On this pageApplication Security Testing MethodsStandards Behind the TestAuthentication and Account RecoverySession HandlingAccess Control Between AccountsInjection and Input HandlingBusiness Logic TestingAPI Penetration TestingScoping a Web Application TestOne Application Test Followed ThroughFindings, Evidence and the ReportRetesting a Fixed WeaknessWhen to Test an ApplicationPlanning the First Application TestWeb Application Testing With ThreatDefenceFrequently Asked Questions It shows which weaknesses in sign-in, sessions, access control, input handling and business rules can be used, and what they expose. Its value depends on the scope, the test accounts and how quickly fixes are retested. This guide covers the testing methods, the standards testers work from, each area of a test, scoping, a worked example, reports, retests and when to test again. Application Security Testing MethodsApplication security testing is the wider family of techniques for finding weaknesses in software, and penetration testing is the one that confirms them against a running application. NIST SP 800-115 describes application security assessment as ranging from source code review to penetration testing of the implemented application. Method What it examines Finds well Tends to miss Static application security testing (SAST) Source code, without running it. Unsafe functions, injectable queries, hard-coded secrets. Flaws that appear only at runtime. Software composition analysis (SCA) Third-party libraries and versions. Components with published vulnerabilities. Flaws in the organization’s own code. Dynamic application security testing (DAST) The running application, through automated requests. Reflected input, missing headers, known injection patterns. Access control and business rules. Penetration testing The running application, examined by a tester. Access control, business logic, chained weaknesses, proof of impact. Code paths the tester never reaches. Documentation or source code-led (hybrid) testing The running application, with design documents and code at hand. Everything above, plus missing controls that leave no visible trace. Little, at the cost of more preparation. SAST, SCA and DAST can run on every build, while a penetration test is placed where human judgment adds most, such as a new application, a redesigned sign-in or a new API. Our comparison of penetration testing and vulnerability scanning shows how the two divide work across the wider estate. Standards Behind the TestA web application test is usually planned against public standards, so that the scope and the result mean the same thing to the tester, the developers and an auditor. Most come from OWASP, the Open Worldwide Application Security Project, a vendor-neutral nonprofit. Standard Published by What it contains How a test uses it OWASP Top 10:2025 OWASP Ranked web application risk categories, led by Broken Access Control. A shared vocabulary for findings. Application Security Verification Standard (ASVS) 5.0.0 OWASP, May 2025 Around 350 requirements for web applications and services, in chapters such as Authentication, Session Management and Authorization, at levels L1 to L3. A test plan, and a pass or fail record per requirement. Web Security Testing Guide (WSTG) 4.2 OWASP Test cases by area, from information gathering through business logic and API testing. The method for each test area. API Security Top 10 2023 OWASP Risks specific to APIs, led by Broken Object Level Authorization. Classification of API findings. SP 800-115 NIST The technical guide to security testing, with an appendix on application testing. Planning, rules of engagement and repeat testing. Information Security Manual (ISM) ASD’s ACSC Australian government controls, including web application development. Control references for Australian organizations. Penetration Testing Guidance PCI Security Standards Council A supplement on application-layer and network-layer testing for PCI DSS. Role coverage and scope for payment applications. The ISM ties several of these together. In its Guidelines for software development, ISM-0971 has the ASVS used in developing web applications, ISM-1850 has the OWASP Top 10 mitigated, and ISM-1851 has the OWASP API Security Top 10 mitigated in developing web APIs. ISM-0402 asks that software is tested for vulnerabilities using SAST, DAST and SCA before its initial release, before each later release, and periodically. Authentication and Account RecoveryAuthentication testing checks whether the application confirms who a user is at every way in, and whether those ways in can be bypassed or abused. The sign-in form receives most attention, and weaknesses often sit in the side doors around it. Sign-in: The tester checks how failed attempts are limited, whether error messages reveal which usernames exist, and whether a multi-factor step can be skipped by calling the next page directly. Registration: Self-registration is checked for who may sign up, whether an email address must be confirmed, and whether a new account can claim a role or an organization it has no right to. Account recovery: Password reset is examined for predictable or reusable tokens, links that stay valid after use, and responses that reveal whether an account exists. Single sign-on: Where sign-in passes through an identity provider, the tester checks that the application validates the tokens it receives and maps each identity to the right local account. Credentials in transit: The tester confirms that credentials travel only over HTTPS, which ISM-1552 requires for all web application content, and that none are left in pages, scripts or logs. Session HandlingSession testing checks how the application remembers a signed-in user between requests, and whether that memory can be stolen, guessed or kept beyond its intended life. A session token works as a temporary password, so a weakness here can bypass even strong sign-in controls. ISM controls added in June 2025 describe what good session handling looks like. Cookie flags: ISM-2063 asks that session cookies set the HttpOnly, Secure and SameSite flags by default where supported. Token content: ISM-2064 asks that session cookies contain only digitally signed opaque bearer tokens, and ISM-2065 asks for non-sequential random identifiers with at least 128 bits of entropy where the tokens are unsigned. Server-side control: ISM-2066 asks that sessions are centrally managed on the server, so that ending a session there ends it everywhere. Single logout: ISM-2067 asks that applications supporting single sign-on equally support single logout. A tester replays a captured token after sign-out, after a password change and after the idle timeout to see whether the server still accepts it. Tokens are also sought in URLs, logs and browser storage that scripts can read. Where the application uses self-contained tokens such as JSON Web Tokens, the checks extend to signature validation and expiry, which ASVS 5.0 covers in a chapter of its own. Access Control Between AccountsAccess control testing checks whether each account can reach only its own data and the functions its role allows. In a multi-user application it is typically where the most serious findings appear, because one missing check can expose every customer’s records. Broken Access Control is first in the OWASP Top 10:2025. The best-known form is the insecure direct object reference (IDOR). ASD’s ACSC, CISA and NSA describe it in their joint advisory, Preventing Web Application Access Control Abuse, as an access control vulnerability in which an application uses an identifier to reach an object without checking the user’s authentication or authorization. Testing needs accounts arranged for comparison. Horizontal checks: A pair of accounts at the same role lets the tester replay each account’s requests with the other account’s identifiers. Vertical checks: An ordinary account calls the routes and functions reserved for administrators or support staff. Unauthenticated checks: Every route that returns data is also called with no session at all. Tenant checks: In a multi-tenant application, an account in one customer organization requests objects that belong to another. The joint advisory asks organizations that run their own web applications to conduct regular, proactive penetration testing, with users of different privilege levels. For payment applications, the PCI guidance asks for testing against all roles, and on multi-tenant platforms for credentials equivalent to a customer user, so the tester can confirm that access stops at that customer’s own data. Injection and Input HandlingInput handling testing checks whether values sent by a user can change what the application does, for example by becoming part of a database query, a system command or a page shown to another user. Injection is fifth in the OWASP Top 10:2025. NIST SP 800-115 gives the classic case: a web application embeds a user’s value in a database query, and unfiltered SQL commands in that value run as part of the query. CISA and the FBI published a Secure by Design Alert on eliminating SQL injection in March 2024, after exploitation of SQL injection defects in a managed file transfer application affected thousands of organizations. The tester works through each place the application accepts input. Database queries: Parameters, form fields, headers and cookies are tested for SQL and NoSQL injection. Browser output: Values returned in a page are checked for cross-site scripting, the risk ISM-1241 addresses by asking for output encoding on all output. Server-side requests: Fields that take a URL, such as a webhook or an image import, are checked for server-side request forgery, which can reach internal services. File uploads: Upload features are checked for file type, size and storage location, and for whether an uploaded file can run on the server or be served to other users. Error handling: Malformed input shows whether the application fails safely, which the OWASP Top 10:2025 now treats as its own category, Mishandling of Exceptional Conditions. Proof stays proportionate: a tester typically shows that injection works with a harmless query, such as one that returns the database version, and stops before reading tables of customer data. Business Logic TestingBusiness logic testing checks whether the application’s rules can be used in ways the business never intended, while every individual request still looks valid. Examples include applying a discount twice, skipping a payment step, approving one’s own request, or ordering a negative quantity. The WSTG treats business logic as a test area of its own, and ASVS 5.0 groups it with validation in chapter V2. These flaws depend on knowing what a process is meant to allow, so the tester gathers that knowledge before the test starts. Process walkthroughs: The product owner explains the workflows that move money, change permissions or release goods. Abuse cases: The tester and the owner list what a dishonest customer, partner or employee would try in each workflow. Order and timing: Steps are taken out of order, repeated, or sent at the same moment from two sessions to look for race conditions. Limits: Quantities, amounts, dates and counts are pushed past the values the interface allows. The ASVS states that automated tools cannot fully verify the more complicated requirements, above all those relating to business logic and access control. Agents confirm many technical weaknesses quickly, and judging whether a permitted action makes business sense is one place where AI penetration testing still hands over to a human tester. API Penetration TestingAPI penetration testing applies the same methods to the routes that a front end, a mobile app or a partner system calls directly. An API often exposes more than the pages built on it, because each route answers to whoever sends a well-formed request, and old routes may stay live after the interface stops using them. The OWASP API Security Top 10 2023 sets out the risks testers check. API risk What the tester does API1 Broken Object Level Authorization Replays a request with another account’s object identifier. API2 Broken Authentication Tests token issue, validation and expiry on every route. API3 Broken Object Property Level Authorization Looks for fields the user should never see, and sends extra fields such as a role. API4 Unrestricted Resource Consumption Checks limits on page size, request rate and upload size, short of denial of service. API5 Broken Function Level Authorization Calls administrative routes from an ordinary account. API6 Unrestricted Access to Sensitive Business Flows Checks whether purchase or sign-up can be automated at scale. API7 Server Side Request Forgery Tests parameters that make the server fetch a URL. API8 Security Misconfiguration Reviews headers, error detail, cross-origin rules and unused methods. API9 Improper Inventory Management Looks for old versions and undocumented routes that still answer. API10 Unsafe Consumption of APIs Checks how data from third-party APIs is handled. Route discovery comes first. The tester collects routes from the application’s scripts, any published API description such as an OpenAPI file, mobile app traffic and older version paths, then compares the list with what the developers believe is live. GraphQL endpoints also get checks for schema introspection and for queries that reach data the equivalent REST routes refuse. Scoping a Web Application TestScoping fixes what will be tested, from where, with which accounts and under what limits, and it decides most of what a test can find. A scope that lists hostnames without roles or accounts typically produces an unauthenticated test of the sign-in page and little else. Hostnames and routes: The scope lists every hostname, API base path and version to test, including separate admin and partner portals. Environment: It states whether testing runs in production or in a staging copy, and how closely that copy matches production in code, configuration and data. Roles and accounts: It names each role to test, with a pair of accounts at the main customer role and an account at each higher role. Knowledge level: It records whether the tester receives documentation, API descriptions or source code, which the ASVS strongly encourages. Third parties: It identifies components run by another provider, such as a hosted payment page or an identity provider, which are excluded or need that provider’s permission. Limits: It sets time windows, request rates, excluded actions such as deleting records or emailing real customers, and the contact to call if something goes wrong. Bought software changes the scope. The PCI guidance notes that a commercial off-the-shelf web application typically gets a network-layer test of its implementation, configuration and maintenance in place of an application-layer test, since the organization does not own its source code. The hosts, ports and services underneath the application usually belong to an external penetration test, and the two scopes are easiest to manage when they are agreed together. One Application Test Followed Throughacme-corp runs an ordering portal at orders.acmecorp.com for its business customers, backed by an API at api.acmecorp.com that its mobile app also calls. One engagement shows how the areas above come together. Scope and accounts. The head of digital signs the authorization for both hostnames, with the hosted payment page excluded. acme-corp creates two customer accounts in separate test companies and one support-agent account, and agrees a weekday window with read-only proof. Mapping. The tester signs in as each account through an intercepting proxy and records every route the portal and the mobile app call. An OpenAPI file at api.acmecorp.com/docs lists an order export route that no screen uses. Access control. Replaying the first company’s order request with an order number from the second test company returns the second company’s order. The tester stops after that single response and records it. Mass assignment. A profile update that includes a role field set to support agent is accepted for the tester’s own account, which then opens the support queue. Session handling. A token copied before sign-out is still accepted an hour later. Business logic. Working from the product owner’s walkthrough, the tester finds that two credit notes can be raised against one order when both requests arrive within the same second. Fix and retest. Developers add an ownership check in shared middleware and reject unknown fields in profile updates. The retest passes on the order route and fails on the export route, which used separate code, so that finding reopens until a second fix passes. The detection side closes the loop. acme-corp’s SOC searched the API gateway logs for runs of sequential order numbers from one account and for bursts of 403 and 404 responses, which the joint advisory associates with enumeration, and found no earlier use of the flaw. The export route was also added to the API inventory. Findings, Evidence and the ReportA web application report is useful when a developer can reproduce each finding without asking the tester a question. The general anatomy of a report is covered in our penetration testing guide, and an application report adds detail at the level of routes, roles and requests. Route and method: The finding names the hostname, path, HTTP method and parameter involved. Role and accounts: It states which test account sent the request and whose data or function it reached. Request and response: It holds the exact request and the relevant part of the response, with personal data masked. Classification: It maps the weakness to an OWASP Top 10 or API Security Top 10 category, and to the ASVS requirement that failed. Severity: The rating reflects what was demonstrated in this application, which may differ from a generic score. Fix: The remediation names the control to add and where, such as an ownership check on every route that takes an order identifier. The ASVS observes that traditional penetration testing reports list failures only. It asks a verification report to include the scope, a summary of all requirements checked, the exceptions and guidance on resolving them. Evidence What it establishes Where it is kept Authorization letter for the hostnames Which hosts, roles and actions were permitted, and when. The engagement file held by the system owner and the tester. Test account register Which accounts existed, their roles, and when they were removed. The application’s user administration screen or the identity provider. Proxy project or traffic capture Every request and response the tester sent. The tester’s intercepting proxy, exported at the end of the test. Route and role matrix Which role reached which route, and the result. The tester’s working papers, attached to the report. API description Published routes, set against the routes found live. The OpenAPI or GraphQL schema in the build repository. Application and gateway logs What the application recorded during the test window, and any alerts. The web server, API gateway or log platform, filtered to the test dates. Code change How each finding was fixed. The pull request or change ticket linked from the finding. Retest record The repeated request, its result and the date. The report appendix or the finding’s history in the tracking system. Retesting a Fixed WeaknessA retest sends the proof from the finding again, now against the fixed application, to confirm that the weakness is gone. NIST SP 800-115 notes that a test team can verify a mitigation only by performing a mirror copy of the original test, which is why the exact request in each finding matters. A fix in one place often leaves the same flaw elsewhere, so a retest goes wider than the single request. Same route, every role: The retest replays the request with each test account, including the higher roles. Sibling routes: Routes that handle the same object, such as list, export and older version paths, are checked for the same flaw. Bypass attempts: Variations of the original input, such as other encodings or identifier formats, show whether the fix is a filter or a real control. Regression: Findings closed in earlier tests are checked again where the code around them has changed. Where a flaw is fundamental to the application, the PCI guidance asks whether the remediation itself is a significant change, which may widen the retest. Each retest ends in a recorded state: closed, reopened, or accepted as a risk by the system owner with a review date. Test accounts created for the engagement are then removed. When to Test an ApplicationAn application is typically tested before it first goes live, after significant changes, and periodically in production. NIST SP 800-115 adds significant changes in the threat environment where the application operates as a reason to test again. A change to sign-in or to roles warrants a test of any system. Changes specific to an application include the following. New API routes: Routes that take object identifiers are a common source of access control flaws. Changed business flows: A new payment, refund or approval flow brings new logic to abuse. New framework or major dependency: A change to the framework underneath can alter session, encoding and routing behavior. In an application that ships every week, an annual test leaves most releases untested. Many teams combine an in-depth test once or twice a year with SAST, DAST and SCA before each release, the pattern ISM-0402 describes, and targeted penetration tests when one of the changes above lands. The frequency rules of PCI DSS, the ISM and APRA CPS 234, and the full list of changes that warrant a test, are set out in the guide to continuous penetration testing. Planning the First Application TestA new program typically starts with an inventory and one well-scoped test. List the applications. Record every web application and API the organization exposes, with its owner, the data it holds, and whether it was built in-house or bought. Rank them. Put first the applications that hold customer data, take payments, or let separate customers sign in to the same system. Set a target level. Choose an ASVS level for each tier, noting that the ASVS expects most applications to strive for L2 and gives online banking as a case where anything less than L3 is hard to justify. Prepare accounts and documents. Set up the role accounts and collect the API descriptions and architecture notes in advance. Run the first test. Test the top-ranked application and agree how findings will reach the developers while the test is still running. Fix and retest. Track each finding to a passing retest or a recorded risk acceptance. Set the cadence. Agree which changes trigger a new test, and add the next applications to the plan. Organizations without in-house testers can keep the cycle running between annual engagements with a subscription model, described in our guide to penetration testing as a service. Web Application Testing With ThreatDefenceBetween releases and annual tests, internet-facing web applications and their APIs can be tested by the testing agent, an offensive agent in the ThreatDefence AI SOC platform. That platform automates security monitoring, compliance and cyber risk management, with SOC analysts on duty 24x7. Scope: Each application enters scope by hostname once the system owner authorizes it. Method: The agent studies the pages, scripts, sign-in flows and configuration, then tests sign-in, sessions, cross-account access, input, exposed API descriptions and exposed files. Accounts: It registers its own test account where registration is open, and client-supplied accounts and leaked-credential sign-in tests wait for approval. Limits: Proof uses the least intrusive request, with data modification, data removal and denial of service excluded. Findings: Each finding carries the request, response, severity by exploitability and fix steps, SOC analysts review significant ones, and a reported fix triggers a retest. Applications reachable only from an internal network fall outside agentic penetration testing, and business logic review or an independent report can come from a human-led test alongside. Frequently Asked QuestionsWhat is web application penetration testing?Web application penetration testing is an authorized test in which a tester tries to break into a web application and its API the way an attacker would. It covers authentication, sessions, access to other accounts’ data, input handling and business rules. The result is a set of confirmed weaknesses, each with the request that proves it, a severity and a fix, followed by a retest once the fix is in place.Is a web app pentest the same as a DAST scan?No. A DAST scan sends automated requests to a running application and matches the responses against known patterns. A penetration test relies on a tester who understands how the application works, so it can find access control and business logic flaws that a scanner has no pattern for. The OWASP ASVS notes that automated tools cannot fully verify requirements on business logic and access control.Does a web application penetration test cover the OWASP Top 10?Yes, in most tests, although coverage depends on the scope and the accounts provided. Categories such as Broken Access Control, Injection and Authentication Failures are tested directly. Others, such as Software Supply Chain Failures and Security Logging and Alerting Failures, may also need a review of build pipelines or logs. A report that lists the ASVS requirements checked shows coverage more precisely.How is API penetration testing different from web application testing?API penetration testing focuses on routes that answer directly to any well-formed request, without the screens that limit what a user can try. Testers first find every route, including undocumented ones and older versions, then check object, property and function level authorization on each. The OWASP API Security Top 10 2023 lists the risks, and a web application test usually includes the API the application calls.Should a web application be tested in production or in staging?Either can work, and the choice depends on risk and on how closely staging matches production. Staging allows more intrusive tests and data changes, but results may differ where configuration, data or third-party connections differ. Production testing finds what an attacker would find, so it typically runs with read-only proof, agreed windows and test accounts that hold no real customer data.How many test accounts does an authenticated web app pentest need?A pair of accounts at the main user role and one account at each higher role is a common minimum. The pair allows horizontal checks, in which one account tries to reach the other’s data, and the higher roles allow vertical checks of administrative functions. In a multi-tenant application the pair should belong to separate customer organizations, so that tenant isolation is tested as well.What is the OWASP ASVS used for in penetration testing?The ASVS gives a tester a list of testable security requirements for web applications and services, grouped by chapter and by levels L1 to L3. A test can be planned against a chosen level and reported requirement by requirement, which makes results comparable between tests and providers. OWASP does not certify vendors or software, so a claim of official ASVS certification deserves caution.Can a SaaS application be penetration tested?Yes, with the provider’s permission. The organization’s own configuration, accounts and integrations can usually be tested inside limits the provider publishes or agrees to. The joint ASD’s ACSC, CISA and NSA advisory on web application access control notes that testing a software-as-a-service product may interfere with the provider’s operations, which is why that permission comes before any test.What is an IDOR vulnerability?An insecure direct object reference (IDOR) is an access control flaw in which an application uses an identifier, such as an order or account number, to fetch a record without checking that the requester may see it. Changing the number in a request then returns someone else’s data. ASD’s ACSC, CISA and NSA published a joint advisory on IDOR in July 2023, and the API Security Top 10 calls it Broken Object Level Authorization.Does every release of a web application need a new penetration test?No. Most releases are covered by automated testing before release and by code review. A targeted penetration test is worth running when a release changes sign-in, roles or access rules, alters a business flow that moves money, or adds API routes that take object identifiers. NIST SP 800-115 ties repeat application testing to significant modifications and to changes in the threat environment.