Artificial intelligence tools are no longer just used to defend European IT systems: they are also used to attack them. THE November 19, 2026there European Commission confirmed that it was targeted in October by a campaign of cyberattacks carried out usingAI tools. Several internal services were disrupted for more than ten days, and the investigation cannot yet determine whether personnel data was stolen.
What the Commission confirmed
“I can confirm that the Commission has recently taken the necessary measures in response to a detected cyber campaign,” said its spokesperson, Thomas Regnier. “As a protective and preventive measure, we temporarily took several apps offline, and most have been restored. The investigation is still ongoing. »
According to the spokesperson, the incident was detected on October 22, 2026. He indicated that there was no no evidence of a data leak affecting all Commission staff, that specific systems had been targeted and that the people concerned had been informed. He refused to name the alleged perpetrators and said he was not aware of any ransom demand.
Internal services cut for more than ten days
As a precaution, the Commission cut several applications. According to Politico, two essential platforms for staff were affected:
| Service | Role | Situation |
|---|---|---|
| Sysper | Human resources platform: pay slips, leave, working time | Restored mid-November, new brief outage on November 18 |
| MyPMO | Reimbursements of staff medical expenses | Recovered mid-November |
| Translation services | Internal translation tools | Disturbed on November 17 |
| Other apps | Not specified | Cut as a precaution, most restored |
Salaries were paid normally for the most part, according to Politico, even if some agents were unable to complete routine procedures. The matter was considered serious enough for the board of directorsthe Commission’s highest internal crisis management body, meets urgently. The Commissioner responsible for technological sovereignty and security, Henna Virkkuneninformed his colleagues on November 18.
ARTEX, a diverted testing tool
According to a European source cited by Politico, the CERT-EUthe team responsible for cybersecurity of the Union institutions, found that hackers were using ARTEX. This is a free penetration testing tool based on AI: it uses large language models to coordinate several agents who probe a system in search of flaws.
ARTEX was intended for authorized safety testing. According to Notebookcheck, its developer removed the code repository in October after reports of abuse, but copies continue to circulate. Several points remain unclear: it is unknown whether the tool allowed the hackers toenter in the systems or if it was only used to look for flawsand the Commission itself did not name the tool in its November 19 statement.
The question of personal data
The main area of concern concerns the personnel data. Shutdown services show which applications were unavailable, not what information the attackers were able to view. The Commission has not confirmed that any pay slips, leave data or medical information was exposed. According to an official cited by Politico, the institution believes it has contained most of the attack, but the exact extent of the damage remains unknown.
This uncertainty partly explains the little information transmitted to staff. It also illustrates a classic difficulty: establishing precisely what an attacker saw or copied can take weeks of analysis of technical logs.
A threat that changes scale
This attack comes in a context where AI is transforming cybersecurity in both directions. On the defenders’ side, ENISA and CERT-EU themselves use AI models to spot flaws, as we explained in relation to the opening of OpenAI’s Daybreak program to Europeans. On the attackers’ side, the same capabilities make it possible to probe faster and wider the systems.
ENISA has also warned that an attacker can now exploit a flaw fifteen minutes after its disclosure. AI agents capable of chaining reconnaissance, vulnerability detection and access attempts further reduce this time, and allow less experienced groups to carry out attacks formerly reserved for experienced teams.
The incident adds to a series of high-profile cyberattacks in 2026. Irish broadcaster RTÉ notably recalls the hacking, in October, of the commercial refrigerators and freezers of hundreds of French supermarkets, but also attacks on ships bound for the United States in August, against American water treatment plants in July and against the Los Angeles transport network in March. In the United States, a nationwide flight shutdown lasting several hours and outages affecting banks and Microsoft applications have also been attributed by experts to hastily deployed patches to close recently discovered flaws. The role of AI is not established in all of these cases, but the rate of incidents is accelerating.
This is not the first incident targeting the Commission: according to Cyber Security News, attacks had already affected the platform of the public Europa site at the beginning of the year, with no established link with the October incident.
Lessons for organizations
The attack on the Commission recalls principles valid for all organizations, public and private:
- apply fixes quickly: AI tools automate the search for known vulnerabilities, and an unpatched vulnerability is spotted much more quickly than before;
- limit exposure: reduce the number of services accessible from the Internet and segment networks so that an intrusion does not spread;
- monitor abnormal behavior: automated agents generate unusual volumes of requests, detectable by careful supervision;
- prepare for crisis management: the ability to quickly cut applications, as the Commission did, limits the damage;
- test your own defenses: AI penetration testing tools, used authoritatively, make it possible to find vulnerabilities before attackers.
A wake-up call for Europe
That an institution as exposed as the European Commission would be hit by an AI-assisted attack shows that the threat is no longer theoretical. It relaunches the debate on European defenders’ access to the most efficient AI tools, and on Europe’s capacity to develop its own. The conclusions of the investigation, expected in the coming weeks, will have to specify the exact role of AI in the intrusion, the possible identity of the authors and the extent of the data affected. Our cybersecurity section will follow the consequences of this affair.

