Join our Newsletter — 33% off our NHI Course
Home› Guides› Internal vs External Penetration Testing: Which Test Answers…
Guide Offensive Security & Testing

Internal vs External Penetration Testing: Which Test Answers Which Risk Question

PDF for newsletter subscribers · free to read online

← All guides
By NHI Mgmt Group, edited by Lalit Choda Updated 14 November 2026 18 min read
On this page

Organisations often commission "a pen test" without deciding whether the risk they want measured sits at the internet edge or inside the network, and then read a clean external report as proof that the whole estate is safe. An external penetration test asks what an unauthenticated attacker on the internet can reach and exploit. An internal penetration test asks what happens once someone is already inside: a phished laptop, a rogue device on a meeting room port, a contractor account or a compromised server. This guide is a scoping decision guide. It sets out the threat model behind each test, how to scope each so the report answers a real risk question, the identity attack paths each one exposes, how often to run them, and how to read each findings report. It is for security leads, risk owners and procurement teams buying or planning tests. For method detail, see the Internal Network Penetration Testing Guide and the External Network Penetration Testing Guide.

Key takeaways

  • Each test answers a different question. External testing measures exposure to an attacker with nothing; internal testing measures how far an attacker with a foothold can go and how quickly you notice.
  • Most real intrusions use both paths in sequence. Stolen credentials on a remote access portal (external) followed by lateral movement and privilege escalation (internal) is the common pattern, so a single test type leaves half the chain unmeasured.
  • Identity dominates both reports. External tests find password-only portals and exposed login surfaces; internal tests find Kerberoastable service accounts, credential reuse and certificate template abuse.
  • Frequency is driven by change, not the calendar alone. Annual testing is the floor in PCI DSS and CIS Controls; significant change to the perimeter or the directory should trigger a new test of the affected side.
  • Read the two reports differently. An external report is mostly a list of exposures to close; an internal report is a story of attack paths, and its value is in the chains and the detection gaps, not the count of findings.

Short answer

Choose an external test when the question is "can someone on the internet get in?" It covers internet-facing hosts, remote access services (VPN, virtual desktop and remote desktop gateways), web applications, email and identity provider login surfaces and cloud endpoints. Choose an internal test when the question is "if someone gets in, how bad does it get?" It starts from a position inside the network or tenant and tests segmentation, Active Directory, privilege escalation, lateral movement and access to crown-jewel systems. If you can fund only one this year, use your threat model: a large or fast-changing perimeter with remote access points to external first; a flat network, a large Active Directory or a history of phishing compromise points to internal first. Most organisations past the smallest scale need both, at least annually.

Side-by-side comparison

DimensionExternal penetration testInternal penetration test
Risk questionCan an attacker with no access get in?How far can an attacker with a foothold get, and do we detect it?
Starting positionInternet, no credentials (sometimes a low-privilege customer or partner account)Network port, VPN connection, standard user laptop, domain user account or compromised server
Typical targetsPublic IP ranges, DNS names, VPN and remote access gateways, web apps, mail, identity provider login pages, cloud servicesActive Directory, file shares, management interfaces, databases, segmentation boundaries, backup and hypervisor platforms
Main ATT&CK tactics exercisedReconnaissance and initial access: Active Scanning (T1595), Exploit Public-Facing Application (T1190), External Remote Services (T1133)Credential access, privilege escalation and lateral movement: Kerberoasting (T1558.003), OS Credential Dumping (T1003), Remote Services (T1021)
Identity findingsPassword-only portals, spraying exposure, user enumeration, leaked credential reuseService accounts with weak passwords, excessive group membership, credential reuse, certificate template abuse, local admin sprawl
Detection valueLimited: internet noise hides most test trafficHigh: shows whether EDR, SIEM and identity threat detection see real attacker behaviour
Operational riskLow to moderate; cloud provider rules applyModerate; production directory and fragile systems need safeguards
Main deliverableList of exposures, rankedAttack path narrative from foothold to objective, plus detection gaps
Typical driverPCI DSS 11.4.3, CIS Safeguard 18.2, customer due diligencePCI DSS 11.4.2, CIS Safeguard 18.5, ransomware resilience

The PCI Security Standards Council's penetration testing supplement defines the external test as covering the exposed perimeter and critical systems reachable from public networks, including remote access, and the internal test as covering the perimeter and critical systems from the internal network's perspective. It also requires both to include application-layer and network-layer work.

External testing in depth

Threat model: an opportunistic or targeted attacker who knows only what is public: your domains, IP ranges registered to you, certificate transparency entries, staff names and email formats. The attacker's goals are an initial foothold, data exposed directly on the internet, or credentials that work on a remote access service.

Architecture in words: the tester sits on the internet and works inward. They enumerate what answers on your address space and domains, fingerprint the services, and test each for misconfiguration, known vulnerabilities and weak authentication. For cloud hosted assets the test covers what is reachable through public endpoints, under the provider's rules. AWS, for example, allows customer testing of listed services such as EC2, RDS, Lambda and API Gateway without prior approval, but prohibits denial of service, DNS zone walking through Route 53 and subdomain takeover testing, and requires a simulated events form for red team or phishing exercises. Microsoft's cloud rules of engagement prohibit using credentials you do not own and post-exploitation actions such as pivoting against Microsoft infrastructure.

What it usually finds:

  • Forgotten assets: test, staging or acquired-company hosts that are not in the asset register.
  • Remote access weaknesses: VPN, virtual desktop or remote desktop gateways accepting a password alone, or running versions with published vulnerabilities.
  • Login surface issues: user enumeration through differing error messages, no lockout or rate limit, legacy authentication still enabled.
  • Exposed management interfaces: administration consoles, database ports and storage endpoints reachable from anywhere.

Failure modes of the test itself: a scope built from the asset register only (the unknown assets are the risky ones), IP addresses that are not yours (shared hosting, previous owners), and a time box too short to go past automated scanning. NCSC notes that closed box tests may miss vulnerabilities within the time allotted.

The Change Healthcare intrusion is the textbook external lesson: attackers logged into a remote access portal with a stolen password because the portal did not require multi-factor authentication. See the Change Healthcare breach. Checking every remote access endpoint for password-only login catches that exposure.

Internal testing in depth

Threat model: an attacker who already has something. That may be a phished workstation, a stolen VPN session, a malicious insider, a compromised supplier connection or a device plugged into an open port. The goal is a defined objective, such as domain control, the payment environment or the backup platform, and to see whether defenders notice.

Architecture in words: the tester starts at an agreed foothold and works outward and upward. They map the reachable network, query the directory as an ordinary user, look for credentials in shares, scripts and memory, test name resolution and relay weaknesses, and try to escalate and move laterally. In a hybrid estate it should also follow synchronised accounts into the cloud tenant.

Choosing the foothold is the most important scoping decision. Options, from least to most realistic:

  1. Network port only: tests network access control and what an unauthenticated device can see.
  2. Standard domain user credentials: the "assumed breach" start, which skips initial access and spends the budget on escalation. See the Assumed Breach Testing Guide.
  3. Corporate build laptop: tests endpoint hardening, EDR and local privilege escalation as well as the network.
  4. Compromised server or supplier connection: tests segmentation from a high-value or third-party foothold.

What it usually finds: service accounts with weak passwords and service principal names that make them crackable offline (Kerberoasting, T1558.003); name resolution poisoning and relay exposure (T1557.001); plaintext credentials in scripts, group policy preferences and file shares; the same local administrator password on many machines; certificate templates that allow a user to request a certificate for another identity (T1649); and flat networks where a user subnet can reach management interfaces and backups.

Failure modes of the test itself: excluding the directory because it is "too sensitive", which removes the main attack surface; starting from a privileged build that real users do not have; and EDR or SOC staff being told in advance, which wastes the detection measurement unless that is the deliberate design.

The 2022 Uber intrusion shows why the internal half matters: after a contractor approved a repeated MFA prompt, the attacker moved from one session to other employee accounts and elevated permissions in internal tools. See the Uber breach.

Identity attack paths: what each test exposes

Identity is where the two tests meet. External testing measures whether a credential is enough to get in; internal testing measures what that credential, and the credentials it can harvest, unlock. Ask the provider to cover both human and non-human identities.

Identity pathSeen in external testSeen in internal testControl that closes it
Password spraying against SSO or VPN (T1110.003)Yes: lockout, rate limit and MFA enforcement on each login surfaceSometimes: against internal services that bypass SSOPhishing-resistant MFA on every external entry point; block legacy authentication
Valid account reuse (T1078)Yes: leaked credentials tried on portals (only with written consent)Yes: one account working on many systemsUnique credentials, conditional access, credential leak monitoring
Cloud account takeover (T1078.004)Yes: exposed tenant endpoints and token handlingYes: synchronised on-premises admins holding cloud rolesCloud-only privileged accounts; separate admin identities
Service account cracking (T1558.003)NoYesManaged service accounts, long random passwords, AES-only Kerberos
Credential dumping (T1003)NoYesCredential Guard style protections, tiered admin, no domain admin logons on workstations
Certificate template abuse (T1649)NoYesAudit and restrict certificate templates and enrolment rights
Secrets in code, scripts and sharesYes, if repositories or storage are publicYesSecrets manager, scanning, rotation of anything found

Testers are given credentials for internal and authenticated tests, and those are themselves non-human or temporary identities. Issue them as named, time-bound accounts, monitor them, and disable them at the end. The Penetration Testing Guide covers this in its section on credentials given to testers. For the service account fix, see the Group Managed Service Accounts (gMSA) Guide.

Scoping each test so the report answers your question

NCSC guidance says scoping should involve risk owners, technical staff who know the systems and the testing team, and produce a plan stating the technical boundaries, test types and effort in days. It also advises sharing your own vulnerability assessment so testers can judge its accuracy. Write the risk question at the top of the statement of work, then derive the scope from it.

Scoping itemExternal testInternal test
Risk question"Which internet-reachable assets could give an attacker access or data?""From a phished user, can an attacker reach [objective] and would we see it?"
Scope sourceAsset register plus attack surface discovery, confirmed against IP ownershipNetwork diagrams, directory structure, crown-jewel list, segmentation design
Knowledge givenClosed box for realism or open box for coverage; state whichUsually grey box: a user account and network access
ExclusionsThird-party hosted assets without their permission; denial of serviceFragile systems (OT, medical, legacy) unless safeguards are agreed
Detection stanceUsually announced to the SOCDecide: announced (coverage) or unannounced (detection measurement)
Success criteriaAll assets enumerated, every login surface tested for authentication strengthObjective reached or not, path documented, time to detection recorded

For segmentation, the PCI supplement says each unique segmentation method should be tested, and that if an out-of-scope network can reach the protected environment, the entity must fix the controls or test the whole network layer. That is an internal test deliverable. Use the Penetration Test Scoping Checklist to build the full statement of work.

Choosing: decision criteria

  • Large or changing perimeter: many internet-facing services, recent acquisitions, cloud sprawl or several remote access products point to external first, ideally backed by continuous attack surface monitoring between tests.
  • Large directory or flat network: thousands of accounts, legacy domains, or user subnets that can reach servers point to internal first.
  • Recent phishing or infostealer incidents: assume the foothold is achievable and fund an internal, assumed breach test.
  • Customer or regulator demand: card payment environments need both under PCI DSS 11.4.2 and 11.4.3; many customer questionnaires ask for an external test summary.
  • Small organisation: with a handful of cloud services and no on-premises directory, an external test plus a cloud configuration review of the identity tenant may cover more risk than a classic internal network test. CIS Controls place external tests (18.2) at Implementation Group 2 and internal tests (18.5) at Implementation Group 3, which reflects that order.
  • Mature detection programme: run internal tests unannounced, or move to red teaming. See the Red Teaming Guide.

Frequency and triggers

PCI DSS requires both tests at least annually and after significant changes, and the PCI supplement notes the standard does not define significant change, so the entity decides based on its risk assessment. NCSC observes that a year or more often passes between tests, and recommends routine security testing in between. A practical pattern:

  1. Annual baseline: one external and one internal test, timed apart so remediation from one informs the other.
  2. Change triggers for external: a new remote access product, a new internet-facing application, a merger, a move of services to a new cloud provider or identity provider.
  3. Change triggers for internal: a directory migration or consolidation, a new segmentation design, a major EDR or SIEM change, a data centre move.
  4. Retest: after remediation of high findings, a focused retest of those findings only. PCI DSS 11.4.4 requires exploitable findings to be corrected and testing repeated.
  5. Between tests: vulnerability scanning and attack surface management for the external side; directory attack path analysis and identity posture checks for the internal side.

Worked scenario: one budget, two questions

A 2,500-person logistics firm has budget for about 20 tester days. It runs a hybrid directory synchronised to a cloud identity provider, two remote access products (one legacy), around 60 internet-facing services and a flat office network. Last year an infostealer took a manager's browser credentials.

  • External, 8 days: discovery across all registered domains and ranges, every login surface tested for MFA enforcement and legacy authentication, the legacy remote access product tested in depth, and the top five internet applications at application layer.
  • Internal, 12 days: assumed breach from a standard user on a corporate laptop, objective "reach the warehouse management database and the backup console", including the path from on-premises admin groups into cloud roles, run unannounced to the outsourced SOC with a named contact in the security team for safety.

The external report finds the legacy gateway accepts passwords alone and a staging server exposes an admin console. The internal report shows a Kerberoastable service account with a 2019 password leading to server admin rights, and the backup console reachable from the user subnet; the SOC raised one alert, four hours after the credential access. Together the reports show a full chain: a sprayed or stolen password on the legacy gateway, then the internal path to backups. That chain, not either list alone, is what goes to the board.

Reading each findings report

NCSC expects a report to include the issues found, the testers' risk rating for each, a resolution method, and an opinion on how accurate your own vulnerability assessment was. Read the two reports with different questions. See How to Read a Penetration Test Report for the general method.

  • External report: coverage first. Check the list of assets actually tested against your scope and against what discovery found. Unknown assets found by the tester are a finding against your asset management, whatever their severity.
  • External report: authentication on every entry point. Look for an explicit statement per remote access and login surface on whether MFA was enforced and legacy protocols blocked.
  • Internal report: the attack path narrative. A good internal report shows each step from foothold to objective, with the misconfiguration that enabled it. Fixing one link in the chain is cheaper than fixing every medium finding.
  • Internal report: detection timeline. Ask for timestamps of key actions so you can compare them with SOC alerts. A path that went unseen is a detection engineering backlog item.
  • Both: severity in context. NCSC notes CVSS may be used alongside, not in place of, a contextual rating. A medium internal finding on the path to domain control outranks a high external finding on an isolated test host.

Detection and measurement

Keep tester source addresses and timings to match against firewall, identity provider, EDR and SIEM records. Track these metrics across tests:

  • Unknown asset rate: internet-facing assets found by external testers that were missing from the register.
  • Entry points without phishing-resistant MFA: count from the external report, target zero.
  • Time to objective: hours from foothold to domain control or crown jewel in internal tests; it should rise year on year.
  • Detection coverage: proportion of key internal test actions that raised an alert, and mean time to detect.
  • Repeat finding rate: findings reappearing from the previous report of the same type.
  • Remediation within target: high and critical findings closed and retested within the agreed window.

Common failure patterns

  • Treating an external pass as whole-estate assurance: a clean perimeter says nothing about what a phished user can reach.
  • Internal tests with the directory out of scope: the main lateral movement surface is excluded, so the report under-reports risk.
  • Scoping only from the asset register: external testers never see the forgotten host that matters.
  • Privileged starting accounts: starting an internal test from an IT admin build hides the escalation path a normal user would take.
  • Tester credentials left active: accounts and VPN profiles issued for the test outlive it and become a standing exposure.
  • Fixing findings, not paths: closing every medium item while the one link that leads to domain control remains.

Framework mapping

FrameworkReferenceExternalInternal
NIST SP 800-53 Rev. 5CA-8 Penetration Testing; CA-8(1) Independent Penetration Testing Agent or Team; CA-8(2) Red Team Exercises; RA-5 Vulnerability Monitoring and ScanningYesYes
NIST CSF 2.0ID.RA-01 vulnerabilities identified, validated and recorded; ID.IM-02 improvements identified from security tests and exercisesYesYes
CIS Controls v8.118.1 programme, 18.2 external tests, 18.3 remediate findings, 18.4 validate security measures, 18.5 internal tests18.218.5
PCI DSS v4.0.111.4.2 internal, 11.4.3 external, 11.4.4 correct and retest, 11.4.5 segmentation11.4.311.4.2, 11.4.5
ISO/IEC 27001:2022 Annex AA.8.8 Management of technical vulnerabilities; A.5.35 Independent review of information securityYesYes
MITRE ATT&CKT1595, T1190, T1133, T1110.003 (external); T1558.003, T1557.001, T1003, T1649, T1021 (internal); T1078 (both)Initial accessPost-compromise

For a full list of regulatory testing requirements, see the Penetration Testing Requirements Map.

Practitioner checklist

  • Write the risk question for each test at the top of the statement of work.
  • Build the external scope from discovery as well as the register, and confirm ownership of every address.
  • Require an explicit MFA and legacy authentication result for every external entry point.
  • Choose the internal foothold deliberately and keep Active Directory and cloud identity in scope.
  • Decide whether the internal test is announced or unannounced, and record the detection timeline either way.
  • Issue tester credentials as named, time-bound accounts and disable them on the last day.
  • Test each segmentation method and record which paths reach protected environments.
  • Run both tests at least annually and on defined change triggers, with focused retests for high findings.
  • Combine the two reports into one attack chain view for leadership.

Standards and references

Related NHI Mgmt Group resources: Penetration Testing Guide, Penetration Testing Provider Buyer's Guide, Penetration Test Scoping Checklist, Assumed Breach Testing Guide, Attack Surface Management Guide.

PDF for subscribers

Take this guide with you

Every section, table and checklist in one PDF, with links to the sources and related guides. Subscribe to our weekly NHI & AI Identity Journal and we’ll email you a download link.

Free weekly newsletter

Subscribe to the NHI & AI Identity Journal

The latest on NHI and Agentic AI security – articles, research, breaches, news and events every week.

Bonus 33% off our NHI Course when you subscribe.

NHIMG Editorial Note
Researched and written by the NHI Mgmt Group editorial team using AI tools, under the editorial direction of Lalit Choda. Facts are drawn from the sources listed on this page and checked by our automated quality process. Last updated 14 November 2026. Spotted an error? Use our contact page.
NHI Mgmt Group, the independent authority on Non-Human Identity, IAM and Agentic AI security. nhimg.org

PDF Downloads for Newsletter Subscribers

Get the guide

Enter your details and we’ll email you a download link.

The full guide is free to read online. The PDF is a thank you for subscribing.

Fill in all fields marked * and tick the PDF box to continue.