- A vulnerability scan identifies possible flaws; a penetration test checks, within an authorised scope, whether and how they can be exploited
- The right first test depends on your attack surface: for many SaaS and cloud-first businesses a web application, API or cloud identity test is more useful than an external network test
- ISO 27001 and NIS2 do not set a universal annual pentest rule; PCI DSS Requirement 11.4 and DORA set specific testing duties for the entities in scope
- Buy on scope, methodology, tester competence, rules of engagement and reporting quality, not on day rate alone
- Every finding needs an owner, a treatment decision, a deadline and evidence that the fix works
What penetration testing actually is
A penetration test is an authorised, objective-driven security assessment in which qualified testers use controlled attack techniques against your systems, applications or people. The tester receives a defined scope and agreed objectives, then tries to identify and validate vulnerabilities, documents each step and reports what was found and why it matters to the business.
A tester does not stop at listing a weakness. Within the rules of engagement, they may try to exploit it, escalate privileges, move between systems and reach the assets named in the objectives. The test ends when the agreed scope and time are used, or earlier if a critical finding needs to be reported to you straight away.
The difference between a pen test and a vulnerability scan
A vulnerability scanner checks systems against known vulnerabilities, missing patches and configuration errors, and produces a list of possible issues ranked by severity. Some scanners also perform limited, controlled validation. What a scan does not do is pursue an objective: it does not show whether an attacker could combine several findings to reach something that matters to you.
A penetration test is an objective-driven, authorised assessment that combines manual and automated techniques to check exploitability and consequences within scope. Penetration testing complements vulnerability management by assessing, within authorised scope, how technical and configuration weaknesses can be exploited individually or chained. A medium-severity misconfiguration plus a reused password, for example, can add up to access to a finance system.
Continuous scanning gives you an ongoing view of known vulnerabilities across your estate. Periodic penetration testing shows what a skilled tester could achieve with them. One does not replace the other; see our vulnerability management guide for how scanning, patching and testing fit together.
These approaches differ in how much information the tester receives, not in how complete the result is. Black-box testing starts from public information only and shows what an outsider could find. Grey-box testing provides partial information, such as user credentials, and suits scenarios like a compromised employee account. White-box testing provides architecture documents, source code or administrative access and allows a deeper review of specific components. Choose by objective. None guarantees full coverage, and combining approaches, such as an authenticated web application test with a short external reconnaissance phase, often gives a better picture.
Types of penetration test
An external network test is not automatically the right first test. Choose by attack surface and main risks. For a SaaS or cloud-first business whose value sits in a web platform and Microsoft 365 or AWS accounts, a web application, API or cloud configuration and identity assessment will usually tell you more than a scan of a handful of public IP addresses.
Web application and API testing
Application tests commonly follow the OWASP Web Security Testing Guide. They check for injection, authentication failures, broken access control such as insecure direct object references (IDOR), security misconfiguration and cross-site scripting. Business logic flaws deserve special attention: a checkout that accepts a manipulated discount, or an account page that shows another customer's data when an ID in the URL changes. Scanners rarely find these; they need a tester who understands how the application is meant to behave.
Social engineering and physical testing
Social engineering tests, usually phishing or targeted spear-phishing simulations, measure whether staff can be persuaded to reveal credentials or approve access, and how quickly the security team notices. Physical tests check whether someone could enter premises, server rooms or unattended workstations. Both involve people, so they need explicit authorisation, careful scoping and agreement with HR and legal before they start.
When your business needs one
Good moments to test are before launching a new application or service, after a significant infrastructure change or cloud migration, after a security incident to confirm the attacker's route is closed, and after a merger or acquisition that brings new systems into your environment. Beyond that, frequency should follow your risk assessment and any contractual or regulatory duties.
ISO 27001. The standard has no universal annual pentest requirement. Annex A controls, such as A.8.8 (management of technical vulnerabilities) and A.5.36 (compliance with policies, rules and standards for information security), are selected through risk treatment and recorded in your Statement of Applicability, and you must show that controls are effective. A pentest is often a sensible way to do that for exposed or critical systems, depending on your risks, selected controls and contractual requirements. See our ISO 27001 guide.
NIS2. Article 21(2)(f) requires policies and procedures to assess the effectiveness of cybersecurity risk-management measures. Penetration tests can contribute, but the Directive sets no universal annual pentest. Check the national transposing law and, for digital infrastructure and digital service providers, Implementing Regulation (EU) 2024/2690. Our NIS2 guide explains scope.
PCI DSS. Requirement 11.4 applies to entities and environments in scope. It covers a defined methodology, internal and external penetration testing at least every 12 months and after significant changes, application-layer testing, correction and retesting of exploitable findings, and testing of segmentation controls. How you validate depends on your cardholder data environment, your SAQ or assessment type and what your acquirer or card brand requires, so not every merchant runs the same programme. See our PCI DSS guide.
DORA. Financial entities need a risk-based, proportionate digital operational resilience testing programme, with appropriate tests at least yearly on ICT systems supporting critical or important functions (Article 24). Penetration tests are one of the possible tests (Article 25). Threat-led penetration testing (TLPT, Article 26) is a distinct, intelligence-led exercise on live production systems, generally every three years, and only for entities identified by their competent authority. A conventional pentest assesses defined systems against agreed scenarios; TLPT emulates real threat actors across critical functions under a formal supervisory framework. See our DORA gap analysis guide.
UK schemes. Cyber Essentials Plus includes hands-on technical verification by an assessor, but that is not a penetration test. For UK public sector and critical national infrastructure work, buyers often ask for providers in the NCSC's CHECK scheme.
A penetration test shows what the testers could identify and validate during the engagement, within its scope and limits. It does not show the full extent of what a real attacker could do.
What happens during a penetration test
Most professional tests follow a similar sequence. Knowing it helps you prepare your team and set realistic expectations.
Scoping and rules of engagement. You agree what is in and out of scope, the testing window, and whether your IT team is told (overt) or not (covert, which also tests detection). The rules of engagement should settle at least the following:
- Who authorises the test, for which systems and which legal entities
- Permissions from third parties and cloud providers where their terms require it
- Permitted and prohibited techniques, such as denial of service or destructive payloads
- Whether social engineering or physical access is in scope
- How sensitive data found during the test is handled
- Emergency stop conditions and escalation contacts on both sides
- How evidence is protected and how the report is delivered securely
Reconnaissance. The tester gathers information before touching the targets: DNS records, certificate transparency logs, public information about the organisation and leaked credentials. This phase often turns up forgotten subdomains or exposed test environments.
Exploitation. The tester tries to validate vulnerabilities in line with the agreed objectives and rules. On critical or fragile systems, proof may be limited to non-destructive techniques. Every step is documented so findings can be reproduced.
Post-exploitation. Where permitted, the tester checks how far initial access leads: from a standard account to domain administrator, from one network segment to another, or to backups and sensitive data stores, always within authorised limits. This shows the potential reach of a compromise.
Reporting. A good report has an executive summary for non-technical readers, reproducible findings with evidence, a risk rating for each finding, the business impact in plain terms and specific remediation advice.
How to buy a penetration test
Quotes are hard to compare on price alone. Use these criteria to compare providers and proposals.
What to do with the results
The report is the start of the more important half of the work. A report that sits unread is documented evidence of known vulnerabilities that were not addressed.
Triage critical findings immediately and apply containment or mitigation where needed, such as disabling an exposed interface or restricting access, while the permanent fix is prepared. Set the final deadline for each fix by exploitability, exposure, impact and dependencies. Each finding needs a designated owner, a treatment decision (fix, mitigate, accept or transfer), a deadline and verification evidence, tracked in your risk register.
A finding marked "fixed" without verification is a risk that has not yet been treated or verified. A retest by the original provider is one way to verify, but not the only acceptable one: an internal test, a targeted rescan or configuration evidence can be enough, depending on the finding. Under PCI DSS, exploitable vulnerabilities found in the pentest must be corrected and the correction verified by repeating the testing.
Finally, look for patterns. Several findings about missing patches point to a process gap in vulnerability management; vulnerabilities on systems nobody knew were online point to an asset inventory gap. A cybersecurity roadmap helps turn findings into a prioritised programme with owners and milestones.
CREST accredits penetration testing providers and maintains a searchable directory of member companies. The NCSC runs the CHECK scheme and publishes guidance on penetration testing for organisations commissioning one.