Sommaire(12)MasquerAfficher
- À quoi sert un pentest, concrètement ?
- Ce qu'un pentest ne fait pas
- Pourquoi les entreprises en ont besoin en 2026 ?
- Quelle surface un pentest doit couvrir en 2026 ?
- Boîte noire, grise ou blanche : quel test pour quel besoin ?
- Tous les combien de temps faut-il tester ?
- Ce que la loi demande vraiment en 2026
- Combien coûte un test d'intrusion ?
- Ce qu'une entreprise doit exiger avant de payer
- Comment vérifier que la correction a eu lieu ?
- Pour aller plus loin
- FAQ
A penetration test is used to determine what an attacker would actually obtain on your system, at a given moment, within a given scope. It is a measurement, not a certificate, and that is precisely what makes it useful: it describes a real exposure, not a security intention.
It does not replace continuous monitoring or a compliance audit. A company that buys it to tick a box therefore pays for a report it will not read, while the benefit lies elsewhere: knowing where someone would get in, before someone does so for real.
What is a pentest actually for?
A pentest answers a question that no one inside the company can settle from within: where would someone get in who has no rights, no documentation, and no courtesy of going through the IT department. The answer is a sequence of paths, each with its own proof.
The PTES describes this process in seven phases, from initial scoping to the report. NIST SP 800-115 frames it on the American institutional side, and ANSSI publishes the corresponding French doctrine. These three frameworks say the same thing: an intrusion without a written mandate is not a test, it is an offense.
What a pentest does not do
It does not prove the absence of a flaw: it proves the presence of the paths it found. A report that states "no access obtained" does not say that the system is secure, it says that the team did not find a door within the paid time. This is a nuance that few providers write, and that changes the reading of the deliverable.
It also does not cover what was not in scope. A web application tested on its own says nothing about the exposure of a forgotten subdomain, a mail server or an administration interface left open. This is the most frequent limitation of short engagements, and the most costly.
Why businesses need it in 2026
Three pressures accumulate, and none of them is resolved by a statement of intent.
- Regulatory: NIS2 broadens the circle of organizations subject to security obligations, and DORA imposes a testing plan on financial actors.
- Insurance-related: cyber insurers increasingly require a dated status report, not a stated intention.
- Technique: the time between the publication of a vulnerability and its exploitation has shortened, so an annual snapshot is no longer sufficient.
The best argument a security manager can show to their leadership is not a score, it is a list: the vulnerabilities actually exploited, and the share of your estate that is exposed to them. Our catalog provides both, with the date of the assessment.
What scope should a pentest cover in 2026?
The perimeter of five years ago fit in one sentence: the public site and its server. That is no longer the case. Programming interfaces expose data without going through a page, and a single one of them that is poorly protected gives the same result as an application flaw.
Four areas come up in almost every engagement: APIs, cloud misconfigurations, the software supply chain (code repositories, continuous integration, deployment rights) and the vendors that touch your data. Added to that over the past two years are applications built on a language model, where prompt injection and training-data leakage are the two flaws found first.
This shift has a direct consequence for the quote: a scope containing four APIs cannot be tested at the price of a brochure site, even when both sit on the same domain name.
Black box, gray box, or white box: which test for which need?
The three approaches do not prove the same thing and do not cost the same.
- Black box: the tester starts from zero, like an outside attacker, and measures what can be found without knowing anything. It is the only one that tells you what someone with no access actually sees.
- Grey box: it receives an account and sometimes part of the documentation, like a contractor or an employee. We measure what a legitimate user can do beyond what they should.
- White box: the code and architecture are open. You find the flaws that the outside never shows, provided you verify that they are reachable from the Internet.
A black-box test on an application whose architecture nobody knows often produces an impressive and not very useful report. A white-box test without a black-box phase misses configuration errors visible from the outside. The details of the three approaches, with what each one proves and when it is justified, are the subject of a separate article, which will appear here once it has been reviewed.
How often should testing be done?
The answer depends on what is changing in your environment, not on a single schedule. Three situations arise, and they do not call for the same cadence.
- A small business with a website and nothing else: a full test every two years, and a version check every month, is enough in most cases. The surface does not change, so repeating the same engagement every year mainly measures the passage of time.
- A mid-sized company with an internal system and online services: a full test once a year, plus a targeted test after each major change (new application, migration, opening an interface) and component monitoring in between.
- A large organization with multiple systems and providers: tests spread throughout the year, continuous monitoring of versions and scopes, and dedicated tests after each incident.
What no longer holds up, however, is the annual report sitting on a shelf. A published vulnerability is exploited within a few hours to a few days, so a document that describes a state of affairs twelve months old says nothing about what is open today. That is exactly the role of our reports.
What the law actually requires in 2026
Four texts recur in the files, and they must be distinguished because they do not impose the same requirements.
- NIS2, currently being transposed in France: risk management measures and evidence that they work, without imposing the method.
- DORA, for the financial sector: an operational resilience testing plan, including penetration testing for the most critical entities.
- PCI DSS 4.0.1: an internal test and an external test at least annually, and after any significant change.
- GDPR, Article 32: no test is imposed, but measures appropriate to the risk. A test report is the most readable evidence a manager can produce.
The same report serves two audiences: the regulator, who asks for proof that the measures exist, and the insurer, who increasingly asks for a recent status report with no critical flaws before granting coverage. A misread text is costly in both directions: you do not publish a penalty figure you have not verified, and you do not promise that one test is enough to be compliant.
How much does a penetration test cost?
A price without a scope means nothing. Four elements change the bill, in this order: the number of applications and interfaces to cover, the depth requested (a black-box test costs more than a guided test, because it requires more reconnaissance time), the intervention window (at night or outside working hours, so as not to disrupt operations) and what is included after the report.
At our end, the first step is a scoping phase at €499, in writing and by phone. An exchange to understand the context, then a document that sets the scope, the addresses concerned and what is excluded, before any request. This amount is deducted from the final quote if the engagement is entrusted to us. The test itself then ranges from a few thousand euros to several tens of thousands, depending on the complexity of the system and the number of interfaces to cover. To discuss it, the contact form is used for scoping, with no commitment.
The last point is the one people forget to compare. A report that lists vulnerabilities without the remediation steps or the final check leaves the company alone with its flaws. A quote that includes verification after remediation costs more at signing and less in total, because it actually closes what it opened.
What a company must require before paying
Four things are checked before signing, and none of them can be fixed afterward.
- A written scope, with the exact addresses and what is excluded.
- A named method, not a promise.
- A report that ranks findings by business impact rather than by technical score.
- A debriefing meeting where the provider explains how it got in, not just what it found.
One criterion never appears on any quote: the relationship. An external audit is not meant to identify someone to blame internally. A provider that fuels this fear makes the company more fragile, because teams then start hiding what they know. An outside perspective has an advantage that no one internally has: it has no stake in the conclusion, and it sees what habit makes invisible. When the feedback session goes well, the internal team comes away with arguments to obtain the resources it was already asking for.
The report is the real deliverable. A two-hundred-page document that no developer can read is not worth ten pages that say: here is the door, here is the proof, here is the fix, here is how to verify it has been done.
How to verify that the fix has taken place
A mission does not end with the delivery of the report, it ends at the retest. The same path must be replayed, with the same proof, and fail. In practice, the retest covers the critical findings, not the entire mission: replaying eighty findings costs more than the mission itself. A provider that bills this retest separately rarely says so at signing, even though it is the only step that turns a finding into a fix.
This is one of the few points where an external measurement helps: a publicly readable server version indicates whether the announced fix is actually in place.
To go further
Three doors, depending on what you are looking for. For choosing the test, the section above details the three approaches. If the question is what is actually exposed today, our vulnerability catalogue gives the version boundary and the remediation for each entry. If the question is whether your site reveals its version, the guide to checking a website's version gives the method in ten minutes.
FAQ
Is a pentest mandatory in France?
No text requires every company to conduct a penetration test. Some regulated sectors and some clients require it contractually, and NIS2 brings several obligations closer together, but the obligation concerns security, not the method.
How long does a pentest take?
It depends on the scope, not on a fixed package. An isolated web application can be tested in a few days, a complete information system in several weeks. A provider who quotes a price without knowing the scope is selling a package, not an engagement.
What is the difference between a pentest and a compliance audit?
An audit compares your organization against a standard and produces a gap. A pentest tests your system and produces proof of access. Both are useful, and one does not replace the other.
What to do with the report next?
Sort it by business impact, fix what is reachable from the outside first, then replay the paths found to verify. A neatly filed report without verification is an expense, not a measure.
