Cyber Security
What Is Penetration Testing? A Plain-English Guide

A penetration test is a controlled, authorized attack on your own systems, carried out by people whose job is to think like criminals and then write down everything they did. It answers one question that no scanner, audit checklist or vendor dashboard can: if someone motivated came after us today, how far would they actually get?
That question matters more every year. In the breaches analyzed for Verizon's 2026 Data Breach Investigations Report, 31% started with an exploited software vulnerability, making it the most common way in, ahead of stolen passwords. Mandiant's M-Trends 2026 puts exploits at the top of its initial infection vectors for the sixth year running, at 32% of intrusions. Attackers are testing your defences either way. A penetration test means the first people to do it are on your side.
What a penetration test actually is
The U.S. National Institute of Standards and Technology, whose 2008 testing guide remains a widely used reference, defines it this way:
"Penetration testing is security testing in which assessors mimic real-world attacks to identify methods for circumventing the security features of an application, system, or network." (NIST SP 800-115)
The important part comes in the next line of that definition. Most penetration tests involve looking for combinations of vulnerabilities that together "gain more access than could be achieved through a single vulnerability." That is the whole point. A scanner will report three medium-risk issues on three different machines. A tester will chain them into a route from a receptionist's laptop to your finance server, and show you the screenshot.
The Canadian Centre for Cyber Security's control catalogue says much the same: a penetration test "attempts to duplicate the actions of adversaries", and all parties agree the rules of engagement before any testing starts. It also recommends that the tester be independent of the team that built and runs the systems being tested. More on why that matters below.
If you are trying to decide between a penetration test and a vulnerability assessment, we cover that in detail in Vulnerability Assessment vs. Penetration Testing: What's the Difference? The short version: a vulnerability assessment tells you which doors might be unlocked. A penetration test walks through them. More precisely, a vulnerability assessment focuses on identifying, validating and prioritizing security weaknesses, while a penetration test goes further by attempting to exploit selected weaknesses within an agreed scope, to show how they could affect the business.

The main types of penetration test
"We'd like a pen test" is not a scope. The first conversation with any good tester is about which of these you need, because each one answers a different question and finds different problems.
- External network. Testing everything you expose to the internet: firewalls, VPN gateways, remote desktop, email servers and anything else with a public IP address. This is the attacker's view from outside, and it is where exploited VPN and firewall flaws show up.
- Internal network. Starting from a foothold inside your network, usually a standard employee laptop, and seeing how far the tester can get. In Windows environments, testers often examine Active Directory permissions, credential exposure and opportunities for privilege escalation. The goal is to show how far an attacker could move from an ordinary user account, within the agreed scope.
- Web application. Testing a specific application, such as a client portal or booking system, for broken access control, injection flaws, weak session handling and business-logic mistakes. Tests are usually run logged in as several different user roles.
- API. The same idea applied to the interfaces behind your apps and integrations, where one customer can often read another customer's data by changing a single number in a request.
- Cloud. Assessing the security of your Microsoft 365, Azure, AWS or Google Cloud environment, including identity permissions, exposed storage, application access, authentication weaknesses and insecure configurations. Testing must stay within your own authorized environment and the cloud provider's permitted-testing rules.
- Wireless. Testing Wi-Fi configuration, guest network separation and rogue access points, usually on site.
- Social engineering. Phishing, phone-based pretexting and help-desk impersonation, measuring how your people and processes hold up rather than your technology.
- Physical. Tailgating, badge cloning and finding unlocked network ports in meeting rooms. Less common for small businesses, but valuable if you hold sensitive data on site.
Testers also describe tests by how much they are told in advance. In a black-box test they start with nothing but your company name. In a grey-box test they get a user account and a scope document. In a white-box test they get documentation, architecture diagrams and sometimes source code.
Black-box testing can be valuable when the goal is to evaluate what an outside attacker could discover and exploit without prior knowledge. But for many smaller businesses, grey-box testing offers a better balance of cost and depth. A real attacker can spend weeks quietly learning about your business for free, while a tester billing by the day would spend those same hours rediscovering it. Giving testers limited information and appropriate user accounts lets them spend more time on real weaknesses, which matters most in web applications, where access-control flaws often only appear when you test with more than one user role. White-box testing goes deeper still for complex applications.
How a penetration test works, step by step
Methodologies differ in how they number the steps. NIST SP 800-115 uses four phases, and the Penetration Testing Execution Standard uses seven. The work itself looks the same in practice.

- Scoping and rules of engagement. Everyone agrees in writing what is in scope, when testing happens, what is off limits (such as anything that could cause an outage or alter production data) and who to call if something goes wrong. Written authorization must come from someone who actually owns the systems. If you use cloud or hosted services, their testing policies matter too, because you cannot authorize an attack on someone else's infrastructure. NIST is blunt about this phase: "No actual testing occurs."
- Reconnaissance. Gathering what is publicly knowable: domains and forgotten subdomains, certificate records, staff names and email formats from LinkedIn, credentials leaked in old breaches, and code accidentally published online.
- Scanning and enumeration. Mapping live systems, open services and software versions, and running vulnerability scans. In a penetration test, a scanner result is only a lead until it has been proven.
- Exploitation. Proving the weakness is real by using it. NIST describes this as exploiting the vulnerability "to confirm its existence." Skilled testers chain small findings together and stop at the agreed point before anything they do could cause damage.
- Post-exploitation. Escalating privileges, harvesting credentials and moving sideways through the network to show what a real attacker could reach. Testers usually prove access to sensitive data with a screenshot or a planted file rather than by taking the data.
- Reporting. A written report and a live debrief. More on what good looks like below.
- Remediation retest. Once you have fixed the findings, the tester checks that the fixes actually worked. The retest letter is often the document your auditor or insurer actually wants.
What testers actually find
Penetration tests regularly uncover weaknesses organizations did not know they had. In Cobalt's State of Pentesting Report 2025, the company reports that every pentest it conducted uncovered at least one reportable finding, with a median of six findings and more than a quarter of tests turning up over ten. Those figures describe Cobalt's own testing data, not every test everywhere, but the pattern is consistent.
What testers find is also rarely exotic. On internal networks, the most common findings are default settings that were never switched off. Vonahi Security, which sells an automated testing platform, reported findings from more than 10,000 of its automated internal network tests in a contributed article. The most common were name-resolution spoofing weaknesses: mDNS spoofing at 78.2% occurrence, NBNS at 73.3% and LLMNR at 65.5%. These name-resolution protocols are switched on by default on many networks, and they let an attacker on your network impersonate other machines and capture login exchanges, which can then be cracked or relayed to gain access. They are often an early step towards wider control of the network.

Web applications tell a similar story. Broken access control remains the number-one category in OWASP's Top 10:2025, the industry's reference list of web application risks. These weaknesses let users reach information or functions they should not be able to. Automated scanners often struggle to identify them, because they cannot tell whether a particular user should be allowed to see a particular record. A human tester logged in as two different customers can.
What a good penetration test report contains
The report is the product you are paying for, so ask to see a redacted sample before you sign anything. A good one includes:
- An executive summary written for owners and leadership, explaining in plain language what was tested, what was found and what it means for the business.
- Attack narratives that walk through how individual findings were chained together, step by step, so your team can see the actual path an attacker would take.
- Risk-rated findings with evidence, the systems affected and clear steps to reproduce each issue.
- Specific remediation guidance your IT team or provider can act on, not generic advice to "apply patches."
- What was tested and what was not, so nobody mistakes a narrow test for a clean bill of health.
If the sample looks like a scanner export with a logo on the front, keep looking.
Finding problems is the easy part
The hardest part of a penetration test is what happens after the report. In Cobalt's State of Pentesting Report 2025, only 48% of identified findings were remediated, while 69% of serious findings were eventually resolved. In 2024, serious findings took a median of 37 days to fix (50 days across Cobalt's full dataset), even though most organizations set themselves a two-week target. Serious internal network findings took longest of all, at a median of 63 days, although Cobalt notes that most of the differences between test types are not statistically significant.

There is a useful detail in that data for smaller businesses: organizations with under 100 employees fix serious findings faster than those with over 5,000, with a median of 27 days compared with 61. Fewer approval layers and fewer systems mean a small business can act on a report quickly. The organizations that get real value from testing are the ones that budget for remediation and the remediation retest, not just the test.
How often should you test?
For many organizations, annual penetration testing is a useful baseline, with additional testing after significant changes to systems or applications. Once identified vulnerabilities have been fixed, a remediation retest confirms the corrections are effective.
For organizations subject to PCI DSS penetration-testing requirements, internal and external tests are required at least every 12 months and after significant changes to the cardholder data environment. Not every small business that accepts credit cards has to commission its own penetration test, though: what applies depends on how payments are processed and which systems fall within scope, so check with your payment processor or acquirer. The CIS Critical Security Controls set an annual baseline for external testing, and even where testing is not mandatory, an annual test gives you a useful security baseline.
Significant change is the part people forget. A new firewall, a move to Microsoft 365, a new client portal or an office relocation can each undo the results of last year's test, and may call for a new or targeted test of what changed. A penetration test is also only a snapshot of your security at a particular moment. Attackers move quickly: Mandiant's M-Trends 2026 estimates the mean time to exploit vulnerabilities has dropped to about minus seven days, which it says means exploitation is routinely happening before a patch is even released. A clean report from the spring tells you little about a VPN flaw announced in the autumn. That is why ongoing vulnerability scanning and timely patching matter between tests.
How to get a quote for a penetration test
Pricing depends almost entirely on scope: how many IP addresses, how many applications, how many user roles and how many locations are being tested, and whether a remediation retest is included. Comparing quotes is only meaningful when the scope is identical, so make sure every provider is pricing the same environment.
The quickest way to a useful number is a short scoping conversation. Contact us for a quote and we will help you define the right scope for your business, then price it.
How to choose a penetration testing provider
- Ask who will actually do the testing. Firms often sell with a senior name and staff the work with juniors. Ask for the testers' names and qualifications. Certifications such as OSCP, PNPT, OSCE3 and GIAC GPEN can indicate relevant expertise, and several include practical, hands-on assessments. Weigh them alongside real-world experience, the tester's methodology and the quality of their previous reports.
- Check company accreditation. CREST accredits testing companies on their processes, personnel security and data handling, not just individual people. Confirm the accreditation covers penetration testing specifically.
- Ask for a sample report. See above. It is the single best predictor of what you will receive.
- Confirm the methodology. Ask which recognized standard they follow, such as NIST SP 800-115, PTES or the OWASP Web Security Testing Guide. PCI DSS requires one.
- Insist on written authorization and rules of engagement before any testing starts, including emergency contacts and stop conditions.
- Check their insurance. The tester should carry professional liability and cyber insurance, and provide a certificate.
- Ask how they handle your data. Testers will collect credentials, password hashes and screenshots of sensitive records. Ask where that evidence is stored, for how long, and in which country.
- Make sure a remediation retest is included, and within what window. Reputable firms typically include one within 30 to 90 days.
- Insist on independence. If the company that manages your firewall also tests it, ask how the testing team is separated from the team that built the configuration. Nobody should grade their own homework.
What penetration testing will not do
A penetration test is not a guarantee. It covers an agreed scope over an agreed number of days, and it shows you how your defences held up on those days. It does not replace patching, monitoring, backups or staff training. Testing live systems also carries a small risk of disruption, which NIST says "can never be fully eliminated". That is why careful scoping, tested backups and a tester who asks about your fragile legacy systems before touching them all matter.
What a penetration test will do is replace assumptions with evidence. Instead of "we think we're fine," you get a specific, prioritized list of the routes an attacker would take, and proof once they are closed.
Sources
- Verizon, 2026 Data Breach Investigations Report, 2026.
- Mandiant (Google Cloud), M-Trends 2026, 2026.
- NIST, SP 800-115: Technical Guide to Information Security Testing and Assessment, September 2008 (current edition).
- Canadian Centre for Cyber Security, ITSG-33 Annex 3A: Security Control Catalogue, CA-8 Penetration Testing.
- Cobalt, State of Pentesting Report 2025, 2025.
- Vonahi Security (contributed article in The Hacker News), 10 Critical Network Pentest Findings IT Teams Overlook, March 2025.
- OWASP, Top 10:2025, A01 Broken Access Control, 2025.
- Center for Internet Security, CIS Control 18: Penetration Testing.
- PCI Security Standards Council, PCI DSS, v4.0.1, Requirement 11.4, 2024.
- Penetration Testing Execution Standard, PTES Technical Guidelines.
Keep Reading
Related articles

Windows 10 End of Life: What It Actually Means for Your Business
Windows 10 support ended in 2025 and ESU Year 1 ended October 13, 2026. What is still patched, what Year 2 costs, and your options for remaining PCs.
Read Article
Vulnerability Assessment vs. Penetration Testing: What's the Difference?
A vulnerability assessment finds known weaknesses. A pen test proves how far an attacker could get. How they differ, how each is priced, and why you need both.
Read ArticleNext step
Ready for a clear-eyed view of your security posture?
Talk to Ark Cyber about a cyber security consultation built around your business.
