any data was copied have not been established.
The three lines
- Confirmed — the Commission acknowledged the attack on November 19, 2026. First signs were detected October 22; the spokesperson called it an AI-generated cyber campaign. Affected individuals were notified and the investigation is open
- Impact — the HR platform Sysper, medical reimbursement system MyPMO and translation services were disrupted for more than 10 days. Sysper came back and went down again on November 18. Salaries were reportedly paid on time
- Not confirmed — the Commission has not named the tool. ARTEX rests on Politico-sourced reporting and a CERT-EU warning of November 5. The attackers, entry point, data exposure and number of staff affected are not public
Key questions
- European Commission AI cyberattack
- **The European Commission confirmed on November 19, 2026 that it was hit by a cyberattack using an AI tool. The first signs were detected on October 22, and staff HR, reimbursement and translation systems were disrupted for more than 10 days.** | Item | Detail | |---|---| | Confirmed | November 19, 2026 (Brussels time) | | First signs detected | October 22, 2026 | | Who spoke | Commission spokesperson Thomas Regnier | | How the Commission describes it | An AI-generated cyber campaign | | Length of disruption | More than 10 days (start and end dates not given) | | Systems affected | Sysper (HR), MyPMO (medical reimbursement), translation services | | Ransom demand | The Commission says it is not aware of one | | Attribution | None |
- What is ARTEX AI tool
- **ARTEX is an open-source AI penetration-testing framework released in July 2026. An operator states an objective in plain language, the software splits it into tasks, and AI agents carry them out with real hacking tools. It has no model of its own and connects to an external one chosen by the operator.** | Item | Detail | Source | |---|---|---| | What it is | Open-source AI penetration-testing framework | The Record, TechRadar-syndicated report | | Released | July 2026 | The Record, TechRadar-syndicated report | | How it works | Plain-language objective, task breakdown, AI agents using real tools | CERT-EU note of November 5, as summarized by Security Affairs | | AI model | None built in; connects to an external model | Same note | | Repository | Removed from GitHub by the developer in October 2026 after misuse reports; copies still circulate | The Register, Security Affairs | | FBI alert | Used by criminal and state-sponsored groups; some intrusions reached privileged access in under three hours | The Record, The Register | | Exposed consoles | More than 1,600 (FBI) / about 1,500 (Censys) | The Register / TechRadar-syndicated report |
- European Commission cyberattack data breach
- **Not established. The Commission spokesperson said there is no evidence of a broad breach of staff data, and the investigation is still open. Whether anyone viewed or copied data has not been made public.** | Question | Commission position | What reports point out | |---|---|---| | Staff data exposure | No evidence of a broad breach | Whether some data was accessed or copied is unknown | | Notification | Affected individuals were informed | How many is not public | | Ransom | Not aware of any demand | No claim of responsibility, nothing posted on the dark web | | Restoration | Most applications restored | Restoring a system does not prove nothing was copied | | Regulator | In contact with the European Data Protection Supervisor | The 72-hour reporting clock starts only once a breach is confirmed |
The European Commission confirmed on November 19, 2026 (Brussels time) that it had been hit by a cyberattack carried out with an artificial intelligence tool. Spokesperson Thomas Regnier told reporters it was an "AI-generated cyber campaign." The first signs were detected on October 22, and the Commission took several applications offline as a precaution. As a result, internal systems used by its staff for HR and pay-related tasks did not work properly for more than 10 days.
The gap between what is confirmed and what is not is wide. The Commission, the executive arm of the European Union that drafts EU law and enforces rules such as the AI Act, acknowledged the attack but did not say which tool was used, who was behind it or why. The name ARTEX comes from Politico, which broke the story, and from a warning issued on November 5 by CERT-EU, the cybersecurity service for EU institutions. Whether data left the building is still under investigation. This article separates what stopped working, what ARTEX is, and where the facts end.
1. What went down — the HR platform, medical reimbursements and translation
The affected systems are not the ones that produce EU policy. They are the internal services Commission employees use for everyday administration. The Record reported that personnel systems including payroll, leave requests, medical reimbursements and translation services were disrupted for more than 10 days. The Register and Security Affairs named the systems.
| System | What it does | Status | Source |
|---|---|---|---|
| Sysper | HR platform: payslips, timesheets, leave requests | Restored, then down again on November 18 | The Register, Security Affairs, Yahoo-syndicated report |
| MyPMO | Medical reimbursements | Restored in the week of November 16 | The Register, Security Affairs |
| Translation services | Document translation | Disrupted; reported offline on November 17 | All four (date from Security Affairs) |
| JSIS | Staff health insurance portal | Reported still down on November 19 | Security Affairs only, citing Euronews |
| Other applications | Not disclosed | Taken offline as a precaution; most restored | The Record, The Register |
Three cautions apply.
First, a system that was taken down is not necessarily a system that was breached. Regnier said the Commission shut several applications as a precaution. That means the list mixes systems the attackers touched with systems the Commission switched off itself to stop the spread. Which is which has not been disclosed. The Register says it asked whether employee data was taken and whether the shutdowns were purely precautionary, and received no answer.
Second, the "more than 10 days" has no start or end date. Detection was October 22 and confirmation was November 19, four weeks apart. No report says which part of that span the 10-plus days covers. Systems came back in stages, and at least one, Sysper, came back and failed again.
Third, salaries reportedly went out. Three outlets citing Politico sources say staff could not file timesheets or request leave but were paid on time. That is not an official Commission statement.
| Date (Brussels time) | Event | Source |
|---|---|---|
| July 2026 | ARTEX released | The Record, Yahoo-syndicated report |
| October | ARTEX developer removes the GitHub repository after misuse reports | The Register, Security Affairs |
| October 22 | Commission detects first signs of the attack | The Record (spokesperson) |
| November 5 | CERT-EU warns it has seen ARTEX-assisted attacks on organizations it covers | The Register, Security Affairs |
| November 13 | FBI issues an alert on ARTEX | The Record ("last week"), date on the alert linked by The Register |
| November 17 | Translation services offline | Security Affairs |
| November 18 | Commissioner Henna Virkkunen briefs the College of Commissioners; Sysper down again | The Register, Security Affairs |
| November 19 | Commission spokesperson confirms the attack | The Record |
2. What the Commission said, and what it did not
The Record carried the fullest account of Regnier's remarks. The other three outlets quote a statement saying the Commission took the necessary steps in response to a detected cyber campaign, that individual systems were targeted and that affected people were informed.
| Question | What the Commission said | What it did not say |
|---|---|---|
| Was there an attack? | Yes, an AI-generated cyber campaign | Which stage of the attack involved AI |
| When did it know? | First signs on October 22 | When the intrusion actually began |
| Which tool? | Not identified | Whether it was ARTEX |
| Why? | Motive not identified | — |
| Who? | Not attributed to a known group | State-backed or criminal |
| Was money demanded? | Not aware of a ransom demand | — |
| Was data exposed? | No evidence of a broad breach of staff data | Whether some data was accessed or copied |
| Current state | Most affected applications restored; other systems being checked preventively | Date of full restoration |
| Why disclose late? | Publicizing defensive measures could undermine them; nothing was hidden | — |
Regnier said the Commission remains in contact with CERT-EU and with the European Data Protection Supervisor (EDPS), the authority that oversees how EU institutions handle personal data. The Register and Security Affairs report that Henna Virkkunen, the commissioner responsible for technology, briefed her fellow commissioners on November 18. No direct statement from her has been published.
An unnamed EU official told Politico that the Commission believes it has contained most of the attack but that the full extent of any damage is not yet clear. The official said some threat actors misuse tools such as ARTEX and have targeted the Commission. Security Affairs and the Yahoo-syndicated article relay those remarks.
The four weeks between detection and disclosure is something the spokesperson addressed directly. His argument is that announcing defensive steps can help an attacker, and he said there was no hiding whatsoever. Readers can weigh that for themselves, but the sequence is a fact: the incident reached the public through a news report before it came from the Commission.
3. What ARTEX is — state a goal in words and AI agents run the hacking tools
ARTEX is an open-source AI penetration-testing framework released in July 2026. Penetration testing is authorized mock hacking that an organization commissions to find its own weaknesses. The tool was built for defense, and the same tool used without permission is an attack tool.
According to CERT-EU's technical note of November 5, as summarized by Security Affairs, it works like this. The operator writes an objective in ordinary language. The software breaks it into tasks. AI agents carry the tasks out using real hacking tools. ARTEX contains no AI model of its own and connects to an external one chosen by the operator. The Record describes it as using commercial AI models to automate vulnerability discovery and exploitation.
| Item | Detail | Source |
|---|---|---|
| Release | July 2026, open source | The Record, Yahoo-syndicated report |
| Input | An objective in plain language | CERT-EU note (via Security Affairs) |
| Processing | Objective split into tasks; AI agents execute with real tools | Same note |
| AI model | None built in; external model connected | Same note |
| Basis | An LLM layer added to ScopeSentry, a Chinese asset-discovery and vulnerability-scanning platform | Yahoo-syndicated report only |
| Repository | Removed by the developer in October 2026; forks, translations and rebrands still circulate | The Register, Security Affairs, Yahoo-syndicated report |
| Speed | Some intrusions went from initial targeting to privileged access in under three hours | FBI alert (The Record, The Register) |
| Users | Criminal and state-sponsored groups | FBI alert (The Record) |
| Exposed consoles | More than 1,600 | FBI alert (The Register) |
| Exposed instances | About 1,500, roughly 10 percent also running an attack tool | Censys (Yahoo-syndicated report) |
The removal of the repository changed little, and that matters. Once open-source code is published, anyone can keep a copy. As three outlets note, forks, translated versions and renamed builds are in circulation. The original is gone and the tool remains.
ARTEX has surfaced before, far from Brussels. It was the tool used in the breaches at South Korean financial firms that we covered on October 7, 2026 in "AI-assisted hack hits 7 Korean financial firms, 66,000 people — through back-office systems," and the FBI alert of November 13 cites the Korean cases. We summarized that alert on November 18, 2026 in "Google exploit_agent — AI that turns patches into exploits, with no confirmed hit or CVE." Financial companies in Seoul and the executive in Brussels are now linked by the name of one tool. In the Commission's case, though, ARTEX remains a reported identification, not an official one.
4. Why an AI tool matters — it changes speed and cost, not skill
Nothing in the reporting suggests AI invented a new attack technique here. The change experts point to lies elsewhere.
Shaun Williams, a senior director at the security company SentinelOne, told The Record that the main point is that AI tools like ARTEX let threat actors find and exploit vulnerabilities faster and more cheaply. Snehal Antani, chief executive of Horizon3.ai, told The Register that such frameworks let a single actor run fleets of agents with little human input. He added that attribution gets harder when state-backed and criminal groups use the same tools.
| What changes | Before | With AI penetration frameworks |
|---|---|---|
| People needed | A team splitting reconnaissance, analysis and exploitation | One operator directing fleets of agents |
| Time | Days to weeks | Under three hours in some cases in the FBI alert |
| Cost | Skilled labor | Fees for an external AI model |
| Attribution | Tools and tradecraft act as a group's fingerprint | Many actors share one public tool, blurring the fingerprint |
| Defender's window | Time to coordinate after detection | The objective may be reached before coordination happens |
The last row connects to the Commission's experience. It caught signs on October 22 and still spent close to a month taking systems down and bringing them back. Whether Sysper failed again on November 18 because the attack continued or because of problems in recovery has not been disclosed.
The attribution point showed up as well. The Commission named no attacker and said it does not know the motive. There is no ransom demand, no claim of responsibility and nothing posted on the dark web. The usual traces of a money-driven crime are missing, but that alone is not evidence of a state actor.
How much of an attack AI can now carry out on its own has come into view over recent weeks. On how such capability is measured, see "What is a CTF (capture the flag)? How a hacker contest became the yardstick for AI risk" (October 24, 2026). On who gets the strongest models first, see "What is trusted access in AI? Why Google, OpenAI and Anthropic give their best hacking-capable models to vetted defenders first" (October 30, 2026).
5. The third this year — the Commission's 2026 incident record
The Record calls this the Commission's third disclosed cyber incident of 2026.
| When | Incident | What is known | Source |
|---|---|---|---|
| January 2026 | Ivanti software breaches | EU and Dutch government bodies affected together | The Record |
| January 30, 2026 | Staff mobile device management system | Some names and phone numbers exposed; contained within nine hours | Security Affairs |
| March 2026 | Cloud infrastructure breach | A poisoned update of the security tool Trivy gave access to an AWS environment; about 91.7 GB of compressed data stolen; at least 29 other EU bodies possibly affected | The Record (links it to TeamPCP), Security Affairs (figures) |
| October–November 2026 | AI-tool campaign | HR, reimbursement and translation systems disrupted for more than 10 days | All four |
The articles do not say whether the two January rows describe the same event. Ivanti makes mobile device management products, so they may be two accounts of one incident, but we could not confirm it and list them separately. The Record says officials have not linked these incidents to one another.
What the incidents share is the kind of target: not policy files or classified material but operational plumbing. Device management, cloud accounts, HR. Systems like these hold names, contact details, pay and health-related information for an entire workforce. Security Affairs lists names, salary, home addresses and health information as categories that could have been exposed this time. That is a list of possibilities, not of confirmed losses.
6. Was personal data exposed? Restored is not the same as untouched
The Commission's sentence deserves a careful reading. The spokesperson said there is no evidence of a broad breach of staff data. He did not say there was no breach, and he did not rule out that some data was viewed. The investigation is open.
Other outlets describe the same situation from the other side. The Register says personal data was potentially exposed. Security Affairs says investigators have not confirmed whether anyone accessed the data. The Yahoo-syndicated article says it is unknown whether data was stolen. These do not contradict the Commission; they stress a different half. There is no evidence of a broad breach, and no evidence that nothing was taken.
Security Affairs adds two points. One is that bringing systems back online does not prove data was not copied. The other concerns the reporting duty. The regulation that applies to EU institutions, Regulation 2018/1725, requires a confirmed personal data breach to be reported to the EDPS within 72 hours, and the article notes that the clock starts only once a breach is confirmed. That is the author's explanation, not a position stated by the Commission.
The Commission says it has notified the people affected. How many, and what they were told was affected, has not been made public.
7. Frequently asked questions
Did an AI attack the EU on its own? Nothing supports that. The tool named in reports, ARTEX, is a framework that carries out tasks after a person types in an objective. A human chose the target. The Commission's own phrase is an AI-generated campaign, not an attack launched by an AI.
Is it certain that ARTEX was used? No. The Commission has not named a tool. Reports citing Politico say CERT-EU identified ARTEX, and CERT-EU warned on November 5 that it had seen ARTEX-assisted attacks against organizations it covers. The circumstantial case is strong, but there is no official confirmation.
Were EU policy documents or classified files taken? No report says so. The known disruption involves staff services: HR, medical reimbursement and translation. The full list of applications taken offline has not been published.
Were Commission staff paid? According to Politico sources, yes, on time. Timesheets and leave requests were blocked. The Commission has not officially confirmed this.
Was it ransomware? The Commission says it is not aware of a ransom demand. Reports say no group has claimed the attack and no data has appeared on the dark web. There is currently no basis for calling it ransomware.
Does this matter outside Europe? The tool is public and its copies are everywhere. The FBI says criminal and state-sponsored groups both use it, and the same name appeared in the South Korean financial breaches of September and October 2026. A framework that anyone can download does not respect borders.
What can an organization do now? The experts quoted all point to speed. If an intrusion can finish in hours, the first things to fall are known vulnerabilities left unpatched, default passwords and management consoles open without authentication. Closing old holes quickly comes before defending against novel techniques.
8. What remains open
| Item | Confirmed | Not confirmed |
|---|---|---|
| The attack | Acknowledged by the Commission on November 19, 2026 | When the intrusion began |
| Detection | October 22 (spokesperson, via The Record) | Corroboration from other outlets |
| Tool | CERT-EU's ARTEX warning of November 5; Politico-sourced reporting | Official confirmation; whether CERT-EU or the Commission's own team detected it |
| Role of AI | The spokesperson's phrase, AI-generated campaign | Which stages AI handled; which model was connected |
| Entry point | — | Not disclosed at all |
| Systems affected | Sysper, MyPMO, translation services | JSIS (single source); the list of other applications taken down |
| Length of disruption | More than 10 days | Start date, end date, full restoration |
| Pay | Sources say salaries arrived on time | Official confirmation |
| Personal data | No evidence of a broad breach (Commission) | Partial access or copying; number of people notified |
| Attackers | No attribution, no ransom demand | State-backed or criminal; motive |
| Earlier incidents | Third disclosed incident of 2026 | Any link to the January and March incidents; officials have drawn none |
| ARTEX exposure | More than 1,600 (FBI), about 1,500 (Censys) | Why the two counts differ |
In short: it is established that the EU's executive has itself acknowledged an attack using an AI tool, and that staff systems were unstable for more than 10 days. That the tool was ARTEX is likely but still at the level of press reports, and who did it, why, and whether data left remain blank. Three things are worth watching next: whether the Commission or CERT-EU names the tool and entry point in an official document, whether a personal data breach is reported to the EDPS, and what explains Sysper going down a second time.
For the Korean cases involving the same tool, see "AI-assisted hack hits 7 Korean financial firms, 66,000 people — through back-office systems" (October 7, 2026). For the FBI alert on ARTEX and an attack framework built to run with no human in the loop, see "Google exploit_agent — AI that turns patches into exploits, with no confirmed hit or CVE" (November 18, 2026).
Sources
- The Record — European Commission confirms cyberattack, says hackers used AI tool (Alexander Martin, 2026-11-19)
- The Register — European Commission scrambles to contain AI-assisted cyberattack (Carly Page, 2026-11-19)
- Security Affairs — European Commission Cyberattack: ARTEX AI Tool Suspected in Prolonged Disruption (Pierluigi Paganini, 2026-11-19)
- Yahoo News (syndicated, TechRadar family) — European Commission says it was hacked by attackers using AI tools (Sead Fadilpašić, 2026-11-20)