Back to glossary

OWASP Top 10

The OWASP Top 10 is OWASP's ranked awareness list of the ten most critical web application security risk categories, revised every few years.

Also known as OWASP Top Ten, OWASP Top 10:2025, OWASP Top 10 2025, OWASP Top 10 2021, Open Worldwide Application Security Project Top 10, Open Web Application Security Project Top 10

The OWASP Top 10, also written OWASP Top Ten, is a ranked list of the ten most critical web application security risk categories, published by OWASP since 2003. Each category groups related weaknesses and is ranked by how often testers find them and how severe their CVEs are. It is an awareness document, not a testable standard.

As of November 2026, the current edition is the OWASP Top 10:2025, the eighth since 2003. OWASP published it as a release candidate on November 6, 2025, and its maintainers removed the release-candidate notice on December 24, 2025. The headline change is A03:2025 Software Supply Chain Failures, which replaces the old "vulnerable and outdated components" category and ranks third. A03 is also where the list's method breaks. In OWASP's own data, A03 has the highest average incidence, exploit, and impact scores of all ten categories, yet only 11 CVEs and the lowest test coverage. And the malicious packages that define modern supply chain attacks are classified under CWE-506, which OWASP files under A08, not A03.

Our view: the OWASP Top 10 is a good shared vocabulary and a poor test plan. Use it to decide what developers should learn first. Do not treat a tool's "OWASP Top 10 coverage" as evidence of protection against supply chain attacks. The list's data comes from testing your code, and supply chain attacks happen in someone else's release process.

What is the OWASP Top 10?

#

The OWASP Top 10 is a ranking of risk categories that OWASP volunteers revise every few years. Editions shipped in 2003, 2004, 2007, 2010, 2013, 2017, 2021, and 2025. Each category is versioned by year, as in A01:2025, so a finding labeled A06 means different things in 2021 and 2025.

Each category is a bucket of Common Weakness Enumeration (CWE) entries, not a single bug. A05:2025 Injection, for example, maps 37 CWEs, from cross-site scripting to SQL injection. The 2025 edition places 248 CWEs in its ten categories, out of 968 in MITRE's dictionary at the time.

OWASP explains why it uses buckets instead of ten individual CWEs, as MITRE's CWE Top 25 does. Not every CWE exists in every language, and different testers file the same flaw under different CWEs. The trade-off is that a category is only as accurate as its CWE mapping. That mapping decides where supply chain malware lands, as we show below.

The OWASP Top 10 cannot be passed or failed. The 2025 text says it directly: "The OWASP Top 10 is primarily an awareness document." It names risks, not requirements a tester can check.

2025 rankCategoryCWEs mappedIn the 2021 edition
A01Broken Access Control40A01, now also absorbs A10:2021 SSRF
A02Security Misconfiguration16A05
A03Software Supply Chain Failures6A06 Vulnerable and Outdated Components, expanded
A04Cryptographic Failures32A02
A05Injection37A03
A06Insecure Design39A04
A07Authentication Failures36A07 Identification and Authentication Failures
A08Software or Data Integrity Failures14A08 Software and Data Integrity Failures
A09Security Logging & Alerting Failures5A09 Security Logging and Monitoring Failures
A10Mishandling of Exceptional Conditions24New in 2025

What changed in the OWASP Top 10 2025?

#

The OWASP Top 10 2025 adds two categories, merges one, and reorders five. Measured against the 2021 edition, the changes are:

  • SSRF merged into A01. Server-side request forgery was its own category in 2021, A10, with a single CWE. It now sits inside A01:2025 Broken Access Control, which stays first.
  • A03 Software Supply Chain Failures is new. It expands A06:2021 Vulnerable and Outdated Components from known-vulnerable libraries to compromises anywhere in dependencies, build systems, and distribution infrastructure. It moved from sixth to third.
  • A10 Mishandling of Exceptional Conditions is new. Its 24 CWEs cover improper error handling, logic errors, and failing open.
  • Security Misconfiguration rose from fifth to second. OWASP found that 3.00% of tested applications had at least one of its 16 CWEs, and attributes the rise to more application behavior moving into configuration.
  • Three categories dropped two places each. Cryptographic Failures fell to fourth, Injection to fifth, and Insecure Design to sixth.
  • Three categories were renamed. A07 is now Authentication Failures. A09 is now Security Logging & Alerting Failures, to stress that logs nobody alerts on are of little use. A08 changed "and" to "or" and is now Software or Data Integrity Failures.

Three more risks missed the cut and appear in a Next Steps appendix. They are Lack of Application Resilience (the renamed Denial of Service), Memory Management Failures, and Inappropriate Trust in AI Generated Code, which OWASP subtitles "Vibe Coding." The AI category has no score table, because there is no testing data for it yet.

How does OWASP rank the Top 10 categories?

#

OWASP ranks Top 10 categories with two datasets, testing results and CVE severity, and then lets a practitioner survey promote two of them.

Testing data. For 2025, a dozen named organizations, plus anonymous donors, contributed results for over 2.8 million applications. For each year since 2021, contributors reported how many applications they tested and how many had at least one instance of each CWE. That yields two numbers per CWE. Incidence rate is the share of tested applications with the weakness. Coverage is the share of applications that were tested for it at all. Frequency is ignored on purpose, so an application with 4 instances counts the same as one with 4,000. The 2025 data contained 589 distinct CWEs, up from almost 400 in 2021.

Severity data. OWASP extracted roughly 175,000 CVE-to-CWE records from the National Vulnerability Database, using OWASP Dependency-Check as the export tool. It then averaged the exploitability and impact sub-scores of the Common Vulnerability Scoring System (CVSS) for each CWE, blending CVSS v2 and v3. It skipped CVSS v4.0, because v4.0 no longer exposes separate exploit and impact scores.

Survey. OWASP takes eight categories from the data and two from a community survey. The reason is in its methodology: "examining the contributed data is essentially looking into the past." OWASP does not say which two categories the survey promoted. The A09 page says it was voted in for the third time. The A03 page says exactly 50% of respondents ranked it first.

The severity half of this method depends on NVD enrichment, which is shrinking. In April 2026, NIST stopped enriching most CVEs and now enriches only those in CISA's KEV catalog, in software the federal government uses, or in critical software. The next edition will have fewer scored CVEs to average.

Every score is published as Markdown in OWASP's repository, so you can pull the per-category numbers yourself:

Python
import glob, re
for f in sorted(glob.glob("A*_2025-*.md")):
    t = open(f).read()
    table = re.search(r"## Score table.*?</table>", t, re.S).group(0)
    c = [re.sub(r"<[^>]+>|\s+", " ", x).strip() for x in re.findall(r"<td>(.*?)</td>", table, re.S)]
    row = dict(zip(c[:len(c)//2], c[len(c)//2:]))
    print(f[:3], *(row[k].rjust(7) for k in ["Avg Incidence Rate", "Avg Coverage",
          "Avg Weighted Exploit", "Avg Weighted Impact", "Total CVEs"]))
Bash
$ git clone --depth 1 https://github.com/OWASP/Top10.git
$ cd Top10/2025/docs/en && python3 scores.py
A01   3.74%  42.93%    7.04    3.84  32,654
A02   3.00%  52.35%    7.96    3.97   1,375
A03   5.72%  27.47%    8.17    5.23      11
A04   3.80%  47.74%    7.23    3.90   2,185
A05   3.08%  42.93%    7.15    4.32  62,445
A06   1.86%  35.18%    6.96    4.05   7,647
A07   2.92%  37.14%    7.69    4.44   7,147
A08   2.75%  45.49%    7.11    4.79   3,331
A09   3.91%  46.48%    7.19    2.65     723
A10   2.95%  37.95%    7.11    3.81   3,416
CategoryAvg incidenceAvg coverageAvg exploitAvg impactCVEsOccurrences
A01 Broken Access Control3.74%42.93%7.043.8432,6541,839,701
A02 Security Misconfiguration3.00%52.35%7.963.971,375719,084
A03 Software Supply Chain Failures5.72%27.47%8.175.2311215,248
A04 Cryptographic Failures3.80%47.74%7.233.902,1851,665,348
A05 Injection3.08%42.93%7.154.3262,4451,404,249
A06 Insecure Design1.86%35.18%6.964.057,647729,882
A07 Authentication Failures2.92%37.14%7.694.447,1471,120,673
A08 Software or Data Integrity Failures2.75%45.49%7.114.793,331501,327
A09 Security Logging & Alerting Failures3.91%46.48%7.192.65723260,288
A10 Mishandling of Exceptional Conditions2.95%37.95%7.113.813,416769,581

The figure above plots five of these columns for all ten categories. A03 has the highest average incidence (5.72%), weighted exploit (8.17), and weighted impact (5.23) of any category. It has the lowest average coverage (27.47%) and the fewest mapped CVEs (11). The next-lowest CVE count belongs to A09, with 723.

A03 tops the risk metrics on 11 CVEs

#

Software Supply Chain Failures is the outlier in the OWASP Top 10 data. It ranks first of ten on incidence, exploit, and impact. It ranks last on coverage, CVE count, and total occurrences, with 215,248 occurrences against at least 260,288 for every other category.

Read together, those numbers say that most testers never looked. On average, about 73% of applications in the dataset were never tested for a given A03 weakness. Where someone did look, they found a problem more often than in any other category.

The A03 exploit and impact scores rest on almost no data. They average 11 CVEs, against 62,445 for Injection and 32,654 for Broken Access Control. In 2021, the same category had zero CVEs, and OWASP substituted a default weight of 5.0 for both scores. Going from 0 to 11 CVEs moved the exploit score from 5.00 to 8.17. That is not a measured trend. It is a small sample that happens to point in the right direction.

OWASP's A03 page also disagrees with itself on the headline number. Its prose gives an average incidence rate of 5.19%, while its score table gives 5.72%. Our coverage of the OWASP Top 10:2025 release candidate quoted the 5.19% figure. We use the table here, because every other number in this article comes from the tables.

What the six A03 CWEs describe matters more than their scores:

  • CWE-477: Use of Obsolete Function
  • CWE-1035: a pointer to A9 of the 2017 Top 10, Using Components with Known Vulnerabilities
  • CWE-1104: Use of Unmaintained Third Party Components
  • CWE-1329: Reliance on Component That is Not Updateable
  • CWE-1357: Reliance on Insufficiently Trustworthy Component
  • CWE-1395: Dependency on Vulnerable Third-Party Component

Every one is a property of a component you chose: old, unmaintained, unpatchable, untrustworthy, or known to be vulnerable. CWE-1357 comes closest, but it describes a component whose trustworthiness you failed to verify, not one that attacks you. None describes a component that is actively hostile. The category's own example attacks are hostile ones: SolarWinds, the Bybit theft, and Shai-Hulud.

Where does malware land in the OWASP Top 10?

#

Malware in an open source package lands in A08:2025 Software or Data Integrity Failures under the OWASP Top 10's own CWE mapping, not in A03. A08 maps 14 CWEs, including CWE-506 (Embedded Malicious Code) and CWE-509 (Replicating Malicious Code, Virus or Worm). OWASP describes A08 as integrity failures "at a lower level than Software Supply Chain Failures."

Advisory databases classify malicious packages as CWE-506. We checked eight malicious npm versions from seven incidents spanning eight years: the event-stream backdoor, the ua-parser-js hijack, the Prettier tooling compromise, the is package hijack, the Qix account compromise that hit chalk and debug, the first Shai-Hulud wave through @ctrl/tinycolor, and the axios compromise. On November 4, 2026, we queried the npm registry's bulk advisory endpoint, the API that npm audit calls:

Bash
$ curl -s https://registry.npmjs.org/-/npm/v1/security/advisories/bulk \
    -H 'content-type: application/json' \
    -d '{"event-stream":["3.3.6"],"ua-parser-js":["0.7.29"],
         "eslint-config-prettier":["8.10.1"],"is":["3.3.1"],
         "chalk":["5.6.1"],"debug":["4.4.2"],
         "@ctrl/tinycolor":["4.1.1"],"axios":["1.14.1"]}' \
  | jq -r 'to_entries[] | .key as $p | .value[]
      | select(.cwe | index("CWE-506"))
      | [$p, (.url | split("/")[-1]), (.cwe | join(",")), .cvss.score] | @tsv'
event-stream            GHSA-mh6f-8j2x-4483  CWE-506          9.8
ua-parser-js            GHSA-pjwm-rvh2-c87w  CWE-506,CWE-912  8.8
eslint-config-prettier  GHSA-f29h-pxvx-f335  CWE-506          7.5
eslint-config-prettier  GHSA-r568-chxc-crfm  CWE-506          0
is                      GHSA-6f4g-3jvw-4j45  CWE-506          0
chalk                   GHSA-2v46-p5h4-248w  CWE-506          0
debug                   GHSA-4x49-vf9v-38px  CWE-506          0
@ctrl/tinycolor         GHSA-qjqf-7j6f-82c4  CWE-506          0
axios                   GHSA-fw8c-xr5c-95f9  CWE-506          0
IncidentMalicious versionAdvisoryCWECVSS scoreOWASP 2025 category by CWE
event-stream backdoor (2018)event-stream@3.3.6GHSA-mh6f-8j2x-4483CWE-5069.8A08
ua-parser-js hijack (2021)ua-parser-js@0.7.29GHSA-pjwm-rvh2-c87wCWE-506, CWE-9128.8A08 (CWE-912 maps to none)
Prettier tooling compromise (July 2025)eslint-config-prettier@8.10.1GHSA-f29h-pxvx-f335 (CVE-2025-54313), GHSA-r568-chxc-crfmCWE-5067.5 and 0A08
is hijack (July 2025)is@3.3.1GHSA-6f4g-3jvw-4j45CWE-5060A08
Qix account compromise (September 2025)chalk@5.6.1GHSA-2v46-p5h4-248wCWE-5060A08
Qix account compromise (September 2025)debug@4.4.2GHSA-4x49-vf9v-38pxCWE-5060A08
Shai-Hulud, first wave (September 2025)@ctrl/tinycolor@4.1.1GHSA-qjqf-7j6f-82c4CWE-5060A08
axios compromise (March 2026)axios@1.14.1GHSA-fw8c-xr5c-95f9CWE-5060A08

Every record carries CWE-506, and none carries any of A03's six CWEs. The second CWE on the ua-parser-js record, CWE-912 (Hidden Functionality), is not mapped to any 2025 category. Three consequences follow.

A03 stays empty. A dashboard that rolls findings up by OWASP category files all eight under A08. That includes the Shai-Hulud version, the attack OWASP itself uses as A03's third example scenario.

CVSS sorts malware last. Five of the eight versions have only records with a CVSS score of 0 and no vector string. A queue sorted by score puts them at the bottom.

Most malware has no CVE. Of the five GitHub advisory pages we opened, only one lists a CVE: CVE-2025-54313, for the Prettier tooling compromise. The axios, chalk, @ctrl/tinycolor, and event-stream advisories list none. Malware does not reliably get a CVE, so a scanner that matches dependencies to CVEs misses most of it, and so do the CVE counts that feed OWASP's severity scores.

All of these signals exist only after someone detects the malware and files an advisory. The attacks themselves ran earlier. Shai-Hulud and the axios compromise both executed from a postinstall script, at install time, before any test ran. Default-on lifecycle scripts were a design mistake for exactly this reason. pnpm 10 stopped running dependency lifecycle scripts by default in January 2025, and npm v12 did the same in July 2026. Those defaults remove an execution path, not the stolen credential behind it. As we wrote when npm v12 shipped, recent worm waves ran from compromised maintainer accounts publishing through legitimate pipelines, "some with valid provenance." The chalk payload shows the other gap: it rewrote cryptocurrency transactions in the browser at runtime, so no install-script policy would have stopped it.

OWASP Top 10 vs ASVS vs CWE Top 25

#

The OWASP Top 10 is the right tool for training and prioritization, and the wrong tool for verification. OWASP says so itself, in its guidance on using the Top 10 as a standard: "OWASP discourages any claims of full coverage of the OWASP Top 10, because it's simply untrue." For requirements a tester can check, OWASP points to the Application Security Verification Standard (ASVS), whose 5.0.0 release shipped in May 2025. Teams choosing a reference list usually weigh those two against MITRE's CWE Top 25, which ranks individual weaknesses from CVE records mapped to CWEs.

ListWhat it isUse it forWhat it misses
OWASP Top 10:2025Ranked awareness list: 10 risk categories covering 248 CWEsDeveloper training, setting priorities, a shared vocabulary with auditorsCannot be passed or failed. A03 measures stale components, and malware lands in A08.
OWASP ASVS 5.0Verification standard: testable requirements grouped by verification levelSecurity requirements, penetration test scope, procurement languageVerifies your application, not what a dependency does when it installs
MITRE CWE Top 25Ranked list of 25 individual weaknesses, built from CVE recordsSecure coding rules per language, tuning static analysisInherits the CVE pipeline's blind spot: malware rarely gets a CVE

Verdict: teach with the OWASP Top 10, specify and test with ASVS, and tune code scanners with the CWE Top 25. For dependencies, none of the three is sufficient on its own, because all three describe weaknesses in code, and a malicious package is working exactly as its author intended. Add a control that inspects what a package does before it runs.

Socket Firewall is one such control. Running sfw npm install routes the package manager through a local proxy that checks every requested package, including transitive dependencies, against Socket's analysis before download, as its documentation describes. The free tier blocks confirmed malware and only warns on AI-detected threats, so a malicious version published minutes ago can install before anyone has classified it. Pre-install inspection narrows the window between publish and detection. It does not close it.

What the next OWASP Top 10 should measure

#

A03's numbers are what you get when you measure supply chain risk with application testing. A tester scanning your code finds an unmaintained library. No test of your code finds a maintainer token that was phished last Tuesday, because that event happened in someone else's release process.

The data to measure supply chain compromise already exists. GitHub's advisory database files malware as its own advisory type, separate from reviewed vulnerabilities. The OpenSSF malicious-packages repository publishes reports in the OSV format. A data call that asked contributors how many applications resolved a package version later flagged as malicious would give A03 a real incidence rate. Cross-listing CWE-506 and CWE-509 under A03 would let the dashboards built on OWASP's mapping show it. Until then, the category named for supply chain failures will be scored on stale libraries and 11 CVEs, and the attacks it was created for will keep landing in A08.