Skip to main content

EU Commission confirms AI-assisted cyberattack on internal systems

Ilustrační obrázek
On 22 October 2026, security teams at the European Commission detected an intrusion that used AI tools. The Commission confirmed it internally four days later, on 26 October. The public learned about it on 19 November, after press inquiries. Several internal administrative applications were taken offline — a set the Commission has not itemised. About 40,000 staff across 27 member states rely on those systems.

The details are sparse, and that matters. The Commission has not named a threat actor or a state. It has not reported a ransom demand. Investigators say they have found no evidence of a data breach affecting staff as a whole. What is confirmed is the use of AI tools in the intrusion and the disruption that followed.

Confirmed details

Commission spokesperson Thomas Regnier confirmed that digital applications were taken offline as a preventative measure, and that most have since been restored, The Independent reported when the Commission confirmed the incident publicly on 19 November.

That gap between detection and disclosure is one of the few firm timelines available. Detection: 22 October. Internal confirmation: 26 October. Public confirmation: 19 November — 28 days after the first detection. The Commission has not published a full incident report, and it has not said when each individual service came back online.

Services taken offline

The affected systems were several internal administrative applications, which the Commission did not itemise. They are not the most sensitive databases in Brussels, but they are the machinery that lets an administration function: access, approvals, records, internal requests. Taking them offline is a containment measure, and it costs working time even when nothing leaks.

The Commission says most services are back. It has not specified when each service returned, so any figure for the total length of the disruption would be an estimate rather than a published fact.

About 40,000 staff across 27 member states work inside the network concerned. Spread evenly, that would be roughly 1,480 people per member state, though the Brussels footprint is much more concentrated. The number matters because it shows the scale of the internal network involved. It is not a single office with a few laptops.

Breach checks

So far, EU investigators have found no evidence of a data breach affecting Commission staff as a whole. No ransom demand has been reported.

That "no breach" framing needs care. Published reports have described attackers claiming to have exfiltrated roughly 350 GB from Commission cloud systems. It is not clear from the public record whether that claim refers to this October incident or to an earlier one. The position as it stands: the Commission has confirmed no breach, while a claim of exfiltration has been reported and remains unverified.

Those facts limit the known damage, but they do not close the case. An intrusion can be disruptive without a confirmed exfiltration of personal data. Systems can be taken offline to stop the spread before investigators know exactly what was touched.

The Commission has also not attributed the incident. That is normal in the early stages of an investigation, but it leaves the public without a clear picture of who was behind it or what they wanted. A state-backed group, a criminal crew and a lone operator are all possible from the public record. The confirmed facts do not choose between them.

CERT-EU warning

On 4 November, CERT-EU, the cybersecurity service for EU institutions, warned about publicly available AI tools that automate vulnerability discovery. The advisory describes a category of software that accelerates the hunt for weaknesses; the publicly available material does not tie a specific product to the Commission incident.

That is a narrower claim than "an AI launched the attack." The warning describes a class of tools that helps people find weaknesses faster. Scanning a network for open services or known bugs used to take time and skill. Automated tools reduce both. For an IT team, the practical change is speed: more probes, more attempts, less time between a weakness appearing and someone trying to use it.

The Commission incident fits that broader trend, but the public facts do not show which tool, if any, was used against it. CERT-EU's advisory is addressed to European organisations. It points to a risk that exists beyond Brussels.

AI Act limits

The EU's AI Act is not a cyber-defence law. Its transparency rules, which took effect on 2 August 2026, require deployers to tell users when they are interacting with a chatbot and require providers to make AI-generated output and deepfakes detectable and visibly labelled. The European AI Office also has enforcement powers over general-purpose AI models, with administrative fines reaching up to €15 million or 3% of worldwide turnover for non-compliance.

Those rules matter for deception, provenance and market oversight. They do not patch a server, manage access rights or detect an intruder inside an internal administrative system. The Commission has not said that this incident triggers a specific AI Act obligation. Cyber incidents involving EU institutions are handled through other security and data protection channels.

For European organisations, the useful distinction is simple. The AI Act asks who built an AI system and how it is used. Network security asks whether an outsider can reach a service they should not reach. The Commission's case touches both areas, but the legal responses are separate.

Open questions

The public record still lacks several details. Which systems were accessed, rather than merely taken offline? How many staff records were exposed, if any? Was any data copied? Does the reported 350 GB claim belong to this incident at all? Did the intruders use an AI tool only for reconnaissance, or also for exploitation? The Commission has not published answers to those questions.

What is confirmed is narrower. An AI-assisted intrusion was detected on 22 October. Several internal administrative applications were taken offline, and the Commission has not said when each was restored. About 40,000 staff across 27 member states were within the affected network. Investigators have found no widespread staff data breach and no reported ransom demand, while a separate claim of data exfiltration has been reported and is unverified. The Commission says most services have been restored. The investigation continues without a public attribution.

Was this attack run by an AI on its own?

The confirmed facts say AI tools were used in the intrusion. They do not show that an AI system planned or ran the operation without human direction. CERT-EU's warning is about publicly available tools that automate vulnerability discovery, which is a narrower claim.

Did any personal data leak?

EU investigators have found no evidence of a data breach affecting Commission staff as a whole. Reports of a claimed exfiltration of roughly 350 GB have been published, but it is unclear whether that claim concerns this October incident or an earlier one, and it remains unverified. "No confirmed breach" is not the same as a guarantee that nothing was touched.

Should small European organisations change anything?

CERT-EU's 4 November advisory is aimed at European organisations, not only EU institutions. It describes AI tools that speed up vulnerability discovery. The advisory does not name a single exploit that a small organisation can block with one setting.

Discussion

No comments yet — be the first to share your thoughts.
X

Don't miss out!

Subscribe for the latest news and updates.