
By Eliot Rabaud
· Updated
What is Ethical hacking (Pentest)?
Deciphering
Ethical hacking, also known as intrusion audit, penetration testing or pentesting, consists in evaluating the security of a computer system or network using the techniques and tools of hackers, but in a legal and controlled way. An intrusion audit (pentest) is a mandatory phase when you want to secure yours or your client’s data. Whether the pentest is conducted on internal, external, cloud infrastructure, or a web application, the goal remains the same: Simulate attacks to assess and improve the security of an informatic system. The point: find and fix vulnerabilities before an attacker exploits them.
Pentesters adopt the perspective of an attacker, using similar approaches and tools. Consequently, the results of an audit reflects what a potential attacker might discover and exploit.
Unlike the services of a “Blue team” or consulting, which focus on setting up technical and organizational defenses to minimize a successful attack (such as firewall rules, monitoring, an application firewall (WAF), etc.), a pentest takes a more direct and technical approach. Recommendations from a pentest aim to eliminate attack vectors upstream of exploitation by correcting vulnerabilities, through means such as modifying the source code, updating systems, or removing problematic configurations.
Contents
Different Approaches to Pentesting
The diversity of penetration testing approaches (black box, gray box, and white box) offers varied but complementary perspectives on system security. Each method brings a different depth of analysis, ranging from total ignorance of the system (black box) to complete knowledge (white box), with an intermediate level of knowledge (gray box). However, regardless of the chosen approach, the objective remains constant: to detect weaknesses and strengthen security.

Black Box
In a black box type pentest, the tester has no prior knowledge of the targeted system. This approach emulates an external attacker attempting to penetrate the system without internal information. A black box audit can be the ideal approach for a company that has never conducted an audit and wants a first assessment of their external defenses.
Further reading: black-box pentesting, what it finds and what it misses.
Gray Box
The gray box type pentest lies between black box and white box approaches. Here, the pentester has some information about the internal infrastructure, such as credentials, database schemas, but not the source code, for example. This method offers a balance between external exploration and internal knowledge. A gray box pentest is perfect for SAAS applications or a company’s internal infrastructure.
Further reading: grey-box pentesting, which access to provide and why.
White Box
In a white box approach, the tester has complete access to system data, including architectural diagrams, source code, and credentials. The white box approach can be interesting for companies that have already conducted a pentest or are confident in their defenses: it finds problems that are hard to spot without this information. It also closes part of the time gap between a pentester and an attacker, who is not limited to a few days. The pentester moves faster and spends more time on exploitation and on chaining vulnerabilities.
Further reading: white-box pentesting and source code review.
Pentest Methodology:
To ensure the quality and effectiveness of a penetration test, it is important to follow a strict methodology. Several organizations have created guides and methodologies to standardize penetration tests. Although the intrinsic quality of the pentester will of course be decisive in the final results. The use of these guides limits the quality gap between different pentests.
At DungeonData, we follow the steps of the OWASP “testing guide” and use the “Common Vulnerability Scoring System (CVSS)” method to standardize the rating of the vulnerabilities, thus allowing effective prioritization of fixes.
The OWASP Testing Guide is primarily oriented towards web application pentests, but the different steps of the penetration test are extracted from the “Penetration Testing Execution Standard” (PTES) and can apply to all types of penetration tests. These steps guide the pentest from start to finish, both technically and organizationally.
The steps include:
- Pre-engagement Interactions
- Intelligence Gathering
- Threat Modeling
- Vulnerability Analysis
- Exploitation
- Post Exploitation
- Reporting
The seven phases in a real engagement
Pre-engagement interactions. Everything starts on paper. The exact scope, the dates, the IP addresses the tests will come from, the excluded actions (denial of service, changing production data…) and the emergency contacts are written into signed rules of engagement. Without that written authorisation, a penetration test is unauthorised access to a computer system, which is a criminal offence in France (article 323-1 of the French Criminal Code) and in most other countries. This is also when the approach is chosen: black, grey or white box.
Intelligence gathering. The pentester lists what exists before looking for what is weak. On the Internet side: domain names, subdomains (public certificate logs reveal a lot of them), IP addresses, exposed services, credentials already leaked. Internally: machines, network shares, Active Directory accounts and groups. A good share of the critical flaws we find sit on an asset the client had forgotten about.
Threat modelling. With the inventory in hand, the question becomes what is valuable and who would want it. Customer data in a SaaS, the payment flow of an online shop, domain admin accounts in a small company. This phase decides where the testing time goes, and testing time is always limited.
Vulnerability analysis. Scanners do the first pass: outdated versions, default settings, known flaws. They don’t understand business logic. A scanner will never notice that a customer can download another customer’s invoice by changing a number in the URL; a human testing with two accounts sees it within minutes. That is why the manual share of a pentest should stay the larger one.
Exploitation. A suspected vulnerability is not a proven one. Exploitation shows the real impact without breaking anything: reading data that should be off limits, running a command, taking over an account. A critical flaw found at this stage is reported to the client straight away, without waiting for the report.
Post-exploitation. Then comes the question: what next? From the compromised machine, what can be reached, what secrets are lying around, can we move on to the domain or to the cloud? Three medium flaws chained together often make a critical attack path. This phase is what separates a list of vulnerabilities from a real attack scenario.
Reporting. An executive summary readable by management, then one sheet per vulnerability with its proof, score, business impact and a fix suited to your stack. PTES stops here. In practice two more steps follow: the debrief with your teams, then the retest that checks the fixes hold. The details are on our methodology page.
The right standard for each scope
PTES gives the sequence. To know what to test, each kind of target has its own reference.
- Web applications and APIs: the OWASP Web Security Testing Guide (WSTG) describes the tests one by one, from session management to business logic. It pairs with the ASVS (Application Security Verification Standard), whose version 5.0 came out in May 2025: about 350 requirements in 17 chapters, with three levels. The WSTG says how to test, the ASVS says what an application should meet. For a development team, ASVS level 1 is a sensible first target.
- Active Directory and internal networks: ANSSI’s Active Directory recommendations and PTES.
- Cloud: Microsoft’s recommendations and the CIS Benchmarks for Azure and Microsoft 365.
Two broader references also show up in tender documents. NIST SP 800-115, the US technical guide to security testing, dates from 2008 but its structure (planning, discovery, attack, reporting) still holds. And in France, ANSSI’s PASSI framework (version 2.2, published on 29 November 2024) governs qualified audit providers; it distinguishes five activities: architecture audit, configuration audit, source code audit, penetration testing, and organisational and physical audit.
The OWASP Top 10 lists: not just the web
The OWASP Top 10 is an awareness document: it ranks the most common risks, based on testing data and a community survey. People often present it as a testing method, wrongly. A pentest that “covers the Top 10” has covered the minimum, and plenty can go wrong in your application without fitting into those ten boxes. It is still very useful for one thing: giving pentesters, developers and clients a shared vocabulary.
And there isn’t just one. OWASP publishes a Top 10 per kind of technology, and the one that matters is the one matching the audited scope.
Web applications: the Top 10:2025
The latest edition of the web Top 10 replaces the 2021 one:
- A01 Broken Access Control
- A02 Security Misconfiguration
- A03 Software Supply Chain Failures
- A04 Cryptographic Failures
- A05 Injection
- A06 Insecure Design
- A07 Authentication Failures
- A08 Software or Data Integrity Failures
- A09 Security Logging and Alerting Failures
- A10 Mishandling of Exceptional Conditions
Three changes are worth a closer look. Access control stays first, which matches what we see in the field: it is the flaw we find most often in business applications. Supply chain (A03) widens the old “vulnerable components” category to everything used to build and ship software: poisoned dependencies, tampered builds, unverified updates. Finally, SSRF no longer has its own line, it now sits under access control, and its slot goes to a new category on error handling: messages that leak internal data, applications that crash or “fail open” when something unexpected happens.
APIs: the API Security Top 10 2023
An API has no interface to hide its functions behind; everything can be called directly. The API Security Top 10 reflects that:
- API1 Broken Object Level Authorization (BOLA)
- API2 Broken Authentication
- API3 Broken Object Property Level Authorization
- API4 Unrestricted Resource Consumption
- API5 Broken Function Level Authorization
- API6 Unrestricted Access to Sensitive Business Flows
- API7 Server Side Request Forgery
- API8 Security Misconfiguration
- API9 Improper Inventory Management
- API10 Unsafe Consumption of APIs
Three of the top five are about authorisation. The invoice example above is API1. And API9 is a classic: an old API version (/v1/), forgotten but still online, without the protections later added to /v2/.
AI: the LLM Top 10 and the Top 10 for agentic applications
Applications built on a language model (chatbots, assistants, search over internal documents) have their own Top 10 for LLM applications, 2025 version. Prompt injection comes first (LLM01): instructions hidden in a document or web page read by the model make it do something else. The list also covers sensitive information disclosure (LLM02), improper output handling (LLM05), excessive agency (LLM06) and system prompt leakage (LLM07).
On 9 December 2025, OWASP added a Top 10 for agentic applications, for agents that no longer just answer but act: send an email, query a database, edit a file. The first risk (ASI01) is agent goal hijack through content the agent reads, followed by tool misuse (ASI02) and identity and privilege abuse (ASI03). In practice the question becomes: if someone takes control of the agent, what can they do with its access? If the answer is “drop the production database”, the problem lies first in the permissions given to the agent.
Delivery pipeline: the CI/CD Top 10
The Top 10 CI/CD Security Risks looks at the chain that builds and deploys your code: a developer who can push to production without review (CICD-SEC-1), a dependency swapped for a poisoned version (CICD-SEC-3), a pipeline that runs code from a mere pull request (CICD-SEC-4), secrets left in build logs (CICD-SEC-6). It is the tooling side of A03 in the web Top 10.
One thing to remember: before an audit, ask which Top 10 matches your scope. A SaaS product with an API and an AI assistant covers three of them.
Scoring vulnerabilities with CVSS
Every vulnerability in our reports gets a CVSS (Common Vulnerability Scoring System) score from 0 to 10, computed from fixed criteria: reachable from the Internet or not, account needed or not, attack complexity, impact on confidentiality, integrity and availability. The score reads as four levels: low, medium, high, critical. We use version 3.1, the most widely supported by tools. Version 4.0, published by FIRST in November 2023, refines the calculation and is starting to replace it.
CVSS measures how severe a flaw is, not how likely it is to be exploited. Two flaws scored 9.8 don’t necessarily carry the same urgency. Other indicators exist to sort patches at scale; we cover them in the article on EPSS and patch prioritisation. In a pentest report, the score comes with the concrete business impact, which often weighs more than the number.
For more information on the CVSS and the OWASP Testing Guide, you can consult the following links:
- CVSS: https://fr.wikipedia.org/wiki/Common_Vulnerability_Scoring_System
- PTES: http://www.pentest-standard.org/index.php/PTES_Technical_Guidelines
- OWASP testing guide: https://owasp.org/www-project-web-security-testing-guide/
Pentest is not an option
By simulating attacks, pentesters show what an attacker could do to your systems, and so what needs fixing first.
Systems change all the time: new versions, new access, new suppliers. A pentest from three years ago says little about your exposure today. Running one regularly, at least once a year and after every major change, lets you fix flaws before someone uses them, and keeps your customers’ trust.
Related articles
- Pentest

White-box pentesting: testing with the source code in hand
White-box pentesting: source code, architecture and accounts handed to the tester. What this approach finds that others miss, what it costs, and when to choose it.
- Pentest

Grey-box pentesting: which access to provide, and why
Grey-box pentesting: test accounts, API documentation, a domain user account. Which access to give the pentester, what it lets them test, and where it stops.
- Pentest

Black-box pentesting: what it finds and what it misses
Black-box pentesting: the tester starts from your domain name, with no account and no code. How it runs, what it finds, its blind spots and when it is enough.
