P.K. SHARMA

Cyber security intelligence, AI governance, practitioner analysis

Commission: AI attack, no evidence of staff data breach, public 28 days after detection; only POLITICO says ARTEX

The Commission's spokesperson, as quoted on 19 November 2026, puts first signs at 22 October, 28 days before the public confirmation (derived), calls the attack AI generated and sees no evidence of a staff data breach. POLITICO prints the name ARTEX; the spokesperson is not quoted saying it.

By · LinkedIn · · 18 min read

28 days from first signs to a public word, and a tool name the Commission has not said

The Record reports that the Commission's spokesperson put the first signs of the attack at 22 October, and the Commission confirmed the attack to journalists on Thursday 19 November 2026. That is 28 days (derived: 22 October to 31 October is 9 days, plus 19 days in November). Infosecurity Magazine adds that the intrusion was confirmed as a cyberattack on 26 October, which is 24 days before the public confirmation (derived: 5 + 19). The Record says the confirmation came "following media reports of the incident", and calls POLITICO the first to report it; the POLITICO page data gives 05:26:00 UTC on 19 November.

What that does not establish. The 28 days is the gap between two dates. It is not a measure of how long anyone had access, and no saved page states an access window. It does not show that the Commission hid anything: asked why the incident was not disclosed earlier, the spokesperson said, as The Record reports, that publicizing defensive measures could undermine their effectiveness, and he told journalists: "There was no hiding whatsoever". It does not show that systems were down for 28 days: POLITICO says "Disruption has lasted over 10 days". And it says nothing about the tool, which is the second thing this brief is about.

The tool name. As quoted, the spokesperson calls the attack "AI generated", done with a "publicly available" tool, and The Record says he did not identify the tool or the attackers' apparent motivation. The name ARTEX appears in POLITICO's item in POLITICO's own voice, with CERT-EU as the detector, and in CERT-EU's own blog post of 5 November, which does not name the Commission. No saved page contains a Commission or CERT-EU document that names ARTEX as the tool used against the Commission. This brief keeps three things apart: the spokesperson's "AI generated", POLITICO's "ARTEX", and CERT-EU's "highly likely" about its own constituency.

Who is speaking, and what each wants. The Commission is the victim and the source of the official statement; its interest is to limit damage and, as with any victim, to minimise. POLITICO's sources are anonymous: it quotes "an EU official, granted anonymity because of the sensitivity of the issue" and a leaked internal mail. CERT-EU is the cybersecurity service for the Union institutions, bodies, offices and agencies, so it serves the Commission among others and is not independent of the victim. SentinelOne sells security products and ISACA is a professional body; both are quoted below as opinion. This brief names no individual: the spokesperson, reporters and commentators appear by role only.

The five dates, kept apart

The rule for an incident like this is to keep the access window, discovery, containment, notification and publication apart and to infer none from another. The table does that with the saved pages. Wording differs between pages, so each wording is shown.

What the saved pages say for each date the breach rules ask for. Sources: The Record, Infosecurity Magazine and POLITICO, 19 November 2026. Day counts are derived.

Date or windowWhat the saved pages sayStatus
Access window (from when to when an intruder had access)Nothing. Infosecurity says no information was given on perpetrators or on whether any data was accessed, with the investigation ongoing.Not stated
Discovery22 October: "the first signs of the attack were detected October 22." (The Record). Infosecurity: the intrusion was detected on 22 October 2026. POLITICO: CERT-EU "detected earlier this month" that hackers used ARTEX, a different detector and a different month.Stated; wording and detector differ
Confirmation as a cyberattack26 October, printed by Infosecurity only. The Record prints no such date.Stated by one page
ContainmentSeveral applications temporarily taken down and "most of them have since then been restored". An anonymous official: the Commission believes it has contained most of the attack. Translation services were hit on Tuesday 17 November and Sysper briefly crashed again on Wednesday 18 November (weekdays derived).Not stated as a date; partial
NotificationSpokesperson to POLITICO: "we have informed those concerned". The Record: notified affected individuals. Infosecurity: impacted teams informed. The European Data Protection Supervisor "has been notified". No date for either.Said to have happened; no dates
Publication19 November: POLITICO's newsletter at 05:26:00 UTC (page data), then the Commission's confirmation to journalists the same day.Stated
Drawn by this site from the four saved pages and the FLASH text saved for brief 709. Axis from the site's own scale module, one tick a day, UTC. Day counts are derived.

Three smaller gaps follow from the same dates. The spokesperson's 22 October and Infosecurity's 26 October are 4 days apart (derived). The CERT-EU post, printed as 10:05:00 PM CET on Thursday 5 November, which is 21:05 UTC because CET is one hour ahead (derived), came 14 days after 22 October and 10 days after 26 October (derived), and 14 days before the public confirmation. The FBI FLASH on ARTEX is dated 13 November, 22 days after 22 October and 6 days before the confirmation (derived). The FLASH says the original developer made the repository unavailable on 8 October 2026, 14 days before the first signs (derived), and CERT-EU says the public versions are now independent forks, so the name covers more than one body of code.

Every derived interval used in this brief, with the arithmetic. Dates are from the saved pages; counts are calendar days.

IntervalArithmeticResult
First signs to public confirmation22 Oct to 31 Oct = 9; plus 19 Nov days = 2828 days
Confirmed as an attack to public26 Oct to 31 Oct = 5; plus 19 = 2424 days
CERT-EU post to public5 Nov to 19 Nov = 1414 days
First signs to confirmation26 less 22 = 44 days
First signs to CERT-EU post9 + 5 = 1414 days
Confirmation to CERT-EU post5 + 5 = 1010 days
First signs to FBI FLASH9 + 13 = 2222 days
FBI FLASH to public19 less 13 = 66 days
Repository unavailable to first signs8 Oct to 22 Oct = 1414 days
Start of a disruption of over 10 days19 Nov less 10 days = 9 Nov9 Nov or earlier
First signs to translation services hit22 Oct to 17 Nov = 9 + 17 = 2626 days

Records, accounts and people. No saved page counts records, accounts, devices or people, and none lists a data field. The systems named are Sysper, which POLITICO calls "the Commission's HR platform for payslips, leave requests and timesheets", MyPMO, its medical reimbursement system, and translation services. POLITICO's opening line says the incident left staff unable to input salary and leave information, but its later paragraph says "salaries landed as usual" and that some officials could not submit timesheets or request leave; The Record says payroll was disrupted. Whether anything was exposed, retrieved, copied or misused is not stated, and the investigation is open.

Who says what: no evidence of a breach against potentially compromised data

Drawn by this site from the saved pages. The spokesperson column is as The Record, Infosecurity Magazine and POLITICO quote the spokesperson; the press briefing itself was not read.

The data question has two answers on two different questions. The spokesperson, as Infosecurity quotes him, said the investigation has not detected "any evidence of a data breach affecting Commission staff" and that he was "not aware" of a ransomware demand; The Record paraphrases the same point as no evidence of a data breach affecting Commission staff "broadly" and no ransom demand. POLITICO's opening line says the incident "has potentially compromised swaths of personal data", in POLITICO's own voice, and its anonymous official says investigators "are still trying to establish whether staff data was compromised". The two do not contradict each other as printed: one reports what investigators have found, the other what may be affected. No evidence is not the same as no breach, and the spokesperson's sentence is scoped to staff; the saved pages say nothing about anyone else. Infosecurity itself writes that the attack "may have breached the personal information of employees".

Individual systems, ten days. A Commission spokesperson, in a statement to Infosecurity and to POLITICO, said "Individual systems were targeted". POLITICO and The Record report disruption across HR, reimbursement and translation for more than 10 days. Several individual systems can be disrupted for a long time, so this is not necessarily a conflict, but no page says how many systems "individual systems" means. POLITICO adds that three officials complained about the lack of communication, and that an IT reply seen by POLITICO about a Sysper problem cited a general issue "without mentioning a cyberattack".

An escalation, and a third incident. POLITICO says, citing a leaked internal mail, that the Corporate Management Board met on "ongoing cyber incidents" and that the digital services department ordered stepped-up security checks; the plural is as printed and is not explained. The Record says the attack is the Commission's third disclosed cyber incident this year, after breaches involving Ivanti software in January and its cloud infrastructure in March, and that officials have not linked the incidents. No saved page says whether the three share a cause, an actor or a weakness.

A tool name does the work of evidence

Three different things are being said, and the headline versions run them together. The spokesperson's claim is about method. Infosecurity quotes him: "AI assisted tools can be used and especially misused to detect vulnerabilities" and "Indeed we can confirm that this was done with AI tools". That is a claim that AI tools were used. It gives no number of targets, no speed, no scale, and no way to tell what the tools did that a human team could not.

POLITICO's claim is a name. Its item says CERT-EU detected earlier this month that international hackers used ARTEX, and in the next sentence its official says, "Some actors misuse such tools and have targeted the Commission" without the name. The Record relays that POLITICO said CERT-EU had identified the attackers' use of ARTEX. If "earlier this month" means November, CERT-EU's detection came after the 22 October first signs, so the tool may have been named later than the intrusion was seen; the saved pages do not say who identified what, or when.

CERT-EU's claim is about its own constituency. Its post of 5 November says it "has observed similar instances of AI-driven activities against its constituency" and that these "highly likely involved the use of ARTEX". It does not name the Commission or any other entity, and the word constituency is not defined in the post beyond the closing paragraph that calls CERT-EU the cybersecurity service for the Union institutions, bodies, offices and agencies. The post is a general warning, written to "raise awareness of the widespread use of ARTEX by threat actors".

Drawn by this site from the four saved pages, read on 19 November 2026.

What the name does not tell a reader. CERT-EU describes ARTEX as open source and says "ARTEX does not include its own AI model", that operators supply a model of their choosing, that the original author made the repository unavailable in October 2026, that "The publicly available versions are now independent forks" and that "Integrations and capabilities may differ between forks". So the name does not identify a code version or a model, and no saved page says which, if any, was used at the Commission. CERT-EU also says the tool "is designed to be observed and steered by a human operator" through a dashboard with an approval queue, and that it relies on rules and the model to respect scope, with "no hard technical enforcement". An attack can therefore be both AI generated and human steered; no saved page says which this was. CERT-EU's statements about speed, scale and persistence are its assessment of the tool in general, not a finding about this incident.

"Chinese" is Infosecurity's word. Infosecurity writes that the POLITICO article claimed the attack used "the Chinese AI penetration-testing tool ARTEX". The POLITICO item this site read, in full, does not contain the word Chinese, and neither does the CERT-EU post outside the web addresses in its reference list. The FBI FLASH, as saved for brief 709, says "Chinese models are most prevalent" in attacker-controlled instances of ARTEX, which is a statement about the models that operators plug in, not about where the tool comes from and not about the Commission. No saved page names an actor for the Commission attack.

Two comments, attributed as opinion. A senior director at SentinelOne, which sells security products, says in The Record that "The key takeaway here is not that threat actors are using Artex" but that widely available AI tools let threat actors find and exploit flaws in internet facing devices faster, more effectively and cheaper. A chief global strategy officer at ISACA, a professional body, tells Infosecurity that AI is enabling attackers to hack "at the speed of intent" and that the Commission will investigate "what weakness hackers were able to use ARTEX to exploit". Both are general comments. The second assumes ARTEX was used, which rests on POLITICO's item; neither adds a fact about this incident.

Where the sources differ

None of these is resolved here. Each row sets one page's wording beside another's. Some rows are small and some are not; all are listed so that a reader can see what the headlines averaged out.

Differences between the saved pages on 19 November 2026. Sources: POLITICO, The Record, Infosecurity Magazine and CERT-EU, as labelled.

PointOne pageAnother page
Data breachSpokesperson via Infosecurity: no evidence of a data breach affecting Commission staff. The Record adds "broadly".POLITICO opening line, own voice: potentially compromised swaths of personal data. Official: whether staff data was compromised is still being established.
How many systemsSpokesperson: individual systems were targeted.POLITICO and The Record: more than 10 days of disruption across HR, reimbursement and translation.
Who named the toolThe Record: the spokesperson did not identify it. Infosecurity: he called it publicly available.POLITICO: ARTEX, in its own voice. Infosecurity attributes "Chinese AI penetration-testing tool ARTEX" to POLITICO; the POLITICO item does not print Chinese.
First signs and confirmationThe Record: first signs 22 October, no 26 October.Infosecurity: detected 22 October and confirmed as a cyberattack 26 October.
DetectionSpokesperson: first signs 22 October.POLITICO: CERT-EU detected the use of ARTEX "earlier this month", which reads as November.
The ransomThe Record: unaware of any ransom demand.Infosecurity: "not aware" of a ransomware demand.
Who was toldThe Record: affected individuals notified.POLITICO: those concerned informed. Infosecurity: impacted teams informed.
PayPOLITICO opening line: staff unable to input salary and leave information. The Record: payroll disrupted.POLITICO body: salaries landed as usual; some could not submit timesheets or request leave.
Senior meetingPOLITICO opening line: senior officials discussed it at their weekly meeting yesterday.POLITICO body: the Corporate Management Board met earlier this month, per a leaked mail; the tech commissioner briefed counterparts at yesterday's College meeting, per two officials.
What was taken offlineSpokesperson via Infosecurity: several applications.The Record: several services offline, as a protective measure.

What the pages state, and what they do not

What the four saved pages state and do not state about the Commission incident, as read on 19 November 2026.

QuestionWhat the pages printStatus
What data, if any, was accessedInvestigation ongoing; no evidence of a staff data breach (spokesperson); potentially compromised personal data (POLITICO)Not stated
The entry route or weaknessNothing. Infosecurity quotes a commentator saying the Commission will investigate it.Not stated
The actor and motiveThe Record: motive not identified. Infosecurity: nothing on perpetrators.Not stated
Whether ARTEX specifically was usedPOLITICO's item says so; the spokesperson, as quoted, does not name the tool.Not stated on the record
What "AI generated" means in practice"AI assisted tools can be used and especially misused", as quoted.Not stated
Whether the three incidents this year are linkedIvanti in January, cloud in March; officials have not linked them.Not linked, not ruled out
Why disclosure came 28 days after first signsSpokesperson: publicizing defensive measures could undermine them; no hiding.An answer, as given
The costNothing.Not stated
How many people or recordsNothing.Not stated
Whether an AI tool did work a human could notNothing. CERT-EU describes the tool in general.Not stated

What CERT-EU prints for defenders, and what is missing

The only technical material on the saved pages is in CERT-EU's post of 5 November, written about ARTEX in general and not about the Commission. It prints three ATT&CK techniques, copied in the table, and a list of what to do in three groups: 4 immediate, 4 short term and 6 continuous items (derived count, 14 in all). The post prints no indicators, no addresses, no hashes and no tool signatures, and none of the four pages prints an indicator for the Commission incident. It tells readers to hunt for ARTEX artefacts such as automated scanning patterns without listing them.

The three ATT&CK techniques CERT-EU prints, as printed. Source: CERT-EU blog post of 5 November 2026, read 19 November 2026.

ATT&CK IDTechnique, as printedWhat CERT-EU attaches
T1595Active ScanningListed under TTPs; no further detail
T1588.002Obtain Capabilities: ToolListed under TTPs; no further detail
T1190Exploit Public-Facing ApplicationListed under TTPs; no further detail

In plain words, the immediate items are to limit internet exposure to the assets that need it, to isolate management and control panels, to patch exposed assets as an emergency and to watch privileged account activity. The short term items are multi factor authentication on critical systems, hunting for connections to known or suspected actor infrastructure and for ARTEX behaviours in logs, and endpoint detection to block unusual processes. The continuous items include sanitising error responses from internet facing devices, a web application firewall, logging for exposed assets, monitoring that also covers AI tools, regular scans and tests, and keeping non severe findings in the fix cycle so that tooling cannot chain small misconfigurations into larger paths.

Context from brief 709, and one open item in it. Brief 709 covers the FBI FLASH of 13 November on ARTEX. The Record, in the Commission story, repeats the FBI warning that some intrusions went from initial targeting to privileged access in under three hours; brief 709 found the FLASH rests that on one vendor's post about one actor's cases, and no saved page says how long any intruder took at the Commission. The same brief could not read the CERT-EU post, which is endnote 8 of the FLASH, cited with two others for approximately 67,000 individuals in South Korean attacks. The copy of the CERT-EU post saved today prints no such figure, no count of individuals and no mention of South Korea outside the addresses in its reference list.

What to do, in order

The first five items are CERT-EU's, in its own order of urgency; the last two are this site's suggestions and are labelled.

Take this with you

Actions, CERT-EU's first

  • CERT-EU, immediate: identify internet exposed assets, limit exposure to those that need it, and isolate management and control panels from the internet or untrusted networks.
  • CERT-EU, immediate: prioritise emergency patching of every internet exposed asset.
  • CERT-EU, immediate: monitor privileged and administrator account activity for anomalous behaviour.
  • CERT-EU, short term: multi factor authentication on all critical systems and services.
  • CERT-EU, short term and continuous: hunt in logs for automated scanning patterns and tool signatures, and sanitise error responses from internet facing devices.
  • This site's suggestion, from the reported disruption: for HR and payroll platforms, keep a manual process for timesheets and leave if the system is offline, and decide in advance who tells staff what is happening, since three officials told POLITICO they lacked communication.
  • This site's suggestion: read "AI generated" as a claim about method, not about scale, until a source gives numbers; ask incident responders and vendors for the access window and the containment date, not the tool name.

The question the Commission's statements leave open

A tool name can be read out in a minute and the Commission has not read it out; a newsletter has. An access window is the number a staff member needs and it is on none of the four pages. Which answer would tell a staff member whether their own data was touched: the name of the tool, which is in the news, or the access window, which is in none of the four pages?

Key facts

Sources

  1. PrimaryCERT-EU blog post 'From penetration testing to cyber threat: beware of ARTEX', Thursday 5 November 2026, 10:05:00 PM CET: ARTEX description, highly likely use against its constituency, three ATT&CK techniques, the advice lists. The Union institutions' own security service. Saved about 17:23 UTC on 19 November 2026CERT-EUaccessed 2026-11-19
  2. PrimaryFLASH-20261113-001 on ARTEX, 13 November 2026: date, the repository made unavailable on 8 October, Chinese models, endnote 8. Read as the text copy saved for brief 709 on 14 November 2026; not fetched again for this briefFBI (Internet Crime Complaint Center)accessed 2026-11-14
  3. Reported byBrussels Playbook newsletter of 19 November 2026 (page data published 05:26:00 UTC), the item on the Commission cyberattack: the first report, anonymous EU officials, a leaked internal mail, ARTEX in POLITICO's own voice, over 10 days of disruption. Only that item was read. Saved about 17:23 UTC on 19 November 2026POLITICO Europeaccessed 2026-11-19
  4. Reported byThe Record, 19 November 2026: the Commission confirmation, first signs 22 October, AI-generated cyber campaign, tool not identified, third incident this year. A secondary report; robots.txt returned 404. Saved about 16:31 UTC on 19 November 2026The Record from Recorded Future Newsaccessed 2026-11-19
  5. Reported byInfosecurity Magazine, 19 November 2026: first signs 22 October, confirmed 26 October, applications taken down, no evidence of a staff data breach, the word Chinese attributed to POLITICO, two comments. A secondary report. Saved about 16:31 UTC on 19 November 2026Infosecurity Magazineaccessed 2026-11-19

Share this briefing

Know someone who owns this problem? Send it to them.

Related briefings

The briefing, in your inbox

Practitioner analysis of cyber and AI security news. No vendor noise.

How often

Every new briefing in one email, at 7am, or at 7am, 12:30pm and 6pm. Nothing is sent when nothing is new. Unsubscribe any time.