External Penetration Testing: Scope, Methods and Reports
Skip to content

External Penetration Testing

External penetration testing is an authorized attempt to get into an organization from the internet, using only what an outsider can reach and find out.

On this page

The testers locate the hosts, services and sign-in pages the organization exposes, then confirm which weaknesses on them lead to data, accounts or the internal network. It is also called an external network penetration test or a perimeter penetration test, and it covers one viewpoint of a wider testing program.

This guide covers what an external pentest includes, from internet-facing hosts and VPN gateways to email, DNS, cloud edges and leaked credentials, then scoping, methods, the report, retesting and frequency.

External and Internal Testing Compared

An external test starts outside the perimeter with no account and no inside knowledge, and an internal test starts inside the network with the access of a user or a compromised device. NIST SP 800-115, the Technical Guide to Information Security Testing and Assessment, describes external testing as work done from outside the security perimeter, usually as seen from the internet, to reveal weaknesses an external attacker could use.

External test Internal test
Starting point A system on the internet, outside the organization’s networks. A device on the internal network, often with an ordinary user account.
Knowledge given Typically the target domains and address ranges, and little else. The network information a user at that level would normally have.
Main targets Hosts, services, remote access gateways, sign-in pages and cloud resources that answer from the internet. Directory services, file shares, internal applications and the paths to administrator rights.
Typical findings Unpatched edge devices, exposed management consoles, weak sign-in protection, open storage. Privilege escalation, weak segmentation, stored passwords, permissive shares.
What it shows Whether an outsider can get a foothold, and where. How far an attacker with a foothold could go.

Order matters when both are planned. NIST advises that the external test usually comes first, so testers do not pick up inside knowledge that an outside attacker would lack.

A foothold found from outside is often where an internal test would start. Some engagements let external testers continue inward up to an agreed stop point, and others end at the perimeter. The types of penetration testing by target and knowledge level are set out in our wider guide.

Attack Surface Management and Testing

Attack surface management keeps the list of what an organization exposes, and an external penetration test proves what that exposure allows. Discovery tells the team that a host answers on a port, and a test tells it whether an outsider could use that host.

  • Discovery feeds the scope: A scope written by hand may omit hosts published without the security team’s knowledge, so it starts from a current discovery inventory.
  • Testing settles priority: A finding confirmed with proof moves ahead of a long list of possible issues from discovery and scanning.
  • Tests find assets too: A host found during the test that was missing from the inventory goes back to discovery with an owner to confirm.

Discovery reads public records and sends light, routine requests all the time. Testing sends deliberate attack traffic to production systems, so it runs only inside a signed authorization with agreed limits.

How discovery works source by source is covered in our guide to external attack surface management.

Hosts and Exposed Services

The core of an external network penetration test is every service that listens on a public address. That takes in the public hostnames under the organization’s domains, the address ranges it owns or rents, and anything that answers on an open port at those addresses.

NIST notes that perimeter firewalls and access control lists usually limit inbound traffic, so initial attacks typically focus on commonly allowed protocols such as FTP, HTTP, SMTP and POP. MITRE ATT&CK records this entry route as Exploit Public-Facing Application (T1190), and its targets include websites, databases, services such as SSH and SMB, and device management protocols such as SNMP.

  • Web servers: The test checks the server software, default pages, exposed directories and files, and the applications behind them.
  • Databases and file services: A database, file share or file transfer service that answers from the internet is checked for access without an account and for default credentials.
  • Management interfaces: An administration console meant for internal use counts as an external exposure once it answers from outside.
  • Forgotten systems: Test copies, old releases and abandoned projects that still answer are often the least maintained hosts in scope.

Management interfaces carry particular weight. CISA’s BOD 23-02 covers the management interfaces of routers, switches, firewalls, VPN concentrators, proxies, load balancers and out-of-band server management. It requires US federal agencies to take such an interface off the public internet, or put it behind zero trust access controls, within 14 days of finding it exposed.

Web applications and APIs on these hosts are in scope too, and the guide to web application penetration testing covers testing them in depth.

VPN and Remote Access Gateways

Remote access gateways are among the highest-value targets in an external test, because they exist to let people from outside into the internal network. They include VPN gateways, remote desktop gateways, terminal servers, SSH services and virtual desktop portals.

Appendix D of NIST SP 800-115 treats remote access testing as a subject of its own and describes the main techniques.

  1. Find unauthorized services. Port scans locate open ports associated with remote access, including services set up outside the normal process.
  2. Review the gateway rules. Rulesets on VPN gateways are reviewed for gaps or misconfigurations that could let unwanted connections through.
  3. Test authentication. Testers confirm that sign-in is required, then try default and maintenance accounts and password guessing within the agreed limits.
  4. Check the traffic. Testers check whether remote access traffic is protected, since unprotected traffic can expose credentials.

NIST adds that intrusive remote access testing should run at times of low demand. MITRE ATT&CK lists the use of VPNs and similar services to get in and stay in as External Remote Services (T1133).

These devices sit at the edge of the network, where attackers look first. ASD’s ACSC and partner agencies, in Mitigation strategies for edge devices, note that both skilled and unskilled actors run reconnaissance against internet-accessible services and often exploit newly released vulnerabilities in edge devices. The guidance asks organizations to know which devices sit at the edge, replace those past end of life, disable unneeded features and ports, keep management interfaces off the internet and use phishing-resistant multi-factor authentication.

A gateway test checks each of those points from outside: the product and version, whether the published fix is in place, whether the management interface answers, and whether every account and sign-in path requires a second factor.

Email and DNS Exposure

Email and DNS are part of the perimeter because anyone can reach them and they decide where traffic and trust go. A perimeter penetration test checks both the servers and the records published about the domain.

  • Mail servers: Testers check whether a mail server relays messages for outsiders, which protocols and encryption it offers, and whether webmail enforces multi-factor authentication.
  • Sender records: The domain’s SPF, DKIM and DMARC records decide whether others can send mail that appears to come from the organization.
  • DNS servers: Testers check whether the organization’s name servers hand out a full copy of a zone or answer recursive queries for anyone.
  • Dangling records: A DNS record that still points at a deleted cloud resource can let someone else claim the name.

A public benchmark exists for the mail settings. CISA’s BOD 18-01 required US federal agencies to enable STARTTLS on internet-facing mail servers, publish SPF and DMARC records, and within one year set a DMARC policy of reject for all second-level domains and mail-sending hosts.

Lookalike domains registered by others are a brand protection concern, outside a test of the organization’s own domains.

Cloud Edges in Scope

Cloud resources are in scope wherever they answer from the internet, even though they sit outside the organization’s own address ranges. CISA’s BOD 26-04 defines a publicly exposed asset as one accessible to unauthenticated or untrusted entities through public networks, regardless of its physical or logical location.

  • Storage: Buckets and containers that list or serve files to anyone are checked with the least intrusive request that shows access.
  • Published services: API gateways, function URLs, managed databases with public endpoints and cloud virtual machines are tested like any other exposed service.
  • Origin servers: Where a content delivery network fronts a site, testers check whether the origin server also answers directly and bypasses that protection.
  • Exposed secrets: Keys and tokens in public files or page source are checked for whether they still work, inside the limits agreed for credentials.

Cloud providers set terms for testing on their platforms. Microsoft’s Security Testing Rules of Engagement, for example, limit testing to the customer’s own tenant or assets it has explicit authorization for, and prohibit denial of service testing and network-intensive automated testing that generates excessive traffic.

The test examines the customer’s configuration, and the provider’s platform stays out of scope. A cloud account inventory is a key scoping input, because cloud addresses change often and rarely appear in registered ranges.

Leaked Credentials and Sign-In

Sign-in pages are part of the external scope because a valid password can be a quieter way in than an exploit. They include single sign-on portals, webmail, VPN and remote desktop portals, customer portals and account recovery pages.

MITRE ATT&CK describes Credential Stuffing (T1110.004) as the use of credentials from breach dumps of unrelated accounts, which works because people reuse passwords. Once a credential works, the attacker holds Valid Accounts (T1078), and later activity may look like an ordinary sign-in.

  • Sign-in inventory: Testers list every page and protocol that accepts a password from the internet, including older portals that a single sign-on rollout left behind.
  • Multi-factor coverage: Testers check whether every sign-in path asks for a second factor, including legacy protocols, older gateways and exempt accounts.
  • Leaked credentials: Credentials from breach data or infostealer logs are matched to current accounts and, with approval, tested once each.

Username discovery, lockout and password recovery on each page are application-level checks, covered under web application testing. Leaked credential tests need the strictest controls in the scope. Each typically waits for written approval, and the test stops at the first page after sign-in.

Where a cloud provider runs the sign-in page, its terms come first: Microsoft’s rules, for instance, prohibit using credentials that are not the tester’s own, including ones leaked publicly. The identity team may then confirm and reset the password while the tester checks the controls around it.

ASD’s Essential Eight maturity model asks for multi-factor authentication for users of an organization’s online services that process, store or communicate its sensitive data. An external test checks directly that every path to those services enforces it.

Scoping and Rules of Engagement

An external test is scoped by listing what may be attacked, confirming it belongs to the organization, and agreeing how the test traffic will behave. The general structure of penetration test rules of engagement is covered in our main guide, and these are the points the perimeter adds.

  1. List the targets. The scope names domains, address ranges, hosts and cloud accounts, plus an exclusion list, such as a payment page under a separate assessment.
  2. Confirm ownership. NIST advises verifying that every public address on the target list is under the organization’s control, using registration records such as WHOIS.
  3. Get third-party consent. An address on shared hosting, a content delivery network or a managed service belongs to a provider, whose terms or consent apply.
  4. Fix the test origin. NIST recommends listing the test systems’ source addresses in the plan, so administrators can tell test traffic from a real attack.
  5. Decide on allowlisting. Letting the testers through a firewall shows more of the systems behind it, though NIST notes the view no longer reflects an outside attacker.

Limits, contacts and stop criteria are then written down as in any rules of engagement, with a request rate cap for each host and time windows for the sensitive ones.

Where testing crosses third-party networks, NIST warns that it may look malicious to their operators and may breach their policies. Telling the hosting or internet service provider in advance avoids blocked traffic and confused incident reports.

Methods From Outside the Network

An external test follows the phases of any penetration test, adapted to a starting point with no access. NIST describes external testing that begins with reconnaissance, moves to enumeration of hosts and listening services, then attacks what is reachable.

  1. Reconnaissance. Testers collect system names, addresses and technical contacts from registration data, DNS and certificate records and other public sources.
  2. Enumeration. Network discovery and port scans show which hosts answer and which services listen. NIST notes that external scans see only the ports the firewall passes.
  3. Identification and analysis. Testers work out each service’s product and version from its responses, pages and certificates, match it against published advisories, and check for default settings, weak authentication and exposed files.
  4. Exploitation. Testers use a weakness with the least intrusive request that proves it. Where the rules allow, the foothold is used to reach hosts hidden from outside, which NIST describes as using minimal access to gain greater access.

Known exploited vulnerabilities come first. For an entry in CISA’s Known Exploited Vulnerabilities Catalog that affects an exposed product, the affected product is matched to the versions recorded on the organization’s hosts.

Each exposed instance is then checked for the fixed build. A patched instance is recorded and closed, and on an affected one the tester proves the weakness and raises a finding naming the host and the fix.

An External Test at Acme-corp

acme-corp, a wholesale distributor, commissions an external test of acmecorp.com and its address range 203.0.113.0/28, asking whether an outsider could reach its internal network or staff mailboxes.

  1. Scoping. The scope lists the domain, the range and the cloud account, and excludes pay.acmecorp.com. A registration check shows one address belongs to the hosting provider’s shared load balancer, so it is removed pending consent.
  2. Reconnaissance. Certificate and DNS records list www, mail, vpn, portal and erp, plus vpn-old.acmecorp.com, which is missing from the asset register.
  3. Enumeration. vpn-old answers with the portal of an older remote access gateway, and a firewall answers on port 8443 with its administration sign-in page.
  4. Analysis. The release on vpn-old matches an entry in CISA’s catalog, with the fixed build absent. The domain’s DMARC policy is set to none.
  5. Credential test. A staff address and password appear in breach data. With written approval, the tester tries it once on vpn-old, it works, and no second factor is requested. The tester stops at the first page and opens no tunnel.
  6. Proof without harm. For the catalog entry, the tester runs the least intrusive check that shows the affected code responds, and keeps the request and response.
  7. Early report. The VPN findings are phoned to the named contact the same day, ahead of the written report.

acme-corp’s IT manager disables vpn-old, resets the password and removes the firewall’s administration page from the internet. The retest from outside shows all three closed, the DMARC change follows, and the SOC searches gateway and sign-in logs for earlier use of the password.

Contents of an External Report

A good external report shows what was tested from the internet, what was proven, and how to check each claim against the organization’s own records. It adds perimeter detail to the usual parts of a report.

  • Asset list as found: The report lists every host and service that answered, with those missing from the inventory marked.
  • Finding location: Each finding gives the hostname or address, port, product and version.
  • Exposure path: Where findings were chained, the report traces the route from the internet to the furthest point reached.
  • Fix: Each fix names the specific change, typically a patch, a configuration change or removal of the service from the internet.
  • Limitations: The report names the hosts left untested and why, including traffic blocked by filtering devices, which NIST notes can hide weaknesses from a test.

The records behind the report are what make it checkable.

Artifact What it confirms Source
Signed scope and authorization The domains, ranges, cloud accounts and exclusions approved. The test plan, signed by the system owner.
Ownership check for each address That each target belongs to the organization or has provider consent. Registration records and the cloud account inventory.
Test source addresses and windows Which traffic came from the testers, and when. The test plan, matched against firewall and web server logs.
Proof of each finding The request and response that showed the weakness. The finding record, matched against the target host’s logs.
Sign-in attempts Which credential tests ran and their result. The identity provider’s sign-in logs.
Detection record Whether monitoring raised alerts during the test. The SIEM or SOC case history for the test window.
Retest result That each fix holds when tested from outside. The dated retest output for each closed finding.

How a test report differs from a scan report is covered in our comparison of penetration testing vs vulnerability scanning.

Retesting and Closing Findings

A finding from an external test closes when a retest from the internet shows the exposure is gone. NIST counts a retest as verification of a fix, alongside an audit of the system.

The retest repeats the original proof from the same outside position. Where the fix was a block rule, a second source address shows whether the rule blocks everyone or only the tester.

  • Patched service: The fixed build is in place and the proof no longer works.
  • Removed service: The host or port no longer answers from the internet, including on its alternative names.
  • Reset credential: The leaked password fails, and multi-factor authentication applies on every path.

A fix closes the weakness, and it leaves open whether someone used it first. BOD 26-04 pairs some remediation deadlines with a forensic triage of the asset to assess whether it was compromised, and the same reasoning applies to any confirmed external weakness. A search of the host’s logs, and of sign-in logs for leaked credentials, answers the question.

Frequency of External Tests

External tests typically run at least once a year and after significant change, and many organizations test the perimeter more often, since it changes fast and attackers scan it constantly.

PCI DSS v4 requirement 11.4.3, for example, asks for external penetration testing at least once every 12 months, and again after any significant infrastructure or application upgrade or change.

Between scheduled tests, a change at the edge is the usual reason to test again: a new hostname, a newly open port, a DNS record that points somewhere new, a changed product version or a new advisory for an exposed product. What each standard expects on frequency, and how change-triggered testing is set up, is covered in our guide to continuous penetration testing.

Setting Up External Testing

An organization starting external testing gets most value by settling its scope and contacts first.

  1. Collect the scope. List the registered domains, address ranges and cloud accounts, and name an owner for each.
  2. Run discovery. Compare what answers from the internet with the asset register, and decide on each unknown host.
  3. Sign the authorization. System owners sign the scope, exclusions, limits, contacts and stop criteria, and hosting providers are told.
  4. Test the edge first. Remote access gateways, management interfaces and services matched to CISA’s catalog of known exploited vulnerabilities lead the list.
  5. Check sign-in. Confirm that every external sign-in path asks for a second factor, and handle leaked credentials under their own approval.
  6. Fix and retest. Each confirmed finding goes to its owner with the proof, and a retest from outside closes it.

By the end of the first month, most organizations have a confirmed external inventory, a short list of proven findings with owners, and a written rule for what happens when the perimeter changes.

External Testing With ThreatDefence

On the perimeter, ThreatDefence keeps discovery and testing in one loop. Its AI SOC platform automates security monitoring, compliance and cyber risk management, and a 24x7 SOC stands behind it.

  • Discovery: The ASM agent keeps a live record of the organization’s internet-facing hosts, services and certificates, and the Brand agent watches for leaked staff credentials.
  • Testing: The testing agent tests systems in the confirmed scope when they change, when a new advisory affects an exposed product, and on a rolling schedule, proving which weaknesses are exploitable and keeping the evidence.
  • Limits: Tests stay inside isolated sandboxes whose network policy enforces the scope, with capped request rates, agreed time windows, least intrusive proof, and approval before higher-impact tests such as a leaked credential check.
  • Follow-up: SOC analysts review significant findings, each problem becomes one tracked action item, and a reported fix is retested from outside.

For monitored customers, the Defender and Hunter agents search telemetry for earlier use of a confirmed weakness. Internal networks fall outside the agents’ testing, so an independent human-led engagement covers them where needed. Details are on the agentic penetration testing page.

Frequently Asked Questions

What is external penetration testing?

External penetration testing is an authorized test of an organization’s internet-facing systems, carried out from outside its network with no account or inside access. Testers find the hosts, services and sign-in pages that answer from the internet and confirm which weaknesses on them could be used to reach data, accounts or the internal network. It is also called an external network penetration test or perimeter test.

Does an external penetration test need access to the organization's network?

No. An external test needs only the list of domains, address ranges and cloud accounts in scope, the signed authorization and named contacts. Everything else is found from the internet, the way an outsider would find it. Test accounts may be created for sign-in checks, and the organization’s logs are useful afterward to confirm what monitoring saw.

Should the testers' IP addresses be allowlisted?

No, in most cases. Allowing the testers through firewalls and intrusion prevention shows more of the systems behind them, though NIST notes that the result no longer reflects an outside attacker. A common approach lists the test source addresses in the plan so the SOC can recognize them, keeps defenses unchanged, and treats what monitoring sees as part of the result.

Are VPN gateways in scope of an external penetration test?

Yes. VPN gateways, remote desktop gateways and other remote access services are among the main targets, because they exist to let people from outside into the network. NIST SP 800-115 describes remote access testing that finds unauthorized services, reviews gateway rules and tests authentication. ASD and partner agencies also single out edge devices as frequent targets of newly released exploits.

Can leaked credentials be used in an external penetration test?

Yes, with written approval and strict limits. A leaked credential is typically tested once against the sign-in page it belongs to, and the test stops at the first page after sign-in. Where a cloud provider runs the sign-in page, its testing terms apply, and the identity team may confirm and reset the password instead while the tester checks the surrounding controls.

Does external penetration testing cover cloud-hosted systems?

Yes, where they answer from the internet and the organization has authorized testing of them. Storage, published services, origin servers behind content delivery networks and cloud virtual machines are all in scope. Testing covers the customer’s configuration, and the provider’s own platform is left alone under the provider’s published rules of engagement.

Does an external penetration test include denial of service testing?

No. Denial of service is normally excluded from an external test, together with anything that changes or removes data. Request rates are capped for each host, and sensitive systems are tested in agreed windows. Cloud providers such as Microsoft prohibit denial of service testing on their platforms, so a load test needs its own separate agreement where it is wanted at all.

Should a new firewall rule or DNS change trigger an external test?

Yes. A firewall change can open a port and a DNS change can point a name at a different server, and either may expose a system that was never tested. Continuous external testing treats such a change as a reason to test the affected system, ahead of the next scheduled engagement.

Should an external test include hosts missing from the asset inventory?

Yes, once the organization confirms they are its own and authorizes testing. Hosts missing from the inventory are often the least maintained in scope. Discovery from the organization’s domains typically surfaces them, and each is checked against registration and cloud records before it is added to the test scope.

Talk to Our Experts

See how ThreatDefence automates security monitoring, compliance and cyber risk management in one platform.

    I'm an MSP, MSSP, or reseller

    By submitting this form, you accept the terms in our privacy policy.