On 21 October 2026, the worst week in years for the world’s most-used server language turned into an emergency for every website owner in France. Two critical flaws in PHP allow an attacker to provoke the remote execution of arbitrary code on the server. European authorities tell organisations to patch immediately. And if your French site handled personal data when it was exposed, a second clock starts: the breach notification to the CNIL. This article explains what happened, what exploitation means under French criminal law, and the exact steps a site owner in France must take now.
Searches for “php vulnerability”, “php patch” and “php websites down” surged across the United Kingdom, the United States, Canada and Australia within hours of the announcement, and the two CVE numbers are now trending queries in their own right. The audience for this piece is international — founders, expatriates and investors running French-facing websites — and the perspective is that of a Paris lawyer: what the flaw is, how French courts punish those who exploit it, and what French law demands of you if your server was exposed.
I. Two critical PHP flaws allow attackers to seize French servers without credentials
A. What CERT-EU confirmed on 21 October 2026 about CVE-2026-109618 and CVE-2026-109619
On 21 October 2026, France’s national cyber-response centre published advisory CERTFR-2026-AVI-1359, “Multiples vulnérabilités dans PHP”, dated the same day the PHP project shipped its emergency security release. The French advisory is unambiguous about the stakes. Under “Risque”, it lists a single line: “Exécution de code arbitraire à distance”. And its summary states that multiple vulnerabilities were discovered in PHP, allowing an attacker to trigger the remote execution of arbitrary code. The documented references attached to the advisory identify the two flaws tracked internationally as CVE-2026-109618 and CVE-2026-109619.
The key point for any site owner is the nature of that risk. Remote execution of arbitrary code means the attacker runs commands on the server itself, and the US National Security Agency warned that the flaws often open a direct route to the server\u2019s file system or database, as reported by the BBC. Your login page and your content management system do not stand between the attacker and the flaw, because the flaw sits underneath them, in the language interpreter itself. The flaws went undetected for more than a decade and surfaced within weeks once researchers aimed advanced AI tools at the PHP codebase, with the BBC reporting that governments and agencies internationally urged websites to fix two critical flaws in a language powering much of the internet, flaws located in PHP itself with the fix released on Wednesday afternoon.
The affected versions named by the CERT-FR advisory cover the whole installed base: all 5.x, 6.x, 7.x, 8.0.x and 8.1.x branches, plus 8.2.x before 8.2.35, 8.3.x before 8.3.36, 8.4.x before 8.4.27 and 8.5.x before 8.5.12. The old branches receive no fix at all, which means sites still running them cannot be repaired by a simple upgrade and must migrate to a supported branch. The prescribed solution is short: refer to the editor\u2019s security bulletin to obtain the patches. Most seriously for the days ahead, exploit code follows every disclosure of this magnitude within days, and the international response shows professionals take that window seriously: multiple cyber-security agencies urged organisations unable to patch in time to consider taking sites offline, with the Finnish government, Edinburgh council and the BBC taking pages down, and the Wikimedia Foundation saying it had taken other systems down as a precaution until they could be patched safely (BBC, 22 October 2026).
Every common way of running PHP on a web server is affected, so patching is not optional whatever your stack. Sites still running branches that reached end of life, PHP 8.1 and older, are in the worst position: the CERT-FR advisory lists them among the affected products, yet no security release will ever ship for them, so the only remediation is migration to a supported branch under emergency conditions. Administrators who postponed that migration for convenience now face it as an incident, with data exposure risk running while developers rebuild. If your French site is in that situation, treat the migration as containment first and modernisation second: move the application to a patched branch on isolated infrastructure, verify integrity against clean backups, and only then reconnect it to the public internet.
For a French website — a Paris boutique’s online shop, an expatriate landlord’s booking page, a startup’s SaaS landing page — the practical meaning is straightforward. If the server runs a vulnerable PHP version and faces the public internet, it must be treated as potentially compromised until patched and checked. Taking the application offline for a few hours costs money; leaving an unauthenticated remote execution flaw open while exploit code circulates costs far more, as the second part of this article shows.
International owners sometimes assume that a server located abroad puts them beyond the reach of French authorities. It does not: the moment the site targets the French public, collects data from persons in France or suffers exploitation affecting them, the CNIL considers itself competent and the French criminal courts can hear the attacker’s case. Geography of hosting has never been a shield, and during an incident is the worst moment to discover it.
B. Why websites in France went down and what exposure means for your liability
Within a day of the disclosure, trending searches included “php websites down” and “websites down”, and reports described sites taken offline worldwide as administrators applied emergency updates or pulled servers from public networks. Some of that downtime is voluntary and healthy: isolating an internet-facing application until the fix is applied and verified is exactly what CERT-EU recommends when immediate patching is not feasible. But some of it is the signature of exploitation — a defaced page, a cryptominer slowing the server, customer data quietly copied out.
Under French law, that distinction matters enormously, and it is the site owner who will have to draw it. France punishes unauthorised access to computer systems with some of the heaviest cybercrime penalties in Europe. Article 323-1 of the Criminal Code provides: “Le fait d’accéder ou de se maintenir, frauduleusement, dans tout ou partie d’un système de traitement automatisé de données est puni de trois ans d’emprisonnement et de 100 000 € d’amende. Lorsqu’il en est résulté soit la suppression ou la modification de données contenues dans le système, soit une altération du fonctionnement de ce système, la peine est de cinq ans d’emprisonnement et de 150 000 € d’amende. Lorsque les infractions prévues aux deux premiers alinéas ont été commises à l’encontre d’un système de traitement automatisé de données à caractère personnel mis en œuvre par l’Etat, la peine est portée à sept ans d’emprisonnement et à 300 000 € d’amende.”
An attacker who fires an exploit at your server commits that offence the moment access is obtained without right. The Cour de cassation has drawn the line strictly: “se rend coupable de l’infraction prévue à l’article 323-1 du code pénal la personne qui, sachant qu’elle n’y est pas autorisée, accède, à l’insu des victimes, à un système de traitement automatisé de données” (Cass. crim., 16 Jan. 2018, No. 16-87.168). Good intentions are no defence — in that case the defendant said he acted to defend his professional situation and reputation, and the Court held those motives “indifférents à la caractérisation des infractions”. Anyone who “tests” the flaw against a French site without authorisation, even a security enthusiast, exposes himself to three years’ imprisonment.
Once inside, each further step is a separate offence. Article 323-2 of the Criminal Code punishes hindering or distorting the operation of an automated data system: “Le fait d’entraver ou de fausser le fonctionnement d’un système de traitement automatisé de données est puni de cinq ans d’emprisonnement et de 150 000 € d’amende.” Article 323-3 covers what attackers usually do next with the data: “Le fait d’introduire frauduleusement des données dans un système de traitement automatisé, d’extraire, de détenir, de reproduire, de transmettre, de supprimer ou de modifier frauduleusement les données qu’il contient est puni de cinq ans d’emprisonnement et de 150 000 € d’amende.” Copying a customer database out through a PHP shell is extraction within the meaning of that article. And the attempt itself is punishable: article 323-7 states that “La tentative des délits prévus par les articles 323-1 à 323-3-1 est punie des mêmes peines.” A failed exploit attempt against your server is already a criminal offence under French law.
Criminal groups industrialise flaws like these within days, which is why the legislator added conspiracy liability: article 323-4 punishes “La participation à un groupement formé ou à une entente établie en vue de la préparation, caractérisée par un ou plusieurs faits matériels, d’une ou de plusieurs des infractions prévues par les articles 323-1 à 323-3-1”, with the same penalties as the planned offence. Convicted individuals also face additional penalties under article 323-5, including bans on civic rights, bans on practising the professional activity through which the offence was committed, and confiscation of the equipment used. French criminal law therefore treats the exploitation of this PHP flaw as serious organised crime, not as a prank — a point worth remembering before anyone suggests “checking” whether a competitor’s site is patched.
II. What French law requires of you if your site was exposed, and how to claim if you were hit
A. Patch, assess, and notify the CNIL within 72 hours where personal data was breached
The first duty is technical and immediate: apply the PHP security release for your branch, or isolate the application from public networks until the fix is applied and verified, as CERT-EU instructs. Then comes the legal assessment, and here French and European law impose a strict timetable. If your site processes personal data — customer accounts, order forms, newsletter lists, analytics — an exploited or plausibly exploited server is a personal data breach, and the General Data Protection Regulation requires notification of the supervisory authority. In France that authority is the CNIL, and its guidance leaves no room for waiting: controllers must notify the CNIL as soon as possible and no later than 72 hours, using its online breach-notification service (CNIL, Quand faut-il notifier une violation de données à la CNIL ?). The 72-hour rule flows from Article 33 of Regulation (EU) 2016/679; the CNIL page is the operational reference every French controller should follow.
Two complements matter. First, documentation: the CNIL expects you to have prepared the notification by documenting the incident internally — nature of the breach, categories and approximate numbers of persons and records concerned, likely consequences, and measures taken or planned. Start that log the moment you suspect exploitation, before memories fade and logs rotate. Second, communication to the individuals: where the incident poses a high risk to the privacy of the individuals, those individuals must be notified as well, under Article 34 of Regulation (EU) 2016/679, and the CNIL adds the pragmatic advice that in case of doubt, you notify and let the authority tell you whether informing users is necessary. A leaked customer table with names, addresses and passwords is the textbook case of high risk; a patched-before-exploitation server with clean logs is the textbook case where notification may not be required — but the assessment itself must be reasoned and written down.
Neglecting security is itself punishable in France. Article 226-17 of the Criminal Code punishes “Le fait de procéder ou de faire procéder à un traitement de données à caractère personnel sans mettre en oeuvre les mesures prescrites aux articles 24,25,30 et 32 du règlement (UE) 2016/679 du 27 avril 2016” with “cinq ans d’emprisonnement et de 300 000 euros d’amende”. Article 32 of the GDPR is the security article: appropriate technical and organisational measures, kept under review. Running an end-of-life PHP branch with a known unauthenticated remote execution flaw, months after the fix, would be difficult to present as an appropriate measure. The same week brought a vivid reminder that attackers harvest the easiest failures first: in Wales, thousands of children’s records were taken after a company left a forgotten test account protected by the password ‘Password1’ open, a story the BBC reported on 21 October 2026 that now trends alongside the PHP flaw in British searches. Patching and credential hygiene are the two halves of the same Article 32 duty, and the CNIL reads both strictly when it audits a controller after a breach. Separate from any criminal risk, the CNIL can impose administrative fines for infringements of the security provisions, and Article 83(5) of Regulation (EU) 2016/679 sets the ceiling at 4% of the total worldwide annual turnover of the preceding financial year. That scale of sanction shapes every negotiation with insurers and processors after a breach.
Our firm has published a step-by-step guide to exercising these rights from France — complaining to the CNIL and claiming damages after a breach — which complements this emergency analysis: EU data breach: complain to the CNIL and claim damages in France. Read it once the immediate containment is done; it details the complaint path and the damages heads that French courts recognise.
B. Preserve evidence, file a criminal complaint, and sue for your loss in France
Everything you do in the first 48 hours determines what you can prove in the following 48 months. Before rebuilding the server, preserve full disk images, web server and PHP logs, firewall records, and the exact version numbers and patch dates. Copy the indicators of compromise — the crafted requests, the spawned processes, the outbound connections — because the criminal complaint and the insurance claim will both turn on one question: was this server actually exploited, when, and what left it. Engage your host with a written request to preserve its own logs, and record every containment action with timestamps. A site owner who wipes and reinstalls on day one often discovers on day thirty that the only proof of the intrusion is gone.
Filing a criminal complaint in France is therefore the hinge of the whole response. Complaints for unauthorised access and data extraction can be filed at any police station or gendarmerie, or directly with the public prosecutor, and in Paris complex cyber cases are handled by specialised divisions of the Paris prosecutor’s office and the Paris court. Filing a criminal complaint serves three purposes: it triggers investigation by specialised units, it creates the dated official record on which the civil claim and the insurance file will rely, and cyber-insurance policies frequently make it a precondition of coverage. Describe the facts precisely — first suspicious request timestamps, affected systems, categories of data concerned — and attach the CNIL notification receipt, which corroborates the timeline. Because attempt is punishable in the same way as the completed offence, even a blocked exploit wave documented in your logs supports a complaint.
On the civil side, the foundation is article 1240 of the Civil Code: “Tout fait quelconque de l’homme, qui cause à autrui un dommage, oblige celui par la faute duquel il est arrivé à le réparer.” Against an identified attacker, that text supports claims for remediation costs, business interruption, customer refunds, forensic expenses and reputational harm. French courts also recognise moral harm flowing directly from cyber offences: in a September 2025 ruling, the criminal chamber held that “la perte de confiance éprouvée par M. [G], suite aux agissements de M. [L], était constitutive d’un préjudice moral découlant directement du délit prévu par l’article 323-1 du code pénal retenu à la charge du prévenu” (Cass. crim., 2 Sept. 2025, No. 24-83.605), upholding 10,000 euros for that head of loss. A company whose administrator credentials were abused by a former insider obtained that result; a company whose customer data was siphoned through an unpatched flaw can reason from the same principle, with the forensic report doing the work the employment relationship did there.
Where the attacker cannot be identified — the common case with automated exploit waves — liability shifts to the chain of providers. Against a host or maintenance contractor, the question is fault: was patching offered and refused, were alerts ignored, was the contract’s security clause honoured. Against a software vendor, French product and contract law may apply depending on the licence terms. In each case the preserved evidence decides the outcome, which is why the forensic image matters more than the invoice. For readers in Paris and the Île-de-France, the practical specifics are local: the competent criminal investigators include the Paris cybercrime brigade for serious cases, civil claims go before the Paris judicial court, and the CNIL’s online breach teleservice is the single entry point for the 72-hour notification wherever in France the controller sits. Notify your cyber-insurer within the time your policy sets — policies make prompt declaration a condition of cover, so read the clause the day of the incident, not the week after. If a processor hosts or maintains the site, demand its own incident report in writing: under data protection law the processor must alert the controller, and its responsiveness or silence will decide how liability is shared between you. Keep every receipt, every log extract and every timestamp; French procedure rewards the party whose file is complete.
Conclusion
The PHP emergency of October 2026 is a stress test of a simple principle: on the French internet, running public code means carrying legal duties. Patch the interpreter now or isolate the application; assess whether personal data was exposed; notify the CNIL within 72 hours where it was; warn the individuals at high risk; preserve the evidence; and complain and claim where loss followed. Attackers who exploit CVE-2026-109618 or CVE-2026-109619 against French servers face up to five years’ imprisonment for access, hindrance and data extraction, with attempt punished as the completed offence — and site owners who ignored the fix will struggle to explain themselves to the CNIL, to their insurer and to their customers. Act this week, document everything, and the flaw remains a bad memory; wait, and it becomes a file.