This is the first of a series of posts where we open up the WPVulnerability database and look at what it says about WordPress security. WPVulnerability is a free and open API (no API key needed) that collects vulnerability information for WordPress core, plugins and themes from several public sources. Every number below comes from a snapshot of the database taken on 21 November 2026.
The short version: more vulnerabilities are being published than ever, nine out of ten of them are in plugins, and the type of bug being reported is changing in a way that matters to everyone who runs a WordPress site.
Where the vulnerabilities live
The database holds 59,713 vulnerability records. 54,367 of them, 91%, are in plugins. Themes account for 4,427 (7.4%) and WordPress core for 919 (1.5%).
Nine out of ten vulnerabilities are in plugins
All 59,713 records by component type, all years
Source: WPVulnerability database, snapshot of 21 November 2026.
This is the most useful fact in this post. WordPress core is built by a large team with a dedicated security process. The plugin and theme ecosystem is tens of thousands of products, maintained by everyone from large companies to a single developer in their spare time. The attack surface of a typical WordPress site is not really WordPress; it is everything installed on top of it.
Twenty years, one direction
Until 2020, the ecosystem typically saw anywhere from a few hundred to about 1,700 published vulnerabilities a year. Then the curve took off: 2,360 in 2021, 5,585 in 2022, 8,839 in 2024 and 11,485 in 2025. 2026 has already passed that with 12,120 records and six weeks still to go. At the current pace it will end the year at around 13,500.
Vulnerabilities published per year
By component type, 2003 to 2026 (2026 up to 21 November)
3,342 records (5.6%) have no published date in any of their sources and are not shown. Hover a column for its breakdown.
Show the data
| Year | Plugins | Themes | Core | Total |
|---|---|---|---|---|
| 2026 | 10,925 | 1,111 | 84 | 12,120 |
| 2025 | 10,519 | 948 | 18 | 11,485 |
| 2024 | 8,485 | 340 | 14 | 8,839 |
| 2023 | 6,968 | 214 | 25 | 7,207 |
| 2022 | 5,193 | 343 | 49 | 5,585 |
| 2021 | 2,172 | 164 | 24 | 2,360 |
| 2020 | 812 | 157 | 53 | 1,022 |
| 2019 | 909 | 70 | 50 | 1,029 |
| 2018 | 433 | 14 | 46 | 493 |
| 2017 | 677 | 13 | 101 | 791 |
| 2016 | 822 | 27 | 40 | 889 |
| 2015 | 1,512 | 150 | 30 | 1,692 |
| 2014 | 856 | 296 | 46 | 1,198 |
| 2013 | 382 | 124 | 35 | 541 |
| 2012 | 370 | 55 | 48 | 473 |
| 2011 | 227 | 23 | 30 | 280 |
| 2010 | 30 | 1 | 16 | 47 |
| 2009 | 31 | 1 | 34 | 66 |
| 2008 | 40 | 0 | 36 | 76 |
| 2007 | 22 | 14 | 79 | 115 |
| 2006 | 3 | 1 | 30 | 34 |
| 2005 | 1 | 0 | 24 | 25 |
| 2004 | 0 | 0 | 2 | 2 |
| 2003 | 0 | 0 | 2 | 2 |
| Undated | 2,978 | 361 | 3 | 3,342 |
More vulnerabilities does not necessarily mean worse code. Much of this growth reflects more people looking: WordPress-focused security companies running bug bounty programmes, more independent researchers, and more organisations issuing CVE identifiers for plugin and theme bugs. What it does mean is more advisories to act on. In 2026 that has been an average of around 37 new vulnerabilities a day.
2026, month by month
Monthly figures are lumpy. Vulnerabilities are often published in batches, when a research team releases the results of a large audit or a source publishes a backlog, so single months can spike well above the trend. The four busiest months on record are February 2022 (2,001), October 2026 (1,908), January 2025 (1,703), July 2023 (1,653).
Vulnerabilities published per month
All component types, 2015 to 2026. Darker means more; months with 1,500 or more are labelled
* 2026 runs to 21 November; December 2026 has not happened yet. Hover a cell for its exact value.
Show the data
| Year | Jan | Feb | Mar | Apr | May | Jun | Jul | Aug | Sep | Oct | Nov | Dec | Total |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2026 | 1,036 | 1,013 | 913 | 751 | 755 | 1,045 | 1,198 | 1,388 | 1,139 | 1,908 | 974 | – | 12,120 |
| 2025 | 1,703 | 897 | 1,056 | 1,390 | 717 | 868 | 661 | 729 | 1,047 | 739 | 621 | 1,057 | 11,485 |
| 2024 | 447 | 438 | 837 | 1,136 | 659 | 772 | 663 | 635 | 501 | 831 | 940 | 980 | 8,839 |
| 2023 | 539 | 465 | 512 | 445 | 598 | 460 | 1,653 | 390 | 393 | 529 | 737 | 486 | 7,207 |
| 2022 | 380 | 2,001 | 803 | 275 | 293 | 250 | 177 | 259 | 192 | 245 | 359 | 351 | 5,585 |
| 2021 | 89 | 108 | 163 | 219 | 167 | 208 | 189 | 326 | 237 | 354 | 154 | 146 | 2,360 |
| 2020 | 119 | 77 | 70 | 84 | 58 | 74 | 99 | 94 | 121 | 104 | 64 | 58 | 1,022 |
| 2019 | 32 | 178 | 59 | 50 | 82 | 107 | 135 | 71 | 82 | 135 | 39 | 59 | 1,029 |
| 2018 | 98 | 32 | 28 | 39 | 39 | 41 | 15 | 27 | 41 | 38 | 36 | 59 | 493 |
| 2017 | 49 | 35 | 71 | 119 | 117 | 46 | 55 | 52 | 75 | 54 | 67 | 51 | 791 |
| 2016 | 42 | 102 | 61 | 91 | 61 | 74 | 165 | 85 | 49 | 40 | 60 | 59 | 889 |
| 2015 | 59 | 82 | 101 | 220 | 514 | 108 | 160 | 214 | 68 | 57 | 70 | 39 | 1,692 |
What stands out about 2026 is its consistency more than any single spike. Every month from January to October saw at least 751 new vulnerabilities, and the second half of the year has been heavier than the first. October, with 1,908, was the busiest month in more than four and a half years.
Plugins: still where the risk is
Plugins account for 10,925 of this year’s vulnerabilities. That is a new annual record, already 4% more than the whole of 2025, with November still in progress. October alone added 1,762 plugin vulnerabilities.
For site owners the arithmetic is simple. Every plugin you install is a dependency with its own release cycle, its own maintainers and its own vulnerability history. A site with 30 plugins doesn’t have one security posture. It has 31.
Themes: a record year, front-loaded
Themes reached 1,111 vulnerabilities in 2026, the highest annual figure so far (the previous record was 948, in 2025) and more than three times the 340 of 2024. The year was unusually front-loaded: 636 of them, 57%, were published between January and March.
Themes are often thought of as “just the design”, but many modern themes ship settings panels, demo importers, page-builder integrations and their own AJAX endpoints. That is code, and it gets the same kinds of bugs as plugins.
WordPress core: small numbers, unusual year
Core has 919 unique vulnerabilities in the database across its whole history, deduplicated across sources and affected versions. Most years see a few dozen at most. 2026 is different: 84 so far, more than 2023, 2024 and 2025 combined (57) and the highest annual figure since 2017 (101). Most arrived late in the year: 74 between August and November.
WordPress core vulnerabilities per year
Unique vulnerabilities, deduplicated across sources and affected versions, 2003 to 2026
2026 runs to 21 November. Hover a column for its value.
Show the data
| Year | Core vulnerabilities |
|---|---|
| 2026 | 84 |
| 2025 | 18 |
| 2024 | 14 |
| 2023 | 25 |
| 2022 | 49 |
| 2021 | 24 |
| 2020 | 53 |
| 2019 | 50 |
| 2018 | 46 |
| 2017 | 101 |
| 2016 | 40 |
| 2015 | 30 |
| 2014 | 46 |
| 2013 | 35 |
| 2012 | 48 |
| 2011 | 30 |
| 2010 | 16 |
| 2009 | 34 |
| 2008 | 36 |
| 2007 | 79 |
| 2006 | 30 |
| 2005 | 24 |
| 2004 | 2 |
| 2003 | 2 |
Core is still the best-defended part of the stack. Minor releases, which carry the security fixes, install automatically by default, so most sites get those patches without anyone lifting a finger. It is worth checking that nothing on your site has quietly switched that off: a hosting panel setting, a constant in wp-config.php or a management tool.
What kinds of vulnerabilities?
When a vulnerability has a CVE, its CWE classification tells us what kind of weakness it is. Across all years, four families account for 76% of all classifications:
- Cross-site scripting (XSS), 41.4%: an attacker gets their own JavaScript to run in someone else’s browser, often an administrator’s.
- Authorization and access control, 16.4%: code does something without checking that the user is allowed to do it.
- Cross-site request forgery (CSRF), 11.3%: a logged-in user is tricked into sending a request they didn’t intend, usually by visiting a malicious page.
- SQL injection, 6.4%: user input ends up inside a database query.
| Attack type | Classifications | Share |
|---|---|---|
| Cross-site scripting (XSS) | 18,145 | 41.4% |
| Authorization / access control | 7,198 | 16.4% |
| CSRF | 4,942 | 11.3% |
| SQL injection | 2,825 | 6.4% |
| PHP file inclusion (LFI/RFI) | 1,244 | 2.8% |
| Path traversal | 1,111 | 2.5% |
| Arbitrary file upload | 1,089 | 2.5% |
| Deserialization (object injection) | 1,017 | 2.3% |
| Sensitive information exposure | 972 | 2.2% |
| IDOR (user-controlled key) | 838 | 1.9% |
The mix, however, is changing.
XSS and CSRF are shrinking; authorization flaws are growing
Share of all CWE classifications per year, 2014 to 2026
2026 is provisional: so far 76.1% of this year’s CVE-linked vulnerabilities have a CWE assigned, against 96.5% for 2025.
Show the data
| Year | XSS | Authorization | CSRF |
|---|---|---|---|
| 2026 | 29.3% | 20.3% | 3.4% |
| 2025 | 40.2% | 16.8% | 12.9% |
| 2024 | 48.1% | 15.4% | 11.0% |
| 2023 | 52.1% | 13.0% | 17.3% |
| 2022 | 31.9% | 28.8% | 16.6% |
| 2021 | 40.5% | 17.7% | 12.3% |
| 2020 | 34.7% | 11.2% | 16.2% |
| 2019 | 43.0% | 3.7% | 14.0% |
| 2018 | 53.1% | 1.3% | 8.7% |
| 2017 | 50.9% | 2.1% | 5.2% |
| 2016 | 54.6% | 1.3% | 11.5% |
| 2015 | 44.0% | 2.3% | 12.5% |
| 2014 | 54.0% | 0.3% | 16.4% |
Cross-site scripting has dominated WordPress vulnerability reports for more than a decade, and it still made up more than half of all classifications as recently as 2023. Its share has fallen for three years in a row, to 29.3% in 2026. CSRF has dropped even faster, from 12.9% in 2025 to 3.4%. Meanwhile, authorization and access-control bugs have climbed to 20.3%, their highest share since 2022.
2026: fewer noisy bugs, more dangerous ones
Comparing 2026 with 2025 makes the shift clearer. Together, XSS and CSRF went from 53.1% of all classifications to 32.7%. Most of the categories that grew are server-side flaws with serious consequences:
How the mix of attack types changed in a year
Share of all CWE classifications, 2025 compared with 2026. Top 12 types in 2026
2026 is provisional (76.1% CWE coverage so far). Hover a row for exact values.
Show the data
| Attack type | 2025 | 2026 |
|---|---|---|
| Cross-site scripting (XSS) | 40.2% | 29.3% |
| Authorization / access control | 16.8% | 20.3% |
| SQL injection | 5.9% | 7.6% |
| PHP file inclusion (LFI/RFI) | 5.8% | 5.5% |
| IDOR (user-controlled key) | 1.2% | 4.7% |
| Deserialization (object injection) | 2.6% | 4.5% |
| Sensitive information exposure | 1.1% | 3.8% |
| CSRF | 12.9% | 3.4% |
| Path traversal | 2.5% | 3.0% |
| Improper privilege management | 0.5% | 2.6% |
| Arbitrary file upload | 2.3% | 2.4% |
| Improper authentication | 0.2% | 2.0% |
- IDOR (authorization bypass through a user-controlled key): 1.2% to 4.7%. Change an ID in a request and get, or change, someone else’s data.
- Deserialization of untrusted data (PHP object injection): 2.6% to 4.5%.
- Sensitive information exposure: 1.1% to 3.8%.
- Improper privilege management: 0.5% to 2.6%. Typically privilege escalation, such as a subscriber account becoming an administrator.
- Improper authentication: 0.2% to 2.0%, from 19 classifications to 188.
- SQL injection: 5.9% to 7.6%.
Why is this happening? Part of the answer is probably that the easy bugs are being caught earlier: automated scanners and code review tools are good at finding missing escaping and missing nonces before a release ships. The other part is incentives. Researchers, and attackers, are spending their time on bugs that pay off. A stored XSS that needs an administrator to visit a page is worth far less than an unauthenticated privilege escalation.
One caution: 2026 figures will move. So far 76.1% of this year’s CVE-linked vulnerabilities have a CWE assigned, against 96.5% for 2025, because classification of recent CVEs is still catching up. The direction, though, matches a trend that has been building for several years.
A small number worth watching: malicious code
25 vulnerabilities in 2026 are classified as embedded malicious code (CWE-506): plugins or themes that shipped with a backdoor or other deliberately malicious code, rather than an accidental bug. That is up from 1 in 2025 and 11 in 2024.
The numbers are small, but this is a different kind of threat. No amount of careful configuration protects a site from code that was written to attack it. Common routes include a compromised developer account and an abandoned plugin changing hands. It is one more reason to know exactly what is installed on your sites, and who maintains it.
What this means for you
If you run WordPress sites
- Know what you run. Keep an inventory of every plugin and theme on every site, including inactive ones: an inactive plugin’s files are still on the server.
- Remove what you don’t use. Fewer components means fewer advisories to track and less code an attacker can reach.
- Turn on auto-updates for plugins and themes you trust, and keep core’s automatic minor updates enabled.
- Watch for abandoned software. A plugin that has been closed in the directory, or not updated in years, will not get a fix when its next vulnerability is found.
- Get alerts. The free WPVulnerability plugin checks core, plugins and themes (plus server software such as PHP, the web server and the database) against the database and can email you when something needs attention.
If you host or manage many sites
- Monitor the whole fleet. The WPVulnerability API is free, needs no API key and returns JSON, so it is easy to wire into your own tooling (documentation).
- Don’t rely on the firewall alone. WAF rules are good at spotting injection payloads. They are much worse at spotting a request that is perfectly well-formed but should not have been allowed. As the mix moves towards authorization and access-control flaws, how fast you update matters more than what you filter.
- Measure time to patch. At around 37 new vulnerabilities a day, the useful question is not whether your sites are affected but how long they stay affected.
If you build plugins or themes
- Check capabilities on every action, with
current_user_can(), including AJAX handlers and admin-post actions. REST routes need a realpermission_callback; use__return_trueonly for endpoints that are public by design. - Check the object, not just the role. Knowing that a user is logged in, or is an editor, is not enough. Check that they may act on that object, for example
current_user_can( 'edit_post', $post_id ). This is what prevents IDOR. - Prepare every query with
$wpdb->prepare(). - Never call
unserialize()on data a user can influence. Use JSON instead. - Never build
includeorrequirepaths from request data. If a file must be chosen dynamically, choose it from a fixed allowlist. - Keep doing the basics: escape output late and verify nonces. XSS and CSRF are shrinking as a share, but they are still the largest categories in absolute numbers.
- Publish a way to report security issues, so researchers can reach you before they publish.
How we count
- Plugins and themes are counted per product. The same CVE affecting two different plugins is two records, one for each product, because each needs its own fix and its own update. These totals match the public API.
- WordPress core is counted per vulnerability. Advisories are deduplicated through their CVE cross-references, so a vulnerability reported by several sources, or affecting many core versions, counts once.
- Dates are the earliest published date found across all the advisories linked to each vulnerability.
- 3,342 records (5.6%) have no published date in any of their sources. They are included in the totals but left out of the per-year and per-month charts.
- Attack types come from CWE classifications and cover CVE-linked vulnerabilities only. A vulnerability with two CWEs counts once for each, so shares are shares of classifications, not of vulnerabilities. Near-duplicate CWEs are grouped into families (for example, all XSS variants under XSS).
- The current year is provisional. CWE classification of recent CVEs takes time, so 2026 attack-type shares will shift as coverage catches up.
Get the data
Everything in this post comes from the WPVulnerability database. The API is open to everyone with no key required, and its documentation covers every endpoint. Live figures are on the statistics page. To get alerts on your own site, install the WPVulnerability plugin. The project is open source on GitHub.
If there is a cut of the data you would like to see in a future post, let us know.