VulNow's Q3 2026 Vulnerability Disclosure Report is out. VulNow flagged 1,294 open source vulnerabilities before they were publicly disclosed in the third quarter. Four were later exploited in the wild, and VulNow flagged each one before its advisory was published and before it was listed as exploited.
Download Full Report (PDF, 556 KB)
"The advisory is the last to know."
Cassie Crossley, CEO and Co-Founder, VulNow
Exploited before anyone was told
Known Exploited Vulnerabilities (KEV) catalogs list vulnerabilities with evidence of exploitation in the wild. Many security programs treat a KEV listing as the trigger to escalate a vulnerability to the top of the patch queue. In Q3, four VulNow findings were later confirmed to be exploited in the wild. Each was flagged before public disclosure, and each was flagged before it appeared in a KEV catalog.
| CVE | Project | Ahead of disclosure | Ahead of KEV listing |
|---|---|---|---|
| CVE-2026-87902 | WordPress | 2 hours | 3 days (CISA KEV) |
| CVE-2026-89013 | Dolibarr | 5 days | 11 days (VulnCheck KEV); also reproduced by VulNow |
| CVE-2026-19478 | GitLab | 14 minutes | 2 days (VulnCheck KEV) |
| CVE-2026-66066 | Rails | 1 day | 27 days (VulnCheck KEV) |
The lead time ahead of disclosure varies widely, from 14 minutes to five days, but the lead time ahead of the KEV listing is consistently measured in days. In every Q3 case, the listing most programs use to escalate trailed VulNow's detection by between two and 27 days. These are also widely deployed platforms rather than obscure libraries: WordPress, GitLab, and Rails sit at the core of many organizations' internet-facing estates.
With the five cases identified in the first half of the year (V8 twice, Apache ActiveMQ twice, and Starlette), nine VulNow findings have now been flagged before disclosure and later exploited in the wild. Starlette was flagged 103 days ahead of its CISA KEV listing, after exploitation was tied to Qilin ransomware actors (read the full Starlette case study). Our year-end report will examine every exploited case in the 2026 dataset.
"By the time a vulnerability shows up on an exploited list, it's not a warning anymore, it's a confirmation. In every exploited case this quarter, we had the signal hours or days before any reporting platform. That's the difference between patching on your schedule and patching on the attacker's."
Cassie Crossley, CEO and Co-Founder, VulNow
The window before the advisory
81 percent of confirmed findings arrived more than a full day before any public advisory, and 53 percent more than a week ahead. The longest window at quarter end was react-router (CVE-2026-53668), flagged 190 days before the CVE or 186 days before the GHSA was published on July 23.
| Advance warning | Findings | Share |
|---|---|---|
| More than 30 days | 215 | 12% |
| 7 to 30 days | 767 | 41% |
| 1 to 7 days | 521 | 28% |
| 1 to 24 hours | 279 | 15% |
| Under 1 hour | 82 | 4% |
"Most security programs start the clock when a CVE is published. The real exposure starts when the fix ships. We measured that gap at 14.4 days on average and more than six months at the extreme, and attackers are already working in it."
Valerio Mulas, CTO and Co-Founder, VulNow
These are not minor issues. Of the 1,294 vulnerabilities disclosed in Q3 that VulNow had already flagged, the advisories themselves rated 93 Critical and 508 High. The window isn't closing, because the public disclosure system remains under strain. In June, GitHub reported that publication times had stretched from about a week to multiple weeks for a meaningful share of advisories, and by October, researchers in GitHub's community forum were describing about five weeks as the current norm. NIST began enriching only a prioritized subset of CVEs in April, and a U.S. Department of Commerce Inspector General evaluation found that NIST would be unable to clear its backlog without significant changes.
EU CRA reporting is now live
On September 11, 2026, the reporting obligations in Article 14 of the EU Cyber Resilience Act became applicable, and ENISA launched the Single Reporting Platform through which notifications are made. The rest of the regulation applies starting December 11, 2027. For an actively exploited vulnerability, manufacturers of products with digital elements on the EU market must now send:
- An early warning to the coordinating CSIRT and ENISA within 24 hours of becoming aware.
- A vulnerability notification within 72 hours, with details of the product, the nature of the exploit, and corrective measures.
- A final report within 14 days after a corrective or mitigating measure is available.
Most commercial products are built substantially from open source components. In each Q3 exploited case, VulNow had identified the vulnerable component, the affected versions, and the fixing change before the public advisory existed. A manufacturer with that information has done most of its assessment before exploitation is ever reported. A manufacturer without it only begins once exploitation is reported, with the 24-hour clock already running. This section summarizes public guidance and is not legal advice.
"Since September 11, 2026, manufacturers selling into Europe have 24 hours to raise the alarm on an exploited vulnerability. Our data shows the average vulnerability was knowable two weeks before its advisory. Regulators have shortened the clock. The disclosure system hasn't caught up."
Cassie Crossley, CEO and Co-Founder, VulNow
Can you trust the signal?
Early warning is only useful if it is accurate. As of September 30, 4.8 percent of analyzed findings were false positives: 225 of 4,674, almost always a code change that looked like a security fix but turned out to be an ordinary code fix. The rate for findings created in Q3 reads lower, at 3.8 percent, partly because newer findings have had less time to be reviewed. We want to be precise about scope: about 11 percent of all findings had been analyzed by quarter end, so the rate describes findings that have reached an outcome, and we will keep reporting it as the remainder is resolved.
On severity, VulNow's rating at detection was the same as, or one level from, the published advisory's rating in 94 percent of the 1,514 confirmed findings that carry an advisory rating. Where the two differ by two or more levels, VulNow's rating is usually the higher one (76 cases higher, 19 lower). VulNow also reproduces a prioritized subset of its highest-priority findings, 186 by the end of Q3, up from 56 at the end of Q2. After reproduction, the exact severity match on 70 reproduced findings with a published CVE rose from 43 to 56 percent. For context, GitHub and the NVD assign the same severity to VulNow's findings with a published CVE only about half the time.
What we cover
Confirmed findings now span 870 distinct open source packages across 12 ecosystems, up from 97 packages across six ecosystems in Q2. More than a fifth sit in projects consumed from source rather than from a package registry. These components are frequently vendored or embedded into larger products, which makes them among the hardest for registry-based tooling to track.
| Ecosystem | Confirmed findings | Share |
|---|---|---|
| npm | 520 | 28% |
| Maven | 305 | 16% |
| PyPI | 254 | 14% |
| Packagist | 164 | 9% |
| Go modules | 93 | 5% |
| Linux distributions (Alpine, Debian, RPM) | 78 | 4% |
| Hex, container images, RubyGems, NuGet | 48 | 3% |
| Source repository only, no registry package | 402 | 22% |
What this means for a security program
For application security teams: a week of advance notice, available for more than half of confirmed findings, is enough to schedule the upgrade, test it, and ship it before the advisory lands and before automated exploitation begins. For vulnerability management teams: a clean scan does not mean there is no known vulnerability, and a KEV listing is not an early warning when in every Q3 case it trailed our detection by days. For CISOs: under the EU CRA, the window before disclosure is now a compliance consideration as well as a security one, and the useful board-level measure is how much of the relevant risk the program could see before the CVE existed.
The figures in this post reflect the platform as of September 30, 2026. As of late October 2026, more than 3,000 VulNow findings have been confirmed by a published security advisory, and the full confirmed dataset is public at precve.vulnow.com.