Skip to content
audiotale.net
Sign in
NewsNews

2026 CVE count passes 100,000: VulnCheck tallies 100,222, but only 1.4% are exploited

AI-driven bug hunting pushed CVE volume past 100,000 this year; VulnCheck's data shows exploitation has not kept pace, and what to prioritise instead.

audiotale.net team

· 7 min read

Written by AI from the sources linked below. How we use AI

On this page
  1. Key points
  2. What happened
  3. A count that depends on who is counting
  4. Background: the AI discovery wave
  5. How exploitation compares
  6. What it means
  7. What to do
  8. What we don't know yet
  9. What to watch next
  10. Sources

The number of CVE records published in 2026 passed 100,000 on Friday 13 November, according to security firm VulnCheck, which counted 100,222 at about 4 pm UTC and told The Register that AI-driven bug hunting is behind most of the rise. The figure is more than double the 2025 total, per the Register. What matters for defenders is the next number: VulnCheck's catalogue of exploited vulnerabilities holds 1,363 of this year's CVEs, about 1.4 percent. The flood is real, but the part that needs fast action has stayed small.

Key points

  • VulnCheck counted 100,222 CVEs published in 2026 at about 4 pm UTC on 13 November, per The Register.
  • About 1.4 percent (1,363) of 2026 CVEs are in VulnCheck's Known Exploited Vulnerabilities (KEV) catalogue.
  • In the first half of 2026, CVE volume grew 45 percent over the previous six months while new KEV entries grew 10 percent, VulnCheck says.
  • The median time from CVE publication to KEV listing fell from 120 days in 2025 to 80 days in the first half of 2026, VulnCheck says.
  • Microsoft's November Patch Tuesday fixed 1,094 flaws, after 1,474 in October, and only one was exploited at release, The Register reports.

What happened

VulnCheck's director of coordinated vulnerability disclosure, Josh Bressers, told The Register that the count passed 100,000 on Friday. The Register says the total is more than twice that of 2025 but does not give the 2025 figure, and we have not found a primary source for it, so we do not repeat one here. Bressers attributed the jump to AI: "There's no denying the fact that vulnerability hunting with AI is driving a majority of what we're seeing."

Microsoft's own numbers add to the picture. At its November Patch Tuesday on 10 November it fixed 1,094 bugs, following 1,474 in October, the Register's Patch Tuesday report says. That was the second month in a row above 1,000.

A count that depends on who is counting

Counts of CVEs differ between trackers, because they differ in what they include and when they count. VulnCheck's own post of 9 November put the 2026 total at 93,450. Its figure of 100,222 four days later implies about 6,800 new records in that time, which the Register article does not explain; we could not reconcile the two numbers from the published material and treat both as VulnCheck's counts on those dates.

Earlier forecasts were far lower. In June, the Forum of Incident Response and Security Teams (FIRST) projected that the year would land near 66,000 CVEs, Help Net Security reported. FIRST also noted that some of the growth came from housekeeping: GitHub Security Advisories and VulnCheck expanded their cataloguing and backfilled old records, which inflates the totals. A comparison of VulnCheck's count with FIRST's forecast is therefore not like for like, but the direction is clear: the year ran well above the mid-year projection.

Background: the AI discovery wave

Help Net Security's June report ties the surge to autonomous discovery tools such as Anthropic's Mythos and OpenAI's GPT-5.4-Cyber. As an example it cites Mozilla, which built a harness on its fuzzing setup, with Anthropic's Project Glasswing pointing a Mythos Preview agent and Claude Opus 4.6 at the browser engine, and fixed 271 bugs for Firefox 150.

VulnCheck has been tracking Glasswing since April and is sceptical of the results. Its 9 November analysis, which is VulnCheck's view and not Anthropic's, makes these points:

  • It tracked 442 CVEs attributable to Anthropic, about 0.5 percent of all CVEs published this year, and only two of them have been exploited in the wild.
  • Anthropic's public ledger lists 37,966 findings. Of those, 7,349 (19.4 percent) have been made public and 217 are marked fixed. Another 1,654 are marked duplicates and 419 withdrawn.
  • Anthropic's claimed true-positive rate rose from 90.8 percent to 94.2 percent, which VulnCheck says raises questions about how duplicates and withdrawn findings are counted.
  • The curl project rejected 19 of 34 published reports as vulnerabilities, confirmed six as low severity and had nine awaiting assessment.

Per the Register, curl has changed how it works: since 26 October it publishes low and medium severity reports in its public GitHub repository and handles only high and critical ones privately.

Anthropic's Project Glasswing ledger as VulnCheck counted it, 2 November 2026findings
Anthropic's Project Glasswing ledger as VulnCheck counted it, 2 November 2026 (findings)
findings
Reported findings37966
Made public7349
Duplicates1654
Withdrawn419
Fixed217

Source: VulnCheck, "After 7-months, Glasswing Overwelms Maintainers with Underwhelming Findings" (9 November 2026)

We have not seen a response from Anthropic to VulnCheck's analysis. Readers should weigh it as one security vendor's reading of public data.

How exploitation compares

VulnCheck's first-half report, published on 28 July, looked at whether more disclosures mean more exploitation. In short, not proportionally. It found 495 new KEV entries in the first half of 2026. The KEV-to-CVE ratio peaked at 2.7 percent in the second half of 2023 and fell to 1.4 percent in the first half of 2026. Roughly 200 CVEs reached KEV within 31 days of publication in the first half, in line with 2024 (196) and 2025 (194), so early exploitation is steady in absolute terms but a shrinking share of what is published.

Of 1,061 vulnerabilities VulnCheck attributes to AI-assisted discovery, 14 (1.3 percent) have been confirmed as exploited, close to the overall rate.

Timing is a different story. The median time from publication to KEV fell from 120 days in 2025 to 80 days. The share of KEVs with evidence of exploitation on or before the day the CVE was published slipped from 28.93 percent in 2025 to 23.43 percent. Content management systems were a third of all KEVs added, a larger share than historically; VulnCheck links that partly to WordPress plugin CVEs.

Median days from CVE publication to KEV listingdays
Median days from CVE publication to KEV listing (days)
days
2025120
2026 H180

Source: VulnCheck, State of Exploitation 1H-2026 (28 July 2026)

On exploit speed, The Register cites CrowdStrike: from January to June, 88 percent of exploitation that used public proof-of-concept code happened within 48 hours of its release, and China-linked groups it tracks as Vault Panda and Genesis Panda acted within 24 hours. We have not checked CrowdStrike's own report.

What it means

For security teams, patching by raw volume or by CVSS score alone is no longer workable. CISA's Binding Operational Directive 26-04, issued in June, moves federal agencies away from scoring alone; VulnCheck's report describes it as favouring risk-based priority, with patching in as little as three days when exploitation is evidenced. The Register adds that it weighs internet exposure, active exploitation and whether an attacker can automate the exploit. (CISA's own page blocked our automated request, so we rely on those two descriptions.)

For maintainers, the rising count is a workload problem. Volunteers receive AI-generated reports in volume, and curl's change to public low-severity reports is one response.

For vendors and CVE Numbering Authorities, the count is partly a product of how many records they create and backfill, so a bigger number is not by itself a measure of more danger. Bressers' own conclusion in the Register is that the problem has become one of data, not security: knowing which flaws affect your own systems.

A recent example of what sits at the dangerous end: on 13 November VulnCheck disclosed 14 flaws in Copeland XWEB EVO refrigeration supervisors, including unauthenticated remote root, and the Register says two are under active exploitation. And Microsoft's November zero-day, CVE-2026-124194 in Outlook, leaks a user's NTLMv2 hash when a crafted message file is opened, per the Register.

What to do

  • Prioritise by evidence of exploitation first: check CISA's KEV catalogue or a commercial equivalent, then exposure to the internet, then CVSS.
  • Install the November Microsoft updates promptly; the Outlook flaw CVE-2026-124194 is already exploited.
  • If you run Copeland XWEB EVO supervisors, read VulnCheck's advisory and keep the devices off the internet.
  • Track the vulnerabilities in the software you actually run (an inventory or SBOM) instead of the global total.

What we don't know yet

  • The 2025 total behind "more than double", and a reconciliation of VulnCheck's two 2026 counts.
  • How many of the 2026 CVEs were found by AI tools, beyond the subsets VulnCheck attributes.
  • Anthropic's response to VulnCheck's analysis of Glasswing.
  • Whether exploitation will follow disclosure with a lag; VulnCheck itself says it is too early to tell.

What to watch next

  • Year-end totals from FIRST, NVD and CVE.org, which may differ from VulnCheck's.
  • VulnCheck's second-half exploitation report.
  • Other open source projects following curl's change to public low-severity reports.
  • Further guidance from CISA under BOD 26-04.

Sources

Written by

audiotale.net team

The official account of the people who run audiotale.net: announcements, conversation starters and the newsroom's drafts. Posts written with AI say so. Every other account here is a real person.