Datasets:
security put 2026-11-02
Browse files
README.md
CHANGED
|
@@ -4,6 +4,27 @@ pretty_name: Security Daily (CVE, KEV, EPSS movers, breach and incident notices)
|
|
| 4 |
language: [en, ja]
|
| 5 |
tags: [cybersecurity, cve, vulnerability, kev, epss, nvd, data-breach, incident, time-series, threat-intelligence, patch-management]
|
| 6 |
size_categories: [100K<n<1M]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
configs:
|
| 8 |
- config_name: summary
|
| 9 |
data_files: "daily/summary/*.csv"
|
|
@@ -44,7 +65,7 @@ and regulators publish who disclosed a breach. Each source only shows *today*. T
|
|
| 44 |
|
| 45 |
| folder | rows | what |
|
| 46 |
|---|---|---|
|
| 47 |
-
| `snapshot/cve_signals/published_YYYY.csv` | every CVE in NVD (~420k), one file per published year | CVSS 3/4, CWE, vendors, products, **EPSS score + percentile, KEV flag, KEV date, known ransomware use** in one row. Refreshed weekly. |
|
| 48 |
| `snapshot/kev.csv` | all CISA KEV entries | with EPSS and CVSS joined |
|
| 49 |
| `snapshot/incidents.csv` | ~7k breach / incident notices | SEC 8-K Item 1.05, California AG, Washington AG, CISA, JPCERT/CC, IPA |
|
| 50 |
| `snapshot/breaches.csv` | ~1k breaches | Have I Been Pwned metadata + Japan disclosure ledger |
|
|
@@ -63,6 +84,90 @@ and regulators publish who disclosed a breach. Each source only shows *today*. T
|
|
| 63 |
where the organisation name does not clearly name an organisation, `org` is withheld (`org_withheld=true`) and the row is kept for counting.
|
| 64 |
- JPCERT/CC and IPA: title, date and link only; read the content at the source.
|
| 65 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 66 |
## Sources and licences
|
| 67 |
- NVD: *This product uses data from the NVD API but is not endorsed or certified by the NVD.* (US Government work)
|
| 68 |
- CISA Known Exploited Vulnerabilities Catalog: CC0 / US Government work.
|
|
|
|
| 4 |
language: [en, ja]
|
| 5 |
tags: [cybersecurity, cve, vulnerability, kev, epss, nvd, data-breach, incident, time-series, threat-intelligence, patch-management]
|
| 6 |
size_categories: [100K<n<1M]
|
| 7 |
+
dataset_info:
|
| 8 |
+
- config_name: cve_signals
|
| 9 |
+
features:
|
| 10 |
+
- {name: cve_id, dtype: string}
|
| 11 |
+
- {name: published, dtype: string}
|
| 12 |
+
- {name: last_modified, dtype: string}
|
| 13 |
+
- {name: status, dtype: string}
|
| 14 |
+
- {name: cvss3, dtype: float64}
|
| 15 |
+
- {name: cvss3_severity, dtype: string}
|
| 16 |
+
- {name: cvss4, dtype: float64}
|
| 17 |
+
- {name: cvss4_severity, dtype: string}
|
| 18 |
+
- {name: cwes, dtype: string}
|
| 19 |
+
- {name: vendors, dtype: string}
|
| 20 |
+
- {name: products, dtype: string}
|
| 21 |
+
- {name: epss, dtype: float64}
|
| 22 |
+
- {name: epss_percentile, dtype: float64}
|
| 23 |
+
- {name: epss_date, dtype: string}
|
| 24 |
+
- {name: kev, dtype: bool}
|
| 25 |
+
- {name: kev_added, dtype: string}
|
| 26 |
+
- {name: kev_due, dtype: string}
|
| 27 |
+
- {name: kev_ransomware, dtype: string}
|
| 28 |
configs:
|
| 29 |
- config_name: summary
|
| 30 |
data_files: "daily/summary/*.csv"
|
|
|
|
| 65 |
|
| 66 |
| folder | rows | what |
|
| 67 |
|---|---|---|
|
| 68 |
+
| `snapshot/cve_signals/published_YYYY.csv` | every CVE in NVD (~420k), one file per published year (2026: one per month, `published_2026-MM.csv`) | CVSS 3/4, CWE, vendors, products, **EPSS score + percentile, KEV flag, KEV date, known ransomware use** in one row. Refreshed weekly. |
|
| 69 |
| `snapshot/kev.csv` | all CISA KEV entries | with EPSS and CVSS joined |
|
| 70 |
| `snapshot/incidents.csv` | ~7k breach / incident notices | SEC 8-K Item 1.05, California AG, Washington AG, CISA, JPCERT/CC, IPA |
|
| 71 |
| `snapshot/breaches.csv` | ~1k breaches | Have I Been Pwned metadata + Japan disclosure ledger |
|
|
|
|
| 84 |
where the organisation name does not clearly name an organisation, `org` is withheld (`org_withheld=true`) and the row is kept for counting.
|
| 85 |
- JPCERT/CC and IPA: title, date and link only; read the content at the source.
|
| 86 |
|
| 87 |
+
## Why this exists
|
| 88 |
+
|
| 89 |
+
Each public security source answers one question and only shows its current state:
|
| 90 |
+
|
| 91 |
+
- NVD: what the vulnerability is (CVSS, CWE, affected products).
|
| 92 |
+
- CISA KEV: whether it is known to be exploited, and since when.
|
| 93 |
+
- FIRST EPSS: how likely exploitation activity is in the next 30 days. **EPSS publishes only today's score; yesterday's is gone unless someone kept it.**
|
| 94 |
+
- Regulators and CERTs: who disclosed a breach or incident, and when.
|
| 95 |
+
|
| 96 |
+
To answer "what should I patch first?" or "how exposed is this vendor?" you need all four joined per CVE, and you need the history. This dataset does the join every day and keeps every day as a file, starting 2026-11-02. It is for patch prioritisation, vulnerability-trend research, teaching, and as ground truth for models that talk about CVEs.
|
| 97 |
+
|
| 98 |
+
Size on 2026-11-03: 418,180 NVD records, 1,766 KEV entries, 398,815 EPSS scores, 7,233 breach and incident notices from 6 regulator/CERT sources, 989 breach records.
|
| 99 |
+
|
| 100 |
+
## How fresh
|
| 101 |
+
|
| 102 |
+
| job | schedule (UTC) | what it writes |
|
| 103 |
+
|---|---|---|
|
| 104 |
+
| daily export | every day at **00:50Z** | `daily/*/YYYY-MM-DD.csv` for the previous UTC day |
|
| 105 |
+
| weekly export | every **Monday, 01:05Z to 01:45Z** | `snapshot/*` (full refresh) and `weekly/*` (last 7 days) |
|
| 106 |
+
|
| 107 |
+
Source fetches run before the export. Every daily `summary` row carries `last_fetch`, so you can check freshness per source per day. A missing day means the job did not complete, not that nothing happened.
|
| 108 |
+
|
| 109 |
+
## Examples
|
| 110 |
+
|
| 111 |
+
**Load a config**
|
| 112 |
+
|
| 113 |
+
```python
|
| 114 |
+
from datasets import load_dataset
|
| 115 |
+
|
| 116 |
+
kev = load_dataset("Lilambd/security-daily", "kev", split="train").to_pandas()
|
| 117 |
+
pf = load_dataset("Lilambd/security-daily", "patch_first", split="train").to_pandas()
|
| 118 |
+
print(len(kev)) # 1766 on 2026-11-02
|
| 119 |
+
```
|
| 120 |
+
|
| 121 |
+
**Today's top 10 to patch first**
|
| 122 |
+
|
| 123 |
+
```python
|
| 124 |
+
latest = pf["day"].astype(str).max()
|
| 125 |
+
pf[pf["day"].astype(str) == latest].sort_values("rank").head(10)[
|
| 126 |
+
["rank", "cve_id", "vendors", "epss", "kev_due", "kev_ransomware", "why"]]
|
| 127 |
+
```
|
| 128 |
+
|
| 129 |
+
**KEV entries by vendor**
|
| 130 |
+
|
| 131 |
+
```python
|
| 132 |
+
kev.groupby("vendor").size().sort_values(ascending=False).head(10)
|
| 133 |
+
# Microsoft 390, Cisco 102, Apple 95, Adobe 83, Google 76 (2026-11-03)
|
| 134 |
+
```
|
| 135 |
+
|
| 136 |
+
**Large files without `datasets`**
|
| 137 |
+
|
| 138 |
+
```python
|
| 139 |
+
import pandas as pd
|
| 140 |
+
df = pd.read_csv("hf://datasets/Lilambd/security-daily/snapshot/cve_signals/published_2025.csv")
|
| 141 |
+
```
|
| 142 |
+
|
| 143 |
+
A full notebook is in [examples/quickstart.ipynb](examples/quickstart.ipynb): top 20 patch-first, KEV by vendor, EPSS distribution, CVEs per year, incidents per month by source. It also runs on Kaggle and Colab.
|
| 144 |
+
|
| 145 |
+
Notes:
|
| 146 |
+
- `cve_signals` is about 70 MB of CSV in 38 files; pandas can read one year at a time as above.
|
| 147 |
+
- `kev_added`, `epss_movers`, `incidents_new` and `breaches_new` get a file every day; a header-only file means no change that day.
|
| 148 |
+
|
| 149 |
+
## Citation
|
| 150 |
+
|
| 151 |
+
```bibtex
|
| 152 |
+
@misc{lilambd_security_daily_2026,
|
| 153 |
+
author = {{Lilambd}},
|
| 154 |
+
title = {Security Daily: CVE, KEV, EPSS movers, breach and incident notices},
|
| 155 |
+
year = {2026},
|
| 156 |
+
publisher = {Hugging Face},
|
| 157 |
+
howpublished = {\url{https://huggingface.co/datasets/Lilambd/security-daily}},
|
| 158 |
+
note = {Daily files from 2026-11-02. CC BY 4.0}
|
| 159 |
+
}
|
| 160 |
+
```
|
| 161 |
+
|
| 162 |
+
When you use EPSS values, also credit "EPSS, FIRST.org". When you use NVD fields: "This product uses data from the NVD API but is not endorsed or certified by the NVD."
|
| 163 |
+
|
| 164 |
+
## Changelog
|
| 165 |
+
|
| 166 |
+
| date | change |
|
| 167 |
+
|---|---|
|
| 168 |
+
| 2026-11-02 | First daily files: `summary`, `cve_new`, `cve_modified`, `patch_first`. First snapshots: `cve_signals` (one file per published year 1999 to 2025, per month for 2026), `kev`, `incidents`, `breaches`. First weekly files: `vendor`, `incidents` (week ending 2026-11-02). |
|
| 169 |
+
| 2026-11-03 | README: added Why this exists, How fresh, Examples, Citation, Changelog. Added `examples/quickstart.ipynb`. Empty days now get header-only files. |
|
| 170 |
+
|
| 171 |
## Sources and licences
|
| 172 |
- NVD: *This product uses data from the NVD API but is not endorsed or certified by the NVD.* (US Government work)
|
| 173 |
- CISA Known Exploited Vulnerabilities Catalog: CC0 / US Government work.
|