Skip to content

Response, New York. Noah Park: Check Point Patches Actively Exploited Management Server Path Traversal. Detection, Dallas. Ava Okello: Report Finds Nearly Half of Deployed Detections Need Attention Before SOCs Can Trust Them.

SOCtember

Always First. Fast SOC News.

Screenshot of the Vulnerability timeline panel on CIRCL's Vulnerability-Lookup for CVE-2025-0282, showing metric cards for time unpatched of 678 days 14 hours, time to exploit of minus 18.93 hours to plus 5.07 hours with order unknown, and exploitation to patch of about 679 days 8.93 hours with the note no patch recorded, above a day-by-day list of 465 recorded events.
Image: CIRCL Vulnerability-Lookup (fair use). The timeline panel for CVE-2025-0282 on vulnerability.circl.lu, captured Nov. 18, 2026. The exploitation date is a KEV listing, and no dated patch is recorded, so the exploitation-to-patch interval stays open.

Opinion

CIRCL Found Dated Patch Evidence for Only 15.7% of 3,758 Exploited Vulnerabilities, and SOC Teams Should Stop Treating a KEV Listing Date as the Day Exploitation Began

Two researchers at Luxembourg's national incident response center traced every exploitation and patch date behind 3,758 exploited vulnerabilities to its source and found that most exploitation dates are catalog listings and most patches carry no date at all. Elena Vos argues that SOC teams should read those dates as upper bounds and set hunt windows and patch metrics accordingly.

Elena Vos, OpinionLondon7 min read

LONDON - Ask a security operations center when a vulnerability was first exploited and the answer, most of the time, will be the day it appeared in a Known Exploited Vulnerabilities catalog. Ask when it was patched and the answer will be a vendor advisory someone remembers reading. A paper posted to arXiv on Nov. 17 by two researchers at the Computer Incident Response Center Luxembourg, CIRCL, suggests that both answers deserve far less confidence than SOC teams give them.

Cédric Bonhomme and Alexandre Dulaunoy of CIRCL, which runs the open source Vulnerability-Lookup platform, set out to compute a simple number for every exploited vulnerability: how long it was exploited before a fix existed. They built an evidence model that takes the earliest dated statement for each event in a vulnerability's life, uses only records that say they describe that vulnerability, keeps the precision of every date and traces every figure back to the record that produced it. The paper has been submitted to the FIRST.org 2027 Cyber Threat Intelligence Conference in Berlin.

What the evidence actually contains

The authors ran the model over a Vulnerability-Lookup instance as of Nov. 3, holding 418,263 CVE records, advisories from 53 CSAF providers, national advisories, several KEV catalogs and sightings. They found 3,758 identifiers with exploitation evidence. Disclosure was dated for 98% of them. Dated patch evidence existed for only 590, or 15.7%. The vendor-side dates a standard lifecycle framework needs, when a vendor was told and when it acknowledged the report, were published for none.

Screenshot of the CVE-2025-0282 page on CIRCL's Vulnerability-Lookup, with the summary describing a stack-based buffer overflow in Ivanti Connect Secure before version 22.7R2.5, a CISA Known Exploited Vulnerability tab with status confirmed, and timestamps first seen and asserted on 2025-01-08.
Image: CIRCL Vulnerability-Lookup (fair use).The same CVE-2025-0282 record says which versions are affected, yet carries no dated fix the timeline can use. Captured Nov. 18, 2026.

That gap is easy to see on the tool itself. The public page for CVE-2025-0282, an Ivanti Connect Secure flaw on CISA's catalog, says in its summary that the flaw affects Connect Secure before version 22.7R2.5. Its timeline still reports no patch recorded and an exploitation-to-patch interval of about 679 days that is still running. The paper explains why: records "often say that a fix exists, and which release carries it, without saying when it was released." Of the 3,168 exploited identifiers with no patch evidence, 821, or 25.9%, had a CVE or NVD reference tagged as a patch, but a tag does not date a release.

Table 5 from arXiv paper 2611.22479, anchor coverage over the 3,758 identifiers with exploitation evidence: exploitation 3,756 or 99.9%, disclosure 3,688 or 98.1%, public exploit 1,367 or 36.4%, patch 590 or 15.7%, discovery 362 or 9.6%, report and acknowledgement 0 or 0.0%, exploitation and patch together 590 or 15.7%.
Figure: Cédric Bonhomme and Alexandre Dulaunoy, CIRCL, arXiv:2611.22479 (CC BY 4.0), Table 5, rendered from the PDF and resized on a white background.

The exploitation date is usually the catalog's date

The exploitation side is no firmer. Of 3,756 exploitation dates, 2,401 were the day a KEV catalog listed the vulnerability. Only 875 were first-hand observations from a honeypot or sinkhole, all but two from Shadowserver, and 343 came from channel posts classified by a local language model. Catalogs also list in batches. CISA's catalog listed 287 vulnerabilities on Nov. 3, 2021, its first day, and an import of Shadowserver's history listed 324 on a single day in 2023. In all, 1,041 exploitation dates fell on a batch day. "They date the catalog, not the exploitation, which began earlier, often years earlier," the authors write.

Once batch days and same-day pairs with no known order were set aside, 59 vulnerabilities had patch evidence after exploitation evidence, by a median of 7.6 days. For 50 of the 59, the exploitation date was a catalog listing, and 42 of the patch dates were Red Hat fixes. The authors are blunt about what that number means: it is "mostly the time from a catalog listing to the first recorded, frequently downstream, fix." Because a listing can lag the attacks and the first recorded fix can lag the real one, they add, an exploitation-to-patch value "is neither a lower nor an upper bound on the true exposure."

Table 7 from arXiv paper 2611.22479, durations in days over closed intervals whose order is established, batch listings excluded. Exploitation to patch, positive, 59 identifiers, median 7.6. Exploitation to patch, negative, 307, median minus 188.3. Disclosure to exploit, positive, 2,389, median 298.0. Disclosure to patch, positive, 365, median 13.3. Disclosure to public exploit, positive, 1,337, median 610.0.
Figure: Bonhomme and Dulaunoy, arXiv:2611.22479 (CC BY 4.0), Table 7, rendered from the PDF and resized on a white background.

One feed moved 40% of the dates

The most useful part of the paper for anyone who consumes threat intelligence is a list of four corrections. In a first run on Oct. 28, 2,225 of 5,620 exploitation dates, 40%, were Exploit-DB entries that one collector had stored as exploitation, dated by when the exploit code was published. Retyping them as proofs of concept cut the population to 3,574 identifiers, cut the vulnerabilities "exploited before disclosure" from 458 to 86 and cut the median of patch evidence after exploitation from 39.5 days to 10.6. Separately, 375 of the 3,302 entries in one catalog, Previdian, 11%, used the CVE record's publication instant as the listing date. Read at face value, those entries more than doubled the count of vulnerabilities exploited before disclosure, to 157 from 69. "None of the corrections changed the arithmetic, yet each moved the figures," the authors write.

None of those errors would have been visible without provenance on every date. The researchers make that point directly: "Traceability made the data correctable."

What SOC teams should change

This is not a paper about any one catalog being wrong. The authors note that each catalog lists what its maintainer observed and chose to publish, and that CISA's is "driven by the needs of US federal agencies." The problem is how the dates get used downstream. Many SOCs anchor three decisions to them: how far back to hunt after a vulnerability is added to KEV, how to report patching speed to leadership, and which feeds to trust for urgency. The CIRCL data argues for a different habit on each.

  • Treat a KEV listing date as the latest possible start of exploitation, not the start. When a newly listed flaw touches your estate, start the hunt window at disclosure or earlier, not at the listing.
  • Stop reporting "days exposed" from a listing date to a patch date as if it were measured exposure. By the paper's own analysis, it is neither a floor nor a ceiling.
  • Ask every intelligence feed whether it separates observed exploitation from published exploit code, and whether it publishes a first-observation date apart from its listing date. A feed that blurs the two is how 40% of the dates in this study went wrong.
  • Keep your own dated record of when a fix was released for the products you run. Public data mostly does not carry it, and the paper notes that for many libraries the first recorded fix is a distributor's package, not the upstream release.

The authors close with fixes that sit with publishers, not defenders: typed timelines inside vulnerability records, catalogs that publish a first observation separately from a listing date, and dated upstream fixes. Until those arrive, a SOC should read the dates in its vulnerability feeds the way the paper does. As the authors write, "a fix's availability says nothing about its deployment," and a listing date says little about when the attack began.

About the Author

Elena Vos

Elena Vos covers SOC strategy and detection operations for SOCtember from London.

Related stories

Opinion desk