Terms used on this page
CVE
A publicly disclosed vulnerability with an ID from the official CVE list. A count of CVEs measures disclosure, not risk.
CISA KEV
CISA's Known Exploited Vulnerabilities catalogue: CVEs with reliable evidence of exploitation and an available fix. A lower bound, focused on products US federal agencies run; the date is when CISA listed it, not when attacks began.
EPSS
FIRST's Exploit Prediction Scoring System: a model's estimate of the probability of exploitation activity in the next 30 days. A ranking aid, not evidence.
Severity (CVSS)
The base score in the CVE record (vendor or CISA-ADP): worst-case impact, not likelihood of attack.
CNA
CVE Numbering Authority: the organisation that assigns and files a CVE.
Bulk filers
Third parties filing CVEs for other people's software in bulk: GitHub (advisories) and VulnCheck.
Surge filers
Any CNA whose filing grew at least fourfold to 1,000+ in a year, including vendors now filing many more CVEs for their own products.
Origin
What kind of software a CVE is in: open-source projects and packages, web CMS plugins, network and security edge devices, OS/browser platforms, or enterprise and other software.
AI-related
A CVE in AI software, MCP servers and agents, AI features, AI compute, or an AI-specific weakness. CVEs merely found by an AI tool do not count.
Zero-day proxy
A KEV entry listed within a day of its CVE's publication: broader than confirmed zero-days (Project Zero) because it also counts flaws attacked as soon as they were disclosed.

CVEs per year, by origin

Published CVEs per calendar year, stacked by origin.

Signal vs noise

Every bar counts the same CVEs: those published in the last 12 months.

Where exploitation happens

CISA KEV additions this year, by origin.

How to adjust your programme

Changes the current data supports, most urgent first. Each cites the numbers behind it; follow a line to the evidence below. Recomputed with every daily rebuild.

    Part 1 · Exploitation Exploitation & zero-days — all CVEs

    Exploited vulnerabilities by vendor

    CISA KEV additions this year per vendor.

    Ransomware-linked exploitation

    Share of each year's CISA KEV additions that CISA links to ransomware campaigns.

    Time from CVE publication to KEV listing

    KEV additions per year by days from CVE publication to KEV.

    Latest exploited vulnerabilities

    The most recent CISA KEV additions, newest first.

    CVEProductOriginAdded to KEV (UTC)Days to KEVEPSS before listing

    Listed in KEV at disclosure (zero-day proxy)

    KEV entries added at most one day after the CVE was published, by publication year.

    Recent zero-days

    Latest vulnerabilities exploited at or before disclosure.

    CVEProductOriginDate (UTC)

    Patch next — highest exploit probability (EPSS), not yet in KEV

    CVEs published in the last 12 months, not in CISA KEV, ranked by FIRST EPSS.

    CVEProductOriginPublished (UTC)EPSS (30-day prob.)

    Part 2 · Volume Volume, severity & weaknesses — all CVEs

    Severity per year

    Published CVEs per year by CVSS severity (from the CVE record).

    Top weakness classes (CWE)

    Last 12 months vs the 12 months before.

    Top vendors

    Last 12 months vs the 12 months before.

    Who files the CVEs (CNAs)

    Last 12 months vs the 12 months before.

    Supply chain: open-source projects & packages

    Open-source CVEs per year

    CVEs whose affected product is an open-source package (npm, PyPI, Maven, Go, crates.io, …).

    Malicious packages per year

    Malicious open-source packages reported to OSV (MAL-* advisories), by ecosystem. These are not CVEs.

    Part 3 · AI-related Which of them are AI-related

    Exploited AI-related vulnerabilities (CISA KEV)

    AI-related CVEs in CISA's Known Exploited Vulnerabilities catalog — confirmed in-the-wild exploitation.

    What to do

      CVEProductAI relationPublished (UTC)Added to KEV (UTC)Days to KEV (days)EPSS before listing

      Patch next — AI-related CVEs with the highest exploit probability (EPSS)

      AI-related CVEs published in the last 12 months, not in CISA KEV, ranked by FIRST EPSS.

      CVEProductAI relationPublished (UTC)EPSS (30-day prob.)

      How CVEs relate to AI

      AI-related CVEs this year by kind of AI relation. A CVE can have more than one.

      AI-related vs all CVEs

      Published CVE records from the official CVE list.

      Explore: aggregation, range, mode
      Agg Range Mode Series

      Where AI-related CVEs come from

      AI-related CVEs per year, stacked by the same origins as Part 1.

      How AI-related CVEs differ from all CVEs

      Weakness classes of AI-related CVEs compared with non-AI CVEs.

      Top weakness classes (CWE)

      Share of CVEs carrying each CWE.

      Who reports AI-related CVEs

      AI-related CVEs per year by assigning CNA (top assigners, rest grouped as "other").

      Method How this is measured

      Methodology

      Public data only, rebuilt daily. Origins say what kind of software a CVE is; AI relations say how it involves AI; exploitation comes from CISA KEV and Project Zero.

      Show definitions, accuracy and caveats

      Origins (Part 1)

      Every published CVE gets exactly one origin — what kind of software it is, regardless of AI. Rules are checked in this order, except that CMS plugins are checked before open-source packages, so a WordPress plugin with a package URL counts as a plugin:

        CISA KEV: what it measures

        "Exploited" on this page means listed in CISA's Known Exploited Vulnerabilities (KEV) catalogue. KEV is a lower bound, not a census: CISA lists a CVE only with reliable evidence of exploitation and an available fix or mitigation, and it focuses on products US federal agencies run. A KEV date is when CISA listed the CVE, not when exploitation began; attacks often start earlier. KEV additions in a year include older CVEs newly listed, so they measure CISA's intake, not the volume of attacks. Commercial catalogues (for example VulnCheck KEV) list more CVEs, often earlier.

        Bulk filers and surge filers

        A CVE Numbering Authority (CNA) assigns CVE IDs. Two kinds of filing change move the counts without more vulnerable software. "Bulk filers" are third parties that file CVEs for other people's software in bulk: GitHub (security advisories for open-source projects) and VulnCheck. "Surge filers" are any CNA whose filing grew at least fourfold to 1,000+ in the last 12 months; besides the bulk filers this includes vendors now filing far more CVEs for their own products. Growth figures say which set they leave out.

        Zero-day proxy

        "KEV-listed at disclosure" = a CISA KEV entry added at most one day after its CVE was published. It undercounts true zero-days (exploitation CISA confirms later is missed and never fills in), rises when CISA lists faster (as in 2026, when its deadlines tightened), and is bucketed by CVE publication year; KEV additions are bucketed by date added. Google Project Zero's in-the-wild tracker is shown alongside as an independent count.

        What counts as AI-related (Part 2)

        A CVE is AI-related when at least one of these relations applies, other than "found by AI": an AI tool finding a bug in ordinary software (curl, OpenSSL) is a discovery trend, counted across all CVEs, not AI attack surface. They are tags, not exclusive buckets — a prompt-injection bug in an MCP server that an AI tool found carries three:

          AI software, MCP agents and AI features are further placed in one layer of the AI stack (first matching rule wins):

            Severity

            One CVSS family per CVE, from the CVE record: the CNA's CVSS v3.1, else CNA v4.0, else CNA v3.0, else CISA-ADP (Vulnrichment) v3.1. Part 1's severity uses only the record; Part 2's tables fall back to NVD's score (v3.1 → v4.0 → v3.0, never v2) when the record has none.

            Caveats

            • The AI tagging is a rule-based heuristic over descriptions, products, credits and CWEs: expect misses and some false matches (see the measured precision and recall).
            • "Found by AI tools or AI labs" relies on the CVE record's credits. Most records credit people, not tools, so this undercounts AI-assisted discovery; it also counts researchers who work at an AI lab.
            • "AI compute" includes NVIDIA's consumer GPU drivers: NVIDIA ships one driver branch for GeForce and data-center GPUs.
            • Counts reflect the CNA mix — new or very active assigners (VulnCheck, huntr, GitHub) change the curve independently of the underlying software.
            • CISA KEV is a lower bound of exploitation: it lists only exploitation CISA has confirmed and judged actionable, and it favours products US agencies run (Microsoft, Cisco, Fortinet…).
            • "Open-source projects & packages" includes every GitHub-assigned (GitHub_M) record. GitHub's CNA issued far more CVEs in 2026 than in 2025, so part of that origin's growth is a reporting change, not only more vulnerable code.
            • EPSS is a model estimate, not an observation; it is shown to prioritise, not to prove exploitation.