skip to content

Midkernel Threat Intel statistics

Freeze tistats-2026-11-17-r1 · Current statistics

About this page

This page publishes counts that Midkernel Threat Intel derived from two public sources: the CISA Known Exploited Vulnerabilities (KEV) catalog and the reviewed advisories in the GitHub Advisory Database. The figures count cataloged and published records. They do not measure incidents, prevalence or affected users.

Each figure has its own as-of date, a one-line caveat and a method note. Windows are complete UTC calendar months. This page is a frozen snapshot. For a live view, use the Threat Intel connector described at the end.

How to cite

Source: Midkernel Threat Intel statistics, data from CISA KEV and the GitHub Advisory Database, as of 2026-11-17.

To cite one figure, add the link to its section and keep its as-of date. GitHub-derived figures are Midkernel counts derived from the GitHub Advisory Database.


M1. CISA KEV additions per month and quarter

Refresh deadline: .

As of: 2026-11-17 UTC. Source version: CISA KEV catalogVersion 2026.11.17. Method: method-v1.0. Window: complete UTC months, by the date CISA added each entry (dateAdded).

Takeaway: CISA added 282 vulnerabilities to its KEV catalog from 2026-01-01 to 2026-10-31, against 214 in the same ten months of 2025. Midkernel counted both periods with the same method and verified both.

Month20252026
011417
022728
033226
041531
052421
062023
072026
081531
091643
103136
Jan to Oct214282
Quarter20252026
Q17371
Q25975
Q351100

Only complete months are shown.

Caveat: Counts are CISA catalog additions by the date CISA added them (UTC), not the date a vulnerability was disclosed or first exploited. They show what CISA chose to catalog, not how many vulnerabilities are exploited.

Data: CISA's catalog, counted and verified by Midkernel. CISA KEV is released under CC0 1.0.


M2. CISA KEV additions by vendor, January to October 2026

Refresh deadline: .

As of: 2026-11-17 UTC. Source version: CISA KEV catalogVersion 2026.11.17. Method: method-v1.0. Sample: the 282 additions in M1, grouped by CISA's vendorProject field (whitespace trimmed, nothing else merged). The catalog lists 106 distinct vendors.

Takeaway: Microsoft has the most KEV additions in this period with 40, followed by Cisco with 20.

Not a security ranking. This table counts what CISA cataloged for each vendor name. It does not rank vendors by how secure they are.

#VendorKEV additions
1Microsoft40
2Cisco20
3=Apple9
3=Fortinet9
3=Google9
3=Linux9
7Ivanti8
8Adobe7
9Citrix6
10Synacor5
All other vendors (96)160
Total (equals the M1 total)282

Ties share a rank (3= is four vendors).

Caveat: Vendor counts reflect what CISA catalogs and how widely a vendor's products are deployed. They are not a measure of how secure a vendor is, and vendor names are CISA's.

Data: CISA KEV, released under CC0 1.0.


M3. CISA KEV additions by weakness type (CWE), January to October 2026

Refresh deadline: .

As of: 2026-11-17 UTC. Source version: CISA KEV catalogVersion 2026.11.17. Method: method-v1.0. Sample: the 282 additions in M1, grouped by each CWE identifier CISA lists for the entry. The catalog lists 102 distinct CWEs. 26 additions list more than one CWE, so rows overlap and do not sum to 282.

Takeaway: CWE-94 is the weakness id CISA lists most often in this period, on 19 additions, followed by CWE-287 on 17 and CWE-78 on 16.

#CWE idKEV additions
1CWE-9419
2CWE-28717
3CWE-7816
4CWE-2214
5=CWE-30613
5=CWE-50213
7CWE-91811
8CWE-78710
9=CWE-4169
9=CWE-899
No CWE given9

Caveat: Weakness types are the CWE identifiers CISA lists for each entry. An entry can list several, so rows overlap, and some entries list none. CISA's CWE assignments can change and are not comparable across years.

Data: CISA KEV (CC0 1.0) for the assignments. CWE identifiers refer to the Common Weakness Enumeration, credit The MITRE Corporation. See MITRE's CWE terms of use. Look up an id on MITRE's CWE site.


M5. Reviewed GitHub advisories published per month, by package ecosystem

Refresh deadline: .

As of: 2026-11-17 UTC (GitHub pull 17:48 to 17:54 UTC). Source version: GitHub Advisory Database, reviewed advisories. Method: method-v1.0. Sample: advisories GitHub has reviewed and first published in the window, with withdrawn advisories removed and one count per CVE (otherwise per GHSA id).

Takeaway: GitHub published 11,635 reviewed advisories, counted once per CVE or GHSA id, from 2026-01-01 to 2026-10-31.

RowYTD01020304050607080910
All (unique canonical ids)11,6354847541,5641,5001,6611,5511,3207218841,196
npm3,412136261603456361339325182288461
PyPI2,472102121195288377361257193211367
Go2,17083148329258338275268109178184
Maven1,041544172140167260130755448
crates.io49931405773813988183240
Other2,14384147317289359285268148138108
RowQ1Q2Q3YTD (Jan to Oct)
All (unique canonical ids)2,8024,7122,92511,635
npm1,0001,1567953,412
PyPI4181,0266612,472
Go5608715552,170
Maven1675672591,041
crates.io128193138499
Other5489335542,143

Ecosystem rows overlap (77 advisories affect more than one) and are not summed. "Other" is composer, nuget, rubygems, erlang, swift, actions and pub.

Caveat: These are advisories published and reviewed by GitHub, not all vulnerabilities. Monthly movement reflects GitHub's review intake and a few large projects' disclosure batches as well as the ecosystems themselves. GitHub can withdraw or re-date advisories after the date shown. Midkernel makes no year-on-year comparison.

Data: Midkernel counts derived from the GitHub Advisory Database, which is licensed CC BY 4.0. Midkernel changed the data by removing withdrawn advisories and counting each CVE (otherwise GHSA id) once.


M6. MCP and agent-tooling advisories per month

Refresh deadline: .

As of: 2026-11-17 UTC (GitHub pull 2026-11-17 UTC). Source version: GitHub Advisory Database, reviewed advisories, with membership list-v1. Method: method-v1.0. Sample: the same reviewed advisories as M5 whose affected package is on Midkernel's published frozen list:

  • (a) MCP SDKs, servers and clients.
  • (b) Agent frameworks, SDKs and skill and plugin systems, judged by each package's own description. OpenClaw is included.
  • (c) A closed list of LLM app and serving runtimes: vLLM, Ollama, LiteLLM and Open WebUI.
  • (d) MCP surfaces inside non-AI products (24), shown separately and not in the headline.

Takeaway: GitHub published 1,586 reviewed advisories for packages on this list from 2026-01-01 to 2026-10-31, and the five largest projects account for 1,056 of them (see concentration below).

RowYTD01020304050607080910
MCP and agent tooling (headline)1,5861311033330918718813810483121
(a) MCP SDKs, servers, clients2976102536383546274628
(b) agent frameworks/SDKs, skill/plugin systems1,061399301261811186158970
(c) LLM app/serving runtimes (closed list)22941712683532192823
(d) MCP surfaces in non-AI products (separate, not in headline)241033125018
RowQ1Q2Q3YTD (Jan to Oct)
MCP and agent tooling (headline)4566843251,586
(a) MCP SDKs, servers, clients41109119297
(b) agent frameworks/SDKs, skill/plugin systems4034601281,061
(c) LLM app/serving runtimes (closed list)1211579229
(d) MCP surfaces in non-AI products (separate)46624

Rows (a) to (c) overlap by one advisory and are not summed.

Project concentration

A few projects drive most months.

  • openclaw/openclaw is the largest project in February (75.5% of that month), March (81.7%), April (52.1%) and July (29.7%).
  • Other months over 25% are led by vllm-project/vllm (Jan, 30.8%), open-webui/open-webui (May, 32.6%), MervinPraison/PraisonAI (Jun, 28.7%) and sooperset/mcp-atlassian (Sep, 30.1%).
  • August (25.0%, FlowiseAI/Flowise) and October (24.0%, MervinPraison/PraisonAI) are not over 25%.
  • Across the year the five largest projects are openclaw/openclaw 601, MervinPraison/PraisonAI 178, open-webui/open-webui 140, FlowiseAI/Flowise 81 and vllm-project/vllm 56, which is 1,056 of 1,586.

Caveat: Reviewed GitHub advisories for MCP and agent-tooling packages on Midkernel's published list. Membership is a judgment, the list is published, and a few large projects dominate some months. Monthly movement reflects those projects' disclosure batches and GitHub's review intake, not a trend in AI security. Midkernel makes no growth claim.

Data: Midkernel counts derived from the GitHub Advisory Database, which is licensed CC BY 4.0. Package descriptions come from each project's own GitHub or registry description. The full membership list, list-v1, is published with this page (columns: class, repo, packages, description).


M7. GitHub advisory withdrawn share, January to July 2026 cohorts

Refresh deadline: .

As of: 2026-11-17 UTC (GitHub pull 2026-11-17 UTC). Source version: GitHub Advisory Database, reviewed advisories. Method: method-v1.0. Sample: every reviewed advisory GitHub published in the cohort month, with withdrawn ones included in the denominator. Only cohorts at least 90 days old at the freeze are published.

Takeaway: Of 9,249 reviewed advisories published in the January to July 2026 cohorts, GitHub had withdrawn 415 (4.5%) as of the pull, and 405 of those 415 (97.6%) carry the "Duplicate Advisory" label.

Cohort (UTC month published)Reviewed advisories publishedWithdrawn as of pullWithdrawn share"Duplicate Advisory" among withdrawn
2026-01494102.0%8 (80.0%)
2026-02764101.3%9 (90.0%)
2026-031,640764.6%73 (96.1%)
2026-041,585855.4%83 (97.6%)
2026-051,705442.6%43 (97.7%)
2026-061,6511006.1%99 (99.0%)
2026-071,410906.4%90 (100.0%)
Jan to Jul pooled9,2494154.5%405 (97.6%)

Caveat: Share of advisories that GitHub later withdrew, as of 2026-11-17. Almost all withdrawals are duplicates of another advisory. Withdrawals keep arriving for more than a year, so older cohorts can rise. This describes GitHub's advisory processing, not the quality of any ecosystem.

Data: Midkernel counts derived from the GitHub Advisory Database, which is licensed CC BY 4.0.


M8. Configured Threat Intel sources and taxonomy version

Refresh deadline: .

As of: 2026-11-17 UTC (registry revision 0392793177b933b286da7d5f5f110741a9a753e6). Method: method-v1.0.

Takeaway: Midkernel Threat Intel has 149 configured public sources (124 web2, 25 web3). The taxonomy version is evidence/3.2.0.

Surface / formatConfigured sources
web2/atom22
web2/csv9
web2/json10
web2/rss69
web2/txt14
web3/atom3
web3/json2
web3/rss20
Total configured149 (web2 124, web3 25)

Caveat: Configured sources are the public feeds Midkernel's collector is set up to read. Configuration does not guarantee that a source produced records in any given period.

Data: Midkernel's own source registry. Source names belong to their publishers.


Method and sources

Midkernel counts are derived from CISA's Known Exploited Vulnerabilities catalog (catalogVersion 2026.11.17) and GitHub's reviewed advisories (pulled 2026-11-17 UTC), checked against Midkernel's own ingest. Windows are complete UTC calendar months. Every figure on this page passed Midkernel's per-figure verification for this freeze: two agreeing source pulls, deduplication by CVE or GHSA ID, and misfile checks.

SourceWhat Midkernel usedLicence and credit
CISA KEV catalogM1, M2, M3CC0 1.0 (CISA licence). CISA's catalog, counted and verified by Midkernel.
GitHub Advisory DatabaseM5, M6, M7CC BY 4.0 (licence). Midkernel counts derived from the database, with the changes described under M5.
MITRE CWECWE identifiers in M3Credit The MITRE Corporation.
Midkernel source registryM8Midkernel's own configuration.

Midkernel's counts have not been reviewed or endorsed by CISA, the U.S. Department of Homeland Security, GitHub or MITRE. This page uses no CISA or DHS logo.

Other notes on method:

  • CISA KEV counts use the date CISA added each entry. Vendor names and CWE assignments are CISA's.
  • GitHub counts use the date GitHub first published each reviewed advisory. In M5 and M6, withdrawn advisories are removed and each CVE (otherwise GHSA id) is counted once. M7 counts every reviewed advisory published in each cohort month, withdrawn ones included, because it measures the withdrawn share.
  • The MCP and agent-tooling list (list-v1) is a published judgment list. The full list is published with this page.

Refresh policy

  • Cadence. Monthly. Each freeze runs in the first scheduled slot that falls at least five UTC days after the month ends, so CISA's edits have settled.
  • Versioning. Every freeze has an id, a method version and a list version. This freeze is tistats-2026-11-17-r1, method-v1.0, list-v1. Older freezes keep their dated permalinks and show a "superseded by" banner. Figure anchors on this page do not change.
  • Changes. A "What changed" section below lists each freeze, which figures changed and why, and any restated figure.
  • Re-freeze triggers. An upstream correction from CISA or GitHub, a change to a definition or to the list, or a figure found wrong.
  • Stale figures. Each figure carries a refresh deadline. After it passes, the page shows "not refreshed since" and the freeze date. If a re-freeze is pending, the figure shows "under review, may be restated" and keeps its old value and date.
  • Next freeze. The first scheduled slot at least five UTC days after the next month ends.

What changed

  • 2026-11-17. First publication. Freeze tistats-2026-11-17-r1, method-v1.0, list-v1.

Freeze version

ItemValue
Freeze idtistats-2026-11-17-r1
As of2026-11-17 (UTC)
Methodmethod-v1.0
MCP and agent-tooling listlist-v1
CISA KEV catalogcatalogVersion 2026.11.17
Source registryrevision 0392793177b933b286da7d5f5f110741a9a753e6
Taxonomyevidence/3.2.0

Query the live data

For current records instead of a monthly snapshot, add the free Threat Intel MCP connector. No login is needed: https://www.midkernel.com/security/app/mcp/threat-intel

Midkernel Threat Intel statistics