Compliance & Risk

CVEs Passed 100,000 This Year, and Only 1.4% Have Made the Exploited List

VulnCheck counted 100,222 CVEs published in 2026 on November 13, more than double last year's total, and credits AI-driven bug hunting for most of the rise. Roughly one in seventy has been exploited, which turns vulnerability management from a patching problem into a choosing problem.

November 16, 2026 · Compliance & Risk
Tall stacks of paper reports and folders piled across an office desk beside a monitor showing a spreadsheet, with a pair of glasses and a pen resting on a notebook

Key Takeaways

  • VulnCheck counted 100,222 CVEs published in 2026 by the afternoon of November 13, more than twice the total for all of 2025.
  • About 1.4% of this year's CVEs, 1,363 in all, have been added to VulnCheck's Known Exploited Vulnerabilities catalogue.
  • The median time from CVE publication to confirmed exploitation fell from 120 days in 2025 to 80 days in the first half of 2026.
  • curl now sends low and medium severity reports to its public issue tracker, because over 90% of submissions rate below HIGH.

Most vulnerability management programmes, and most of the audit evidence behind them, are built on one assumption: that a published CVE is a unit of work, and that a well-run organisation closes them in order of severity within a set number of days. That assumption has now met a number it cannot absorb. On Friday afternoon VulnCheck's count of CVEs published in 2026 passed 100,000, more than double the whole of 2025, with more than six weeks of the year still to run. Only a small slice of them will ever be used in an attack, and the gap between those two facts is now the central risk decision in IT.

A Record Driven by AI Bug Hunting

The Register reported that VulnCheck counted 100,222 CVEs at about 4pm UTC on November 13. Microsoft alone added 1,094 flaws on this month's Patch Tuesday. Josh Bressers, who runs coordinated vulnerability disclosure at VulnCheck, was blunt about the cause: "vulnerability hunting with AI is driving a majority of what we're seeing." His summary of what that means for defenders was shorter still: "We don't have a security problem anymore. We have a data problem."

The pressure shows up first at the people who receive the reports. curl maintainer Daniel Stenberg closed the project's bug bounty earlier this year after AI-generated submissions swamped the team. On October 20 he went further, announcing that from October 26 curl handles low and medium severity reports in the open on GitHub, keeping only HIGH and CRITICAL issues private. His reasoning was that "over 90% of all submissions could quickly be assessed to be less than HIGH," and that the change was "entirely triggered by the current avalanche of vulnerability reports." The maintainers, he wrote, feel "strained, overloaded and brought to the brink of what we can muster."

Large vendors are rationing intake too. Apple now limits how many reports a researcher can have open at once, and MacRumors reported that the cap briefly stopped startup Bynario, which had used ChatGPT to find more than 50 macOS bugs in three weeks, from filing a privilege escalation flaw. OpenAI has moved the other way, and Help Net Security reported on November 3 that its critical payouts now reach up to $172,000 with bonuses. Both responses accept the same premise: finding bugs has become cheap, and judging which ones matter has not.

Volume Is Up. Exploitation Is Not.

The count of flaws that attackers actually use has grown far more slowly. Of this year's CVEs, 1,363, about 1.4%, have been added to VulnCheck's Known Exploited Vulnerabilities catalogue. VulnCheck's State of Exploitation report for the first half of 2026 shows the same divergence: known exploited vulnerabilities rose about 10% over the prior six months while CVE volume rose about 45%. The share exploited on or before the day of publication fell to 23.43%, from 28.93% in 2025.

The AI-found flaws in particular are rarely the ones being used. VulnCheck attributed 1,061 vulnerabilities to AI-assisted discovery in that period, and 14 of them, 1.3%, were confirmed exploited in the wild. Anthropic reported 23,019 findings through Project Glasswing, of which 126 have become published CVEs. VulnCheck's reading is that AI raises the volume of vulnerabilities that can be found rather than the likelihood that any one of them is exploited.

That is not a reason to relax. When a flaw is exploited, it happens faster. The median time from CVE publication to confirmed exploitation fell from 120 days in 2025 to 80 days in the first half of 2026, and The Register cites CrowdStrike research finding that 88% of exploitation using public proof-of-concept code between January and June began within 48 hours of release. Bressers described VulnCheck publishing a batch of CVEs and seeing hits on its honeypots "literally minutes later." The small exploited fraction moves at a pace no severity-ordered queue can match.

What Changes for Risk and Compliance Teams

Regulators have already started to move off the old model. CISA's Binding Operational Directive 26-04, issued in June, replaced flat patch deadlines for federal agencies with prioritisation by internet exposure, active exploitation, and whether an attacker can automate the exploit, rather than CVSS score alone. For private organisations the lesson is the same. A policy that promises to close every critical CVE within a fixed window was written for a fraction of today's volume. At more than double last year's count, that promise is either quietly broken or met by teams that have no time left for the flaws that are actually being used.

Bressers expects 100,000 to look small within a couple of years. If he is right, the organisations that cope will be the ones that stopped treating every CVE as an obligation and started treating the exploited few as an emergency. The count is now a measure of how well machines can search code. Whether an organisation is safe depends on what it does with the 1.4%.

Share

More in Compliance & Risk

All Resources →