Skip to main content
Security guideReviewed

The OWASP Top 10:2025, explained for business owners

What each of the ten risk categories means for a website or web application, how a tester checks for it without harming your site, what a fix usually involves, and what the list can and cannot tell you when you buy a security test.

OWASP Top 10:2025 business-owner matrix

One row per 2025 category. Read across for what the category means for your business, a published advisory whose weakness class OWASP maps to that category, a safe test shown as an illustration on a lab or staging system, the evidence a tester should keep, and the check that confirms a fix.

How to use it

Every advisory was opened in the GitHub Advisory Database and in the vendor's own repository on 13 November 2026.

10 rows

OWASP Top 10:2025 business-owner matrix

CategoryWhat it means for your businessDated example and why it fitsSafe lab test (illustration)Evidence a tester keepsCheck that confirms the fix
A01:2025 Broken Access ControlA customer, a lower role or someone signed out can see or change data and actions meant for someone else.CVE-2025-29927 in Next.js (GHSA-f82v-jwr5-mffw), published 21 March 2025, critical, CVSS v3.1 9.1: authorisation checks made in middleware could be bypassed. Fixed in Next.js 12.3.5, 13.5.9, 14.2.25 and 15.2.3. The vendor advisory lists CWE-285, which the 2025 edition maps to A01.Illustration on staging: sign in as test account A, open one of A's records, then request the same address with test account B's record ID (WSTG-ATHZ-04). Repeat signed out and as a lower role against an admin page. Only test records are read.Each request and response, the account and role used, and the record IDs, marking any response that returned another account's data.The same requests on the fixed build return a refusal and no data, and the ownership check sits in server-side code next to the data, not only in middleware or the page.
A02:2025 Security MisconfigurationSoftware is set up insecurely: defaults left on, internal detail exposed, headers missing, or access granted more widely than intended.CVE-2026-54290 in the hono npm package (GHSA-88fw-hqm2-52qc), published 9 June 2026, high, CVSS v3.1 7.1: with credentials enabled and no origin set, its CORS middleware reflected any requesting origin, so a site a signed-in user visited could read cookie-authenticated endpoints. Fixed in hono 4.12.25. Both advisory pages list CWE-942, which the 2025 edition maps to A02.Illustration: send a test account's request to a signed-in API endpoint with an Origin header naming a site the tester controls, and read the CORS headers in the response (WSTG-CLNT-07). Nothing is changed; only headers are read.The request with its Origin header, and the Access-Control-Allow-Origin and Access-Control-Allow-Credentials values returned for it.The same request gets no permission for the untrusted origin while your own front end still works, and allowed origins are an explicit list in configuration.
A03:2025 Software Supply Chain FailuresA plugin, library or build step you did not write is vulnerable, unmaintained or compromised, and your site inherits the problem.CVE-2026-105642 in Ghost (GHSA-788w-68h3-cvxp), published 1 October 2026, high, CVSS v3.1 8.8: an image processing library bundled with Ghost had a flaw in its SVG handling that let any staff user, including Contributors, run commands on the Ghost server through a bookmark card. Both advisory pages list versions from 6.56.0 up to, but not including, 6.67.0 as affected, and 6.67.0 as patched. Both pages list CWE-1395, which the 2025 edition maps to A03, as well as CWE-94, which it maps to A05.Illustration: read installed versions from the lockfile, admin screen or build, and match each component against published advisories. Where a flaw runs code on the server, no exploit is sent; version and configuration are the evidence.A component inventory with versions, where each version was read from, and the advisories each one matches.The deployed build reports a patched version for every matched advisory, and the inventory, such as a software bill of materials, is kept current.
A04:2025 Cryptographic FailuresData is readable in transit or at rest, passwords can be cracked, or signed tokens can be forged, because cryptography is missing, weak or misused.CVE-2022-23540 in the jsonwebtoken npm package (GHSA-qwph-4952-7xr6), published 21 December 2022, moderate: in versions up to 8.5.1, calling jwt.verify() with no algorithm specified and a falsy secret or key could accept a token with no signature. Fixed in jsonwebtoken 9.0.0. The database entry lists CWE-327 and CWE-347, mapped to A04, and CWE-287, mapped to A07; the vendor page lists no CWE.Illustration: take a token issued to a test account, change one value in its payload or strip its signature, and send it back. Separately, read the TLS versions offered and the flags on session cookies.The original and altered tokens with secrets redacted, the server's response to each, and the TLS and cookie settings observed.Altered and unsigned tokens are refused, the accepted signing algorithm is set explicitly in code, and old TLS versions are switched off.
A05:2025 InjectionText typed into your site is run as part of a database query, script or command, which can expose or alter data or attack other users.CVE-2026-105845 in Payload (GHSA-v49j-62m6-pgrr), published 22 September 2026, critical, CVSS v3.1 9.8: SQL injection on SQLite and Postgres where untrusted users can query readable collections with dynamic filters or joins; MongoDB users are not affected. Fixed in Payload 3.88.0 and 4.0.0-canary.27. Both pages list CWE-89, which the 2025 edition maps to A05.Illustration on staging: send search, filter and form fields characters that change how a query reads, such as a single quote, and compare the responses and error messages (WSTG-INPV-05). Nothing that modifies or deletes data is sent.The parameter, the input sent, and the difference in response that shows the input changed the query.The same inputs are treated as plain text, queries use parameters or the framework's query builder, and a code search finds no query built by joining input into text.
A06:2025 Insecure DesignA business rule, such as a discount limit, booking cap or approval step, can be abused even though the code does what it was written to do.CVE-2026-31824 in Sylius (GHSA-7mp4-25j8-hp5q), published 9 March 2026, high, CVSS v3.1 8.2: a race condition in its promotion and coupon limits let a single-use promotion or coupon be redeemed an arbitrary number of times by completing several orders at once, with no authentication required; the advisory says this may lead to direct financial loss. Fixed in Sylius 1.9.12, 1.10.16, 1.11.17, 1.12.23, 1.13.15, 1.14.18, 2.0.16, 2.1.12 and 2.2.3. Both pages list CWE-362, which the 2025 edition maps to A06.Illustration on staging: create a test coupon limited to one use, prepare several carts with it, complete them at the same moment, and count the redemptions (WSTG-BUSL-05). Test orders only, never real payments.The coupon's limit, the timed requests, and the number of orders that received the discount.The repeated run records one redemption, the limit is enforced in the database with an atomic update or lock, and a test for the misuse case sits in the test suite.
A07:2025 Authentication FailuresSomeone is accepted as a user they are not, through guessed or reused passwords, a weak recovery flow, or a session that outlives logout.CVE-2026-114607 in Better Auth (GHSA-965c-763c-88jm), published 30 September 2026, critical, CVSS v3.1 9.1: with the Magic Link plugin and social or Generic OAuth sign-in configured as the advisory describes, an attacker who knew a user's email address could get a session for that user without access to the mailbox. The vendor advisory says better-auth 1.7.7 fixes it; on 13 November 2026 the database text still called that release planned. Both pages list CWE-287, which the 2025 edition maps to A07.Illustration on staging: with two test accounts, try each sign-in and recovery path, including password reset (WSTG-ATHN-09), magic links and social sign-in, to reach account B without its password or inbox. Lockout checks use a small, agreed number of attempts.Each path tried, the requests sent, and whether a session was issued for account B.Every path refuses the sign-in, sessions end on the server at logout, and multi-factor authentication applies wherever it is enabled.
A08:2025 Software or Data Integrity FailuresCode or data is trusted without being checked, such as unverified scripts or updates, or serialised data from the browser rebuilt on the server.CVE-2025-55182 in React Server Components (GHSA-fv66-9v8q-g76r), published 3 December 2025, critical, CVSS v3.1 10.0: unauthenticated remote code execution in versions 19.0.0, 19.1.0, 19.1.1 and 19.2.0 of the react-server-dom packages, fixed in 19.0.1, 19.1.2 and 19.2.1. CISA has added it to its Known Exploited Vulnerabilities catalogue, based on evidence of active exploitation. The vendor advisory lists CWE-502, which the 2025 edition maps to A08.Illustration: confirm from the deployed build which framework packages and versions are present, and list where the application loads scripts or accepts serialised data from outside. No exploit is sent, because the flaw runs code on the server.Package versions from the deployed build, each third-party script with its source, and any endpoint that accepts serialised objects.The build reports patched versions, external scripts come from sources you control or are verified, and serialised data from clients is refused unless it carries an integrity check.
A09:2025 Security Logging and Alerting FailuresAn attack happens and nobody finds out, or the logs themselves leak passwords, tokens or personal data.CVE-2024-32474 in self-hosted Sentry (GHSA-6cjm-4pxw-7xp9), published 18 April 2024, high, CVSS v3.1 7.3: a superuser's password was written to logs in cleartext when they signed in with a username and password, so anyone with access to the log data could use it. Fixed in Sentry 24.4.1; Sentry's hosted service was not affected. The vendor page lists CWE-532 and the database entry lists CWE-117 and CWE-312; the 2025 edition maps CWE-532 and CWE-117 to A09, and CWE-312 to A06.Illustration: agree a short series of failed logins, access denials and one admin action on test accounts, then ask your team what the logs recorded and who was alerted. OWASP notes this category is confirmed through interviews and samples of incident response more than by testing.The time of each agreed event, the matching log entries or their absence, any alert raised, and log samples checked for passwords, tokens and personal data.The same events appear in the logs with user and time, an alert reaches a named person, and the samples contain no secrets.
A10:2025 Mishandling of Exceptional ConditionsWhen something goes wrong, the site reveals internal detail, leaves a transaction half done, or lets people through a check that has failed.CVE-2026-73421 in Auth.js, the next-auth package (GHSA-8fpg-xm3f-6cx3), published 20 July 2026, critical in the GitHub Advisory Database, CVSS v4 9.1: when the Auth.js configuration produced a server error, apps that only checked whether the auth object existed treated every request as signed in, failing open. It affects next-auth 5.0.0-beta.0 to 5.0.0-beta.31, and both pages list 5.0.0-beta.32 as patched. Both pages list CWE-636, mapped to A10, and CWE-285, mapped to A01.Illustration on staging: send malformed, missing and oversized values to sign-in, payment and API endpoints and compare the error responses (WSTG-ERRH-01). With your team's agreement, briefly break a dependency such as the auth configuration and confirm protected pages still refuse access.Each malformed request, the error returned and any internal detail it exposed, and whether protected pages stayed protected when a component failed.Errors return a generic message with detail kept in the logs, protected routes deny access when the auth layer errors, and interrupted transactions roll back in full.

Category

What it means for your business

  1. A01:2025 Broken Access Control
    What it means for your business
    A customer, a lower role or someone signed out can see or change data and actions meant for someone else.
    Dated example and why it fits
    CVE-2025-29927 in Next.js (GHSA-f82v-jwr5-mffw), published 21 March 2025, critical, CVSS v3.1 9.1: authorisation checks made in middleware could be bypassed. Fixed in Next.js 12.3.5, 13.5.9, 14.2.25 and 15.2.3. The vendor advisory lists CWE-285, which the 2025 edition maps to A01.
    Safe lab test (illustration)
    Illustration on staging: sign in as test account A, open one of A's records, then request the same address with test account B's record ID (WSTG-ATHZ-04). Repeat signed out and as a lower role against an admin page. Only test records are read.
    Evidence a tester keeps
    Each request and response, the account and role used, and the record IDs, marking any response that returned another account's data.
    Check that confirms the fix
    The same requests on the fixed build return a refusal and no data, and the ownership check sits in server-side code next to the data, not only in middleware or the page.
  2. A02:2025 Security Misconfiguration
    What it means for your business
    Software is set up insecurely: defaults left on, internal detail exposed, headers missing, or access granted more widely than intended.
    Dated example and why it fits
    CVE-2026-54290 in the hono npm package (GHSA-88fw-hqm2-52qc), published 9 June 2026, high, CVSS v3.1 7.1: with credentials enabled and no origin set, its CORS middleware reflected any requesting origin, so a site a signed-in user visited could read cookie-authenticated endpoints. Fixed in hono 4.12.25. Both advisory pages list CWE-942, which the 2025 edition maps to A02.
    Safe lab test (illustration)
    Illustration: send a test account's request to a signed-in API endpoint with an Origin header naming a site the tester controls, and read the CORS headers in the response (WSTG-CLNT-07). Nothing is changed; only headers are read.
    Evidence a tester keeps
    The request with its Origin header, and the Access-Control-Allow-Origin and Access-Control-Allow-Credentials values returned for it.
    Check that confirms the fix
    The same request gets no permission for the untrusted origin while your own front end still works, and allowed origins are an explicit list in configuration.
  3. A03:2025 Software Supply Chain Failures
    What it means for your business
    A plugin, library or build step you did not write is vulnerable, unmaintained or compromised, and your site inherits the problem.
    Dated example and why it fits
    CVE-2026-105642 in Ghost (GHSA-788w-68h3-cvxp), published 1 October 2026, high, CVSS v3.1 8.8: an image processing library bundled with Ghost had a flaw in its SVG handling that let any staff user, including Contributors, run commands on the Ghost server through a bookmark card. Both advisory pages list versions from 6.56.0 up to, but not including, 6.67.0 as affected, and 6.67.0 as patched. Both pages list CWE-1395, which the 2025 edition maps to A03, as well as CWE-94, which it maps to A05.
    Safe lab test (illustration)
    Illustration: read installed versions from the lockfile, admin screen or build, and match each component against published advisories. Where a flaw runs code on the server, no exploit is sent; version and configuration are the evidence.
    Evidence a tester keeps
    A component inventory with versions, where each version was read from, and the advisories each one matches.
    Check that confirms the fix
    The deployed build reports a patched version for every matched advisory, and the inventory, such as a software bill of materials, is kept current.
  4. A04:2025 Cryptographic Failures
    What it means for your business
    Data is readable in transit or at rest, passwords can be cracked, or signed tokens can be forged, because cryptography is missing, weak or misused.
    Dated example and why it fits
    CVE-2022-23540 in the jsonwebtoken npm package (GHSA-qwph-4952-7xr6), published 21 December 2022, moderate: in versions up to 8.5.1, calling jwt.verify() with no algorithm specified and a falsy secret or key could accept a token with no signature. Fixed in jsonwebtoken 9.0.0. The database entry lists CWE-327 and CWE-347, mapped to A04, and CWE-287, mapped to A07; the vendor page lists no CWE.
    Safe lab test (illustration)
    Illustration: take a token issued to a test account, change one value in its payload or strip its signature, and send it back. Separately, read the TLS versions offered and the flags on session cookies.
    Evidence a tester keeps
    The original and altered tokens with secrets redacted, the server's response to each, and the TLS and cookie settings observed.
    Check that confirms the fix
    Altered and unsigned tokens are refused, the accepted signing algorithm is set explicitly in code, and old TLS versions are switched off.
  5. A05:2025 Injection
    What it means for your business
    Text typed into your site is run as part of a database query, script or command, which can expose or alter data or attack other users.
    Dated example and why it fits
    CVE-2026-105845 in Payload (GHSA-v49j-62m6-pgrr), published 22 September 2026, critical, CVSS v3.1 9.8: SQL injection on SQLite and Postgres where untrusted users can query readable collections with dynamic filters or joins; MongoDB users are not affected. Fixed in Payload 3.88.0 and 4.0.0-canary.27. Both pages list CWE-89, which the 2025 edition maps to A05.
    Safe lab test (illustration)
    Illustration on staging: send search, filter and form fields characters that change how a query reads, such as a single quote, and compare the responses and error messages (WSTG-INPV-05). Nothing that modifies or deletes data is sent.
    Evidence a tester keeps
    The parameter, the input sent, and the difference in response that shows the input changed the query.
    Check that confirms the fix
    The same inputs are treated as plain text, queries use parameters or the framework's query builder, and a code search finds no query built by joining input into text.
  6. A06:2025 Insecure Design
    What it means for your business
    A business rule, such as a discount limit, booking cap or approval step, can be abused even though the code does what it was written to do.
    Dated example and why it fits
    CVE-2026-31824 in Sylius (GHSA-7mp4-25j8-hp5q), published 9 March 2026, high, CVSS v3.1 8.2: a race condition in its promotion and coupon limits let a single-use promotion or coupon be redeemed an arbitrary number of times by completing several orders at once, with no authentication required; the advisory says this may lead to direct financial loss. Fixed in Sylius 1.9.12, 1.10.16, 1.11.17, 1.12.23, 1.13.15, 1.14.18, 2.0.16, 2.1.12 and 2.2.3. Both pages list CWE-362, which the 2025 edition maps to A06.
    Safe lab test (illustration)
    Illustration on staging: create a test coupon limited to one use, prepare several carts with it, complete them at the same moment, and count the redemptions (WSTG-BUSL-05). Test orders only, never real payments.
    Evidence a tester keeps
    The coupon's limit, the timed requests, and the number of orders that received the discount.
    Check that confirms the fix
    The repeated run records one redemption, the limit is enforced in the database with an atomic update or lock, and a test for the misuse case sits in the test suite.
  7. A07:2025 Authentication Failures
    What it means for your business
    Someone is accepted as a user they are not, through guessed or reused passwords, a weak recovery flow, or a session that outlives logout.
    Dated example and why it fits
    CVE-2026-114607 in Better Auth (GHSA-965c-763c-88jm), published 30 September 2026, critical, CVSS v3.1 9.1: with the Magic Link plugin and social or Generic OAuth sign-in configured as the advisory describes, an attacker who knew a user's email address could get a session for that user without access to the mailbox. The vendor advisory says better-auth 1.7.7 fixes it; on 13 November 2026 the database text still called that release planned. Both pages list CWE-287, which the 2025 edition maps to A07.
    Safe lab test (illustration)
    Illustration on staging: with two test accounts, try each sign-in and recovery path, including password reset (WSTG-ATHN-09), magic links and social sign-in, to reach account B without its password or inbox. Lockout checks use a small, agreed number of attempts.
    Evidence a tester keeps
    Each path tried, the requests sent, and whether a session was issued for account B.
    Check that confirms the fix
    Every path refuses the sign-in, sessions end on the server at logout, and multi-factor authentication applies wherever it is enabled.
  8. A08:2025 Software or Data Integrity Failures
    What it means for your business
    Code or data is trusted without being checked, such as unverified scripts or updates, or serialised data from the browser rebuilt on the server.
    Dated example and why it fits
    CVE-2025-55182 in React Server Components (GHSA-fv66-9v8q-g76r), published 3 December 2025, critical, CVSS v3.1 10.0: unauthenticated remote code execution in versions 19.0.0, 19.1.0, 19.1.1 and 19.2.0 of the react-server-dom packages, fixed in 19.0.1, 19.1.2 and 19.2.1. CISA has added it to its Known Exploited Vulnerabilities catalogue, based on evidence of active exploitation. The vendor advisory lists CWE-502, which the 2025 edition maps to A08.
    Safe lab test (illustration)
    Illustration: confirm from the deployed build which framework packages and versions are present, and list where the application loads scripts or accepts serialised data from outside. No exploit is sent, because the flaw runs code on the server.
    Evidence a tester keeps
    Package versions from the deployed build, each third-party script with its source, and any endpoint that accepts serialised objects.
    Check that confirms the fix
    The build reports patched versions, external scripts come from sources you control or are verified, and serialised data from clients is refused unless it carries an integrity check.
  9. A09:2025 Security Logging and Alerting Failures
    What it means for your business
    An attack happens and nobody finds out, or the logs themselves leak passwords, tokens or personal data.
    Dated example and why it fits
    CVE-2024-32474 in self-hosted Sentry (GHSA-6cjm-4pxw-7xp9), published 18 April 2024, high, CVSS v3.1 7.3: a superuser's password was written to logs in cleartext when they signed in with a username and password, so anyone with access to the log data could use it. Fixed in Sentry 24.4.1; Sentry's hosted service was not affected. The vendor page lists CWE-532 and the database entry lists CWE-117 and CWE-312; the 2025 edition maps CWE-532 and CWE-117 to A09, and CWE-312 to A06.
    Safe lab test (illustration)
    Illustration: agree a short series of failed logins, access denials and one admin action on test accounts, then ask your team what the logs recorded and who was alerted. OWASP notes this category is confirmed through interviews and samples of incident response more than by testing.
    Evidence a tester keeps
    The time of each agreed event, the matching log entries or their absence, any alert raised, and log samples checked for passwords, tokens and personal data.
    Check that confirms the fix
    The same events appear in the logs with user and time, an alert reaches a named person, and the samples contain no secrets.
  10. A10:2025 Mishandling of Exceptional Conditions
    What it means for your business
    When something goes wrong, the site reveals internal detail, leaves a transaction half done, or lets people through a check that has failed.
    Dated example and why it fits
    CVE-2026-73421 in Auth.js, the next-auth package (GHSA-8fpg-xm3f-6cx3), published 20 July 2026, critical in the GitHub Advisory Database, CVSS v4 9.1: when the Auth.js configuration produced a server error, apps that only checked whether the auth object existed treated every request as signed in, failing open. It affects next-auth 5.0.0-beta.0 to 5.0.0-beta.31, and both pages list 5.0.0-beta.32 as patched. Both pages list CWE-636, mapped to A10, and CWE-285, mapped to A01.
    Safe lab test (illustration)
    Illustration on staging: send malformed, missing and oversized values to sign-in, payment and API endpoints and compare the error responses (WSTG-ERRH-01). With your team's agreement, briefly break a dependency such as the auth configuration and confirm protected pages still refuse access.
    Evidence a tester keeps
    Each malformed request, the error returned and any internal detail it exposed, and whether protected pages stayed protected when a component failed.
    Check that confirms the fix
    Errors return a generic message with detail kept in the logs, protected routes deny access when the auth layer errors, and interrupted transactions roll back in full.

What the OWASP Top 10 is, in OWASP's own words

The OWASP Top 10 is a ranked list of ten categories of web application security risk, published by the OWASP Foundation.

In OWASP's words it is "a standard awareness document for developers and web application security" that represents "a broad consensus about the most critical security risks to web applications". The current edition, the OWASP Top 10:2025, is the eighth since the list began in 2003.

What it is not matters as much. OWASP calls the Top 10 "primarily an awareness document" and says that anyone using it as a coding or testing standard should know "it is the bare minimum and just a starting point".

Treat it as a list of risks to understand, not a standard your site can pass. There is no OWASP certificate behind it either: in its Application Security Verification Standard, OWASP states that it does not certify any vendors, verifiers or software.

A proposal or badge that calls a site "OWASP Top 10 certified" is describing something OWASP does not issue.

OWASP also warns against the opposite overreach. Its guidance says tools cannot comprehensively detect, test or protect against the Top 10, and that it discourages any claim of full Top 10 coverage "because it's simply untrue". A test can show which categories it examined, how, and what it found in each. It cannot show that a category of weakness is absent from your application.

What changed in the OWASP Top 10 2025

The 2025 edition reworks the 2021 list rather than replacing it.

OWASP describes two new categories and one consolidation, and says it focused on root causes over symptoms wherever it could. In order of the new ranking:

  • A01:2025 Broken Access Control stays first, and Server-Side Request Forgery (SSRF) has been rolled into it.
  • A02:2025 Security Misconfiguration rises from fifth to second, which OWASP links to how much of an application's behaviour now depends on configuration.
  • A03:2025 Software Supply Chain Failures expands the 2021 category Vulnerable and Outdated Components to cover compromises across dependencies, build systems and distribution infrastructure.
  • A04:2025 Cryptographic Failures, A05:2025 Injection and A06:2025 Insecure Design each fall two places.
  • A07:2025 Authentication Failures keeps seventh place under a shorter name, previously Identification and Authentication Failures.
  • A08:2025 Software or Data Integrity Failures stays eighth.
  • A09:2025 Security Logging and Alerting Failures was Security Logging and Monitoring Failures. The new name stresses alerting, because, as OWASP puts it, great logging with no alerting is of minimal value in identifying security incidents.
  • A10:2025 Mishandling of Exceptional Conditions is new: improper error handling, logic errors and systems that fail open when something goes wrong.

How the ranking was built

The list is data-informed rather than purely data-driven. Contributors supplied testing data for over 2.8 million applications, and eight of the ten categories come from that data. The other two come from a community survey of practitioners, because, in OWASP's words, examining contributed data is essentially looking into the past.

Each category groups many related weaknesses, 248 CWEs across the ten, so a category name describes a family of problems rather than a single bug. OWASP adds that it is basically impossible to create ten categories without some overlap, which is why one flaw can sit in two of them.

The ten OWASP Top 10 categories in plain terms

Each entry gives the category's exact number and name, what it means for a website or web application, and what fixing it usually involves, based on the prevention guidance on OWASP's page for that category.

The matrix on this page adds a dated example, a safe test and the evidence for each.

A01:2025 Broken Access Control

Access control decides who may see or change what. 1 min read

A02:2025 Security Misconfiguration

Misconfiguration is software set up insecurely: default admin accounts left active, features and sample pages that should have been removed, error pages that print internal detail, missing security headers, or cloud storage shared more widely than intended. 1 min read

A03:2025 Software Supply Chain Failures

This covers the code you did not write: frameworks, libraries, plugins, themes and the build pipeline that assembles and ships your site. 1 min read

A04:2025 Cryptographic Failures

This is about protecting data with cryptography and getting the details right: encryption in transit and at rest, password storage, key handling and signature checks. 1 min read

A05:2025 Injection

Injection happens when input from a user reaches an interpreter, such as a database, a browser or the server's command line, and is run as part of a command. 1 min read

A06:2025 Insecure Design

Insecure design is a control that was never designed in, as opposed to one that was built badly. 1 min read

A07:2025 Authentication Failures

Authentication failures let someone be accepted as a user they are not. 1 min read

A08:2025 Software or Data Integrity Failures

Integrity failures happen when code or data is trusted without being checked: scripts loaded from a source you do not control, updates applied without a signature check, a build pipeline that pulls unverified artefacts, or an application that accepts serialised objects from the browser and rebuilds them on the server. 1 min read

A09:2025 Security Logging and Alerting Failures

This category is about finding out. 1 min read

A10:2025 Mishandling of Exceptional Conditions

The new category covers what an application does when something unexpected happens. 1 min read

How a tester checks each category without harming your site

A good test proves a weakness with the least force that proves it.

On a system your customers depend on, that rule shapes how each category is approached.

  • Test accounts and test records stand in for real customers. Access control (A01) is checked with two accounts per role, each asking for the other's records, so the only data read or changed is test data.
  • Staging comes first for anything that pushes the application into an unusual state. Business logic (A06) and error handling (A10) tests do exactly that, so they run on a staging copy that matches production where one exists, and otherwise in an agreed window.
  • Version evidence replaces exploits where a known flaw runs code on the server, as several A03 and A08 advisories do. The tester confirms the version and configuration in the deployed build instead of sending the attack.
  • Injection (A05) is probed first with input that changes how a query reads, never with input that deletes or overwrites data.
  • Lockout and rate-limit checks under A07 and A10 use a small, agreed number of requests. Clever Ops does not run load or denial-of-service testing.
  • Every target has written authorisation from its owner, and a technical contact can pause testing if anything looks wrong.

What testing reaches, and what needs code or conversation

OWASP is candid that the Top 10 documents risks rather than neatly testable issues. It says Insecure Design is beyond the scope of most forms of testing, and that whether logging and monitoring work can only be confirmed through interviews and by sampling real incident responses; a penetration tester may only learn that they triggered incident response in a test environment. A useful report therefore says how each category was covered, not only whether something was found.

  • A01, A02, A05 and A07 show up in requests and responses, so testing the running application as each role reaches them well.
  • A03 and A08 depend on knowing exactly what is installed and loaded, which is quickest from the lockfile, plugin list or build.
  • A04 is partly visible from outside, such as TLS and cookie settings, and partly only in code, such as how passwords are hashed and where keys live.
  • A06 needs the business rules written down so the tester can try to break them, and read-only code access helps.
  • A09 needs agreed test events and a conversation with whoever reads the logs.
  • A10 is tested from outside for error output and fail-open behaviour, and in code for how interrupted transactions roll back.

How the matrix pairs each category with a real advisory

Every row in the matrix names a published product vulnerability rather than a breach at a named business, because an advisory states the affected versions, the fix and the class of weakness, which a breach report rarely does.

A row's category follows from that class. Each advisory lists one or more CWE weaknesses, and OWASP's page for each category lists the CWEs it maps there. Where an advisory lists CWEs from two categories, the row says so, and the overlap is real: OWASP lists CORS misconfiguration as an access control problem under A01, while the CORS weakness in the A02 row, CWE-942, is mapped to A02.

Each advisory was opened on 13 November 2026 in two places: the GitHub Advisory Database entry and the vendor's own advisory in its code repository. The two do not always agree. For CVE-2026-114607 in Better Auth, the database entry's text still described the fixed release as planned and not yet available, while the vendor's advisory said to upgrade to it.

The jsonwebtoken database entry lists three CWEs and the vendor's page lists none, and the Sentry entry lists different CWEs in each place.

Where they differ, the row says which page states what. Before relying on any advisory for your own systems, open both pages yourself, because both can change after publication.

The safe test column is an illustration of method on a lab or staging system. It is not a record of testing any of these products, and the products named are examples of each category, not a list of software to avoid.

What the Top 10 tells you when you buy a test, and what it does not

For a buyer, the Top 10 is most useful as a shared vocabulary.

A finding labelled A01 tells a non-technical reader the family of problem at a glance, and category labels make it easier to answer customer security questionnaires that ask about specific types of risk. Clever Ops maps each finding in a report to its OWASP Top 10:2025 category for that reason.

What the list cannot tell you is just as important:

  • What was in scope. "Tested against the OWASP Top 10" says nothing about which pages, roles, APIs or environments were covered. Ask for the scope in writing.
  • That nothing else is wrong. OWASP calls its Top Ten lists awareness documents that are not meant to be a complete list, only a starting place, and it discourages claims of full coverage.
  • Your priorities. The ranking reflects data from many applications. Your order of risk depends on what your application does with money, accounts and personal data.
  • Certification. OWASP does not certify software. If a tender or customer requires an accreditation such as CREST, that is a separate requirement: Clever Ops is not CREST accredited, so we refer that work to a provider that is.
  • AI features. Prompt injection against chatbots and agents sits in OWASP's separate LLM Top 10, and needs a test designed for it.

Questions to ask a provider about Top 10 coverage

These questions turn a vague promise into a scope you can compare between providers.

  • Which categories will you test directly, which will you assess from code or configuration, and which will you only ask us about?
  • How will you test access between our user roles, and between separate customer accounts?
  • Will the report map each finding to a 2025 category and say how each category was covered, including where it was not?
  • Will you confirm component versions from our deployed build, and which advisory sources will you check them against?
  • What will you do to avoid changing real data or affecting real customers while testing?

The Top 10, the OWASP Testing Guide and ASVS

OWASP publishes three documents that buyers often mix up.

The Top 10 names the risks. The Web Security Testing Guide, whose current stable release is version 4.2, describes how to test, one test at a time, each with an ID such as WSTG-ATHZ-04 for insecure direct object references. The Application Security Verification Standard, version 5.0.0 since May 2025, sets out requirements an application can be verified against.

OWASP's own table of use cases makes the division plain. For penetration testing, secure code review and coding standards it marks the Top 10 as "bare minimum" and the ASVS as "yes", and it encourages anyone adopting an application security standard to use the ASVS because it is designed to be verifiable and tested. A test plan that cites the Top 10 alone is naming its labels, not its method.

Clever Ops reports reference WSTG and ASVS 5.0 coverage to describe what was checked, and use Top 10:2025 categories as labels for what was found. That describes method and coverage. It is not a certificate or a compliance opinion.

An OWASP Top 10 checklist for your next developer meeting

Ten questions, one per category, that a business owner can ask without technical knowledge.

A specific, confident answer is a good sign. A vague one shows where a test should look first.

  • A01:If I change the number in the address bar on my account page, could I see another customer's account? How do we know?
  • A02:Which default accounts, sample pages and debug settings were removed before launch, and what checks that they stay removed?
  • A03:Do we have a current list of every plugin, library and component we run, and who reads the advisories for them?
  • A04:How are customer passwords stored, and are any keys or secrets kept in our code repository?
  • A05:Does every database query use parameters or the framework's query builder, or is any query built by joining text together?
  • A06:Which rules protect our money, such as coupon limits, refunds and free trials, and has anyone tried to break them?
  • A07:Do staff and admin accounts use multi-factor authentication, and what stops someone trying thousands of passwords?
  • A08:Which outside scripts does the site load, and is anything we install or deploy checked for tampering?
  • A09:If someone tried to break into an admin account tonight, who would be told, and how?
  • A10:What does a customer see when the site hits an error, and what happens to a payment that fails halfway?

Where a Clever Ops test fits

A web application and API penetration test is the service that covers the Top 10 most directly: signed-in testing as each role, the API called directly rather than only through the screens, and business rules such as payments and discounts.

A website security check covers the outside of a site, which mostly touches misconfiguration and visible component versions. Read-only code access can be added for the parts testing reaches poorly, such as password storage, transaction handling and how errors are caught.

If Clever Ops built or hosts your application, the work is a security review rather than an independent test, and the report states who built and who hosts it. If a customer or auditor needs an independent tester, choose a provider with no part in either.

Frequently asked questions

The 2025 categories are A01 Broken Access Control, A02 Security Misconfiguration, A03 Software Supply Chain Failures, A04 Cryptographic Failures, A05 Injection, A06 Insecure Design, A07 Authentication Failures, A08 Software or Data Integrity Failures, A09 Security Logging and Alerting Failures and A10 Mishandling of Exceptional Conditions. Each is a family of related weaknesses rather than a single vulnerability.

No. OWASP describes it as primarily an awareness document, and says that anyone using it as a coding or testing standard should treat it as the bare minimum and a starting point. OWASP does not certify software, and for anyone adopting a standard it recommends the Application Security Verification Standard, which is designed to be verified and tested.

The 2025 edition is the eighth since the list began in 2003, and it follows the 2021 edition. When a report, contract or questionnaire mentions the Top 10, check which edition it means, because category numbers and names changed between 2021 and 2025.

The Top 10 is a short list of risk categories meant to raise awareness. The Web Security Testing Guide, version 4.2, describes how to test a web application, with an ID for each test. A good report uses the Top 10 to label findings and the Testing Guide, with ASVS, to describe what was checked.

Only in part. OWASP says tools cannot comprehensively detect, test or protect against the Top 10 because of the nature of several of its risks, and points to Insecure Design as an example. A scanner is useful for known issues and visible misconfiguration. Access control between users, business logic and logging need a person.

Not directly. OWASP covers prompt injection and related AI risks in its separate LLM Top 10, which the 2025 Injection page points to. If your website or app includes a chatbot or an AI agent, an AI chatbot and agent security test covers the bot, the data it can read and the actions it can take.

Start with A01 Broken Access Control if customers log in: it ranks first again in 2025, and it covers whether one customer can reach another customer's data. If you run an online store, add A06 for discounts and refunds. If your site runs on plugins, start with A03. The checklist on this page helps you decide.

Scope the right test for your application

Tell us what you have built, who uses it and when it needs to launch. You will get a plain-English recommendation: a website audit, a web application test or a referral to a specialist.