Sample report
Real pentest reports. Anonymised.
Two black-box assessments we delivered: one site in critical shape, one that mostly held up. Client details removed. Findings kept.
Penetration testing and security assessment
- Prepared for
- Assessment
- External, unauthenticated
- Report
- Report A, 41 pagesReport B, 59 pages
- Reference
- 3High
- 7Medium
- 7Low
- 5Info
- 0High
- 11Medium
- 10Low
- 7Info
ConfidentialPage 1 of 41Page 1 of 59
Choose a report
01Assessment resultsReport A
Security at a glance
Overall risk
Critical
Rated critical: a server-level exposure could not be ruled out from outside, and several findings combine.
Findings by severity
22 findings
- 10high and medium findings to fix first
- 41pages in the delivered report
- 12attack classes that held up
02Executive summary
What the owner reads first
A server-level exposure that could not be ruled out from outside, open backend access and weak email protection. Patch first, then close the doors.
Business implications
The main risks: compromise of the publishing server, takeover of editor accounts and bulk harvesting of the site's directory. Missing email authentication makes the brand easy to impersonate.
Immediate priorities
Confirm and patch the server runtime. Update the CMS, protect the login and fix upload validation. Review what the public API returns and switch on email authentication.
Controls that held
No SQL injection, cross-site scripting, file inclusion, open redirect or permissive cross-origin access was confirmed in the paths examined.
03Scope and methodology
What was tested, and how
- Target
- A publishing website on a popular CMS, with a separate API host
- Test
- External black-box penetration test
- Access
- Unauthenticated, no source code
- Window
- A single overnight testing window
Method
- 01DNS and certificate reconnaissance
- 02Content and route discovery
- 03Parameter and injection testing
- 04Header and cookie review
- 05TLS protocol checks
- 06API enumeration
- 07Upload validation
- 08Targeted exploit attempts
In scope
- Public website
- API and CMS host
- TLS endpoints
- DNS and email domain records
Rules of engagement
- No content, account or setting was edited or deleted
- Only additive test submissions, to exercise the upload pipeline
- Request rates kept low to avoid disruption
- Retrieved data handed back to the client
Out of scope
- Third-party service infrastructure
- Signed-in CMS testing
- Social engineering
- Physical testing
- Denial-of-service load
Stated limits
- The exact runtime patch level cannot be proven from outside
- The site's web firewall stopped one exploit probe at its first stage
- Whether an uploaded file could run was not verifiable
Attack surface
- 01
Web firewall and edge
Public HTTP and TLS entry point for both hosts.
- 02
Public website
Dynamic pages, search, filters and a public submission form with uploads.
- 03
API and CMS host
Content API, the editors' login and integration endpoints.
- 04
Email domain
Sender policy, signing keys and DNS assurance records.
04Findings register
22 findings, ranked
Showing 22 of 22
- F-01HighServer runtime needs urgent patch verificationServer
F-02HighAnonymous API access exposes the content archive and contact dataAPIFull write-up
7.5CVSS 3.1estimatedConfirmedIn plain English
The publishing API lets anyone download the content archive and directory records without signing in, contact details included.
Business implications
An attacker can collect the directory in bulk and use named contacts and organisation details to write convincing phishing emails or phone scams. The archive itself can be copied at scale.
What the evidence shows
Anonymous requests returned thousands of records across several collections, including email addresses, phone numbers and postal addresses. Three requests were enough to list the collections and page through them.
Captured requests, generalised GET /api/types 200 every content type listedGET /api/directory?limit=1 200 total: GET /api/directory?limit=100 200 full records, contact fieldsRecommended action
Agree which records and fields are meant to be public. Withhold private fields and unneeded metadata without breaking the public site, and add proportionate request limits.
How to verify the fix
Compare anonymous API responses with the agreed publication rules. Private fields are withheld; public pages still work.
- F-03HighCMS core is behind the current security releaseCMS
F-04MediumA trailing slash bypasses the login-page blockSign-inFull write-up
5.3CVSS 3.1estimatedObservedIn plain English
A rule blocks the normal login address. Add a slash to the end and the sign-in form loads anyway.
Business implications
The block meant to hide the login form can be walked around. Together with the missing throttling in F-05, editor accounts are open to password guessing and credential stuffing.
What the evidence shows
The login address returned 403 Forbidden. The same address with a trailing slash returned 200 and the full sign-in form. The admin routes stayed blocked.
Captured requests, generalised GET /login 403 ForbiddenGET /login/ 200 full sign-in formGET /admin/ 403 ForbiddenRecommended action
Apply the restriction to every equivalent path. Normalise paths and test trailing-slash, encoded and double-slash variants. Keep MFA and login throttling.
How to verify the fix
Every equivalent login address gets the same decision, and real editors can still sign in.
- F-05MediumLogin form has no observed throttlingSign-in
F-06MediumUpload form accepts falsely labelled image filesUploadsFull write-up
5.4CVSS 3.1estimatedValidation failureIn plain English
The upload form accepted a file that was not an image, because the upload said it was a JPEG. The label a browser sends cannot be trusted.
Business implications
Trusting the sender's label lets non-image content into the upload pipeline. If such a file were stored and run as code, it could lead to server-side code execution.
What the evidence shows
A harmless script file labelled image/jpeg was accepted and redirected to the success page. The same bytes labelled honestly were rejected. Size and count limits held. Where the file was stored could not be recovered.
Captured requests, generalised POST /submit note.txt type=text/plain rejectedPOST /submit probe type=image/jpeg 302 successPOST /submit oversized photo rejected, size limitRecommended action
Check content with a maintained image decoder, allow only the formats you need, re-encode images and generate safe names. Store uploads outside executable paths with script execution off.
How to verify the fix
Mislabelled, malformed and double-extension files are rejected, valid images still upload, and stored uploads cannot run as code.
- F-07MediumNo HTTPS-only policy (HSTS)Transport
- F-08MediumNo content security policyBrowser
- F-09MediumEmail authentication lacks DMARC and DKIMEmail
- F-19MediumAsset-style URLs cause HTML pages to be cachedCaching
- F-10LowUnexpected parameter types cause server errorsInput
- F-11LowLegacy TLS 1.0 and 1.1 are still acceptedTransport
- F-12LowA pre-login cookie has no explicit SameSite policySessions
- F-13LowDNS assurance (CAA and DNSSEC) is not publishedDNS
- F-14LowBackend error pages disclose the server versionDisclosure
- F-15LowA legacy integration endpoint returns an origin errorCMS
- F-20LowUnpublished article IDs are distinguishableContent
- F-16InfoPage fragments can be requested directlyDisclosure
- F-17InfoSecurity baselines differ between the two hostsConfiguration
- F-18InfoPublic assets expose software version hintsDisclosure
- F-21InfoUpload folders give distinguishable responsesDisclosure
- F-22InfoPublic contact addresses can be recovered by scriptsEmail
In the delivered report every finding gets this full write-up on its own page.
05Remediation plan
What to fix, in order
- 01First 72 hours
Now
3findings- Prove the runtime and CMS versions, stage the security updates and record the build after restart.
- Check where uploads land and what may run there. Reject non-image content.
- 02First week
Next
4findings- Close the login bypass, add MFA and bounded throttling.
- Agree which API fields are public. Confirm every mail sender, then roll out DMARC.
- 03Within 30 days
Later
15findings- Redirect to HTTPS, stage HSTS, test a content policy, retire legacy TLS and fix path caching.
- Tidy input handling, cookies, DNS assurance and banners as planned maintenance.
06Positive observations
What held up
Attempted against the paths examined. None confirmed.
- SQL injectionClean
- Reflected cross-site scriptingClean
- Stored cross-site scriptingClean
- File inclusion and path traversalClean
- Open redirectClean
- Permissive cross-origin accessClean
- Header injectionClean
- Server-side request forgeryClean
- Host-header and cache poisoningClean
- Directory listingHardened
- Sensitive file exposureHardened
- ClickjackingCovered
01Assessment resultsReport B
Security at a glance
Overall risk
Medium
Rated medium: nothing critical or high. The work is availability, request cost and directory access.
Findings by severity
28 findings
- 11high and medium findings to fix first
- 59pages in the delivered report
- 18attack classes that held up
02Executive summary
What the owner reads first
The core application held up. The work is availability, request cost and how much of the directory a script can collect.
Business implications
An alternate hostname serves blank pages. Search and asset delivery make automated collection cheap and repeat traffic expensive. Email impersonation is only monitored, not blocked.
Immediate priorities
Fix hostname routing and monitor real content. Cut uncached repeat fetches, set request budgets and agree what the directory publishes. Stage HTTPS-only and email enforcement.
Controls that held
Output encoding, method limits, body limits and strict JSON validation resisted every tested attack. Framing is blocked. Concurrent duplicate submissions were handled correctly.
03Scope and methodology
What was tested, and how
- Target
- A creative directory website with search, profiles and a JSON route
- Test
- External black-box penetration test
- Access
- Unauthenticated, no source code or accounts
- Window
- A single testing window
Method
- 01DNS and TLS reconnaissance
- 02Route and parameter discovery
- 03Injection, encoding and method checks
- 04Request-parsing checks
- 05Search and JSON enumeration
- 06Browser header review
- 07Form validation
- 08Bounded rate and connection measurements
In scope
- Public website
- Alternate hostname
- JSON and form endpoints
- TLS endpoints
- DNS and email domain records
Rules of engagement
- No existing record, file or setting was edited or deleted
- The public form was exercised with test identities only
- Rate and connection trials kept bounded
- Directory data came from public endpoints only
Out of scope
- The edge provider's infrastructure
- Their mail provider
- Server administration
- Signed-in testing
- Denial-of-service load
Stated limits
- Effects on the origin's bandwidth cannot be measured from outside
- Rate results reflect one testing window
- No login or upload interface was found to test
Attack surface
- 01
Web firewall and edge
Public HTTP and TLS entry point, in front of a reverse proxy.
- 02
Public website
Search, profile pages and discovery files.
- 03
JSON and form endpoints
A search route that answers in JSON and one documented write form.
- 04
Email domain
Sender policy, signing keys and DNS assurance records.
04Findings register
28 findings, ranked
Showing 28 of 28
- F-01MediumNo HTTPS-only policy (HSTS)Transport
- F-02MediumDMARC monitors impersonation but does not block itEmail
- F-03MediumNo certificate-authority restriction (CAA)DNS
- F-04MediumSearch showed no throttling in a burst testAbuse
- F-05MediumNumeric profile links make the directory easy to walkDirectory
- F-10MediumUnfinished requests hold connections until a timeoutConnections
F-13MediumAn alternate hostname serves blank pagesAvailabilityFull write-up
MediumTypeAvailability defectConfirmedIn plain English
Visitors who use the other common form of the address get a successful response with nothing on it, and no redirect to the working site.
Business implications
Those visitors hit a dead end. Uptime checks that only read the status code miss it, and search engines see an empty page instead of the real one.
What the evidence shows
Repeated homepage requests on the alternate hostname returned success with an empty body. Search, profile, asset and discovery paths did the same. Sending the main hostname in the request header brought the content back, so the fault sits in routing for that one hostname.
Captured requests, generalised GET main host / 200 full pageGET alternate host / 200 empty bodyGET alternate host / 200 empty bodyGET alternate, main Host 200 full pageRecommended action
Redirect the alternate hostname to the main one with path and query kept, or route it to the same application. Add a content check to monitoring.
How to verify the fix
Homepage, search, profile and asset links on the alternate hostname reach real content, and a monitor flags an empty successful response.
- F-14MediumA forwarded visitor-IP header triggers an edge blockIntegration
F-20MediumStatic assets bypass browser and edge cachesCachingFull write-up
MediumTypeAvailability and performanceConfirmedIn plain English
Stylesheets, icons and other common files are downloaded in full on every visit instead of being reused. Some media is cached. The main static bundle is not.
Business implications
Repeat visits cost more bandwidth and server work. With no throttling in front, it also gives anyone a cheap way to make the server do expensive work.
What the evidence shows
Repeated stylesheet requests were never served from the edge cache; repeated media requests were. A unique query string on a cached media file forced a full uncached copy. Two serving layers sent overlapping cache headers.
Captured requests, generalised GET stylesheet, repeated edge: not cachedGET media file, repeated edge: cachedGET media file + new query edge: miss, full bodyRecommended action
Set caching by resource type: long, immutable caching only for versioned file names, validators and a sensible freshness period for the rest. Check edge rules, cache keys and app headers together.
How to verify the fix
Cold and warm requests show cache hits or not-modified responses where expected, and changing a file still refreshes it.
- F-23MediumSearch pages continue past the published limitDirectory
F-28MediumAn undocumented JSON route allows bulk directory downloadDirectoryFull write-up
MediumTypeDirectory access policyConfirmedIn plain English
A JSON search route returns profile names, links, aliases and platform links in a format scripts can collect page by page.
Business implications
Copying the whole directory takes little effort, and the copy is a ready list for impersonation or unwanted contact. Access rules need to cover every format the directory is published in.
What the evidence shows
Anonymous requests returned paged records with a running total, and paging continued to the end of the directory. The records held no email addresses, phone numbers or postal addresses.
Captured requests, generalised GET /?page=1 200 records, total, next pageGET /?page=2 200 next recordsGET /?page=last 200 final records, no sign-inRecommended action
Decide who this route is for and which fields it returns. Document it as a public API, or put it behind the intended access control. Add pagination and request limits.
How to verify the fix
Anonymous responses match the agreed fields and access policy, and request limits apply to the JSON route and public search alike.
- F-06LowDNS answers are not signed (DNSSEC)DNS
- F-07LowOptional browser isolation headers are missingBrowser
- F-08LowPublic metrics reveal overall traffic volumeDisclosure
- F-09LowValidation errors return a full page, not a short messageInput
- F-11LowSeveral held-open connections were accepted at onceConnections
- F-15LowStatic files lack revalidation headersCaching
- F-17LowForm rate limiting behaves differently between runsRate limits
- F-21LowAn oversized image is shown as a small iconPerformance
- F-22LowTrailing-slash and case handling differ between routesRouting
- F-26LowSearch totals understate the real result countSearch
- F-12InfoInput limits and malformed-request handling workedPositive
- F-16InfoCertificate revocation follows the issuer's current modelCertificates
- F-18InfoArray-style query parameters are silently ignoredInput
- F-19InfoRecent profiles are missing from the sitemapPublishing
- F-24InfoCache policy is set by more than one layerConfiguration
- F-25InfoForm validation handles international text and bad JSON safelyPositive
- F-27InfoDuplicate form submissions are handled atomicallyPositive
In the delivered report every finding gets this full write-up on its own page.
05Attack scenarios
What we tried, and what we measured
- 01Confirmed
Cache bypass multiplies serving work
- Attempted
- Requested a cached media file with a fresh query string each time, at rising concurrency.
- Measured
- Every request missed the cache. A tiny request pulled back a response hundreds of times its size, and throughput kept rising with no throttling.
- Outcome
- Confirmed. Fix the cache key and set request budgets first.
- 02Confirmed
Walking the directory
- Attempted
- Paged through public search results in order, without signing in.
- Measured
- A dozen requests returned a couple of hundred unique profiles. The whole directory was a few hundred requests away.
- Outcome
- Confirmed. Agree what the directory publishes, then limit every format.
- 03Confirmed
Hostname and forwarded headers
- Attempted
- Requested the alternate hostname, and sent a forwarded visitor-IP header the edge reserves for itself.
- Measured
- The alternate hostname returned empty pages every time. The forwarded header made the edge refuse the request.
- Outcome
- Availability defects. Neither opened a way past the firewall.
- 04Held up
Holding connections open
- Attempted
- Opened requests that announce a body and never send it, in growing batches.
- Measured
- The edge timeout cut each one off. While they were held, every ordinary request still succeeded.
- Outcome
- Held up. The timeout bounds the exposure.
06Remediation plan
What to fix, in order
- 01First week
Next
7findings- Restore hostname routing with paths kept, and monitor for empty pages.
- Set request budgets and one cache policy, and resize the oversized icon.
- Stage HSTS. Confirm every mail sender, then enforce DMARC.
- 02Within 30 days
Later
10findings- Agree which directory fields, limits and formats are public.
- Coordinate CAA and DNSSEC with certificate renewal.
- Document request deadlines, limiter behaviour and header forwarding.
- 03Planned maintenance
Maintain
11findings- Make browser features, metrics access and error pages deliberate.
- Fix validators, parameters, sitemaps and route variants. Keep the working checks as regression tests.
07Positive observations
What held up
Attempted against the paths examined. None confirmed.
- Reflected cross-site scriptingClean
- Stored cross-site scriptingClean
- SQL injectionClean
- NoSQL injectionClean
- Template injectionClean
- File inclusion and path traversalClean
- Server-side request forgeryClean
- Header injectionClean
- Open redirectClean
- Permissive cross-origin accessClean
- Cache poisoning and deceptionClean
- Request smugglingClean
- Host-header injectionClean
- Subdomain takeoverClean
- Request body limit bypassHardened
- Error-page disclosureHardened
- ClickjackingCovered
- MIME sniffingCovered
Black-box penetration test
Want this report for your site?
Same method, same depth, your scope. Every finding ranked, explained and paired with its fix.
Black-box penetration test
$500
- Executive summary
- Findings
- Remediation plan
- What held up


