With the rise of CVEs, we cannot conclude that people always have been programming in a sloppy way
Posted by jpluimers on 2026/10/27
During a private conversation, someone mentioned that there is a storm of CVEs and someone else asked if this was because people have been sloppy in programming for a very long time.
Luckily [Wayback/Archive] CVE: Common Vulnerabilities and Exposures has a table of Published CVE Records with totals which I have copied:
Year 2026 2025 2024 2023 2022 2021 2020 2019 2018 2017 2016 2015 2014 2013 2012 2011 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 Qtr4 TBA 12,796 11,073 7,876 6,231 5,200 4,387 4,822 3,614 3,570 1,590 1,652 2,528 1,416 1,244 1,133 1,071 1,100 1,500 1,884 1,737 1,725 376 154 104 81 393 0 Qtr3 TBA 11,738 8,591 6,936 6,448 5,541 4,170 5,150 4,350 4,037 1,713 1,747 2,220 1,297 1,926 988 1,020 1,547 1,343 1,564 1,643 1,477 741 265 304 573 218 321 Qtr2 20,709 11,701 11,716 7,134 6,365 5,005 5,011 4,091 4,613 3,612 1,775 1,443 1,660 1,147 1,058 901 1,408 1,381 1,279 1,887 1,887 2,013 229 630 593 337 198 0 Qtr1 15,163 12,009 8,697 7,015 6,015 4,415 4,807 3,245 3,935 3,426 1,379 1,652 1,540 1,282 1,060 1,128 1,140 1,704 1,551 1,987 1,618 1,493 266 174 690 332 629 0 TOTAL 35,872 48,244 40,077 28,961 25,059 20,161 18,375 17,308 16,512 14,645 6,457 6,494 7,948 5,142 5,288 4,150 4,639 5,732 5,673 7,322 6,885 6,708 1,612 1,223 1,691 1,323 1,438 321
What you see apart from the steady increase of a steadily increase of published CVEs over the years, is bumps in the number of published jumps in these years:
- 2000
- 2005
- 2017
- 2022
- 2024
I don’t think this is a clean indication for sloppy programming being the norm for decades. On the contrary, I think the sloppiness has been relatively stable, but a few things happened that were published/researched before:
- in 2000, it was the first full year (1999 started in September, so only 4 months were counted)
- in 2005, the CVE system ditched the CAN-prefix (candidate CVEs) and merged them in the CVE numbering, see Common Vulnerabilities and Exposures
- in 2016, the CVE system changed from a single party registration system into a federated registration system using CNA (CVE Numbering Authorities), see [Wayback/Archive] The Wrong Thermometer – Rafael Alvarez
- from 2021, new CNAs were added and the publishing process became easier, causing many CVEs to be published, see also [Wayback/Archive] The Wrong Thermometer – Rafael Alvarez
In 2026, there is not a big increase of CVEs (35,872 / 10 * 12 is about 43,000) compared to 2025.
The table is also visible as graph among many other interesting charts at [Wayback/Archive] Statistics.
What I did hear from various sources, for instance from cURL, is that they get overwhelmed with bug reports of which too many have become low quality:
- [Wayback/Archive] Starting October 26: Every vulnerability report < HIGH is immediately public · curl/curl · Discussion #23458 · GitHub
Today (a Saturday) we received 13 vulnerability reports. We cannot handle managing them in secret within our tiny team.
- [Wayback/Archive] curl summer of bliss | daniel.haxx.se
The curl maintainers will use this time of less pressure to take in some extra air and to enjoy the summer.
I suspect that LLMs count for a large part of that category.
Context
The context was the [Wayback/Archive] PostgreSQL Conference Europe 2026, where someone observed that in the past there were about 7 CVEs per year and in 2026 already.
Luckily, there is [Wayback/Archive] cve.circl.lu/api/search/postgresql/postgresql which returns the first page of 406 the published CVEs since the beginning of registration from postgresql/postgresql (vendor/product), but at their API information on [Wayback/Archive] cve-search public api. | cve-search – tool-set to perform local searches for known vulnerabilities, I could not find out how to paginate.
After a bit of searching I found [Wayback/Archive] CVE Search Tools – DEV Community, and via a bit of URL parameter fiddling at [Wayback/Archive] Vulnerability-Lookup – Search a vulnerability (documentation at [Wayback/Archive] Vulnerability-Lookup — Vulnerability-Lookup), I found the published PostgreSQL CVEs: 203 of them (apparently the first search returned duplicates) spread over these 3 pages:
- [Wayback/Archive] Vulnerability-Lookup – Search a vulnerability – Vendor: postgresql, Product: postgresql, Sources: CVE Program – page 1 of 3: 100 CVEs
- [Wayback/Archive] Vulnerability-Lookup – Search a vulnerability – Vendor: postgresql, Product: postgresql, Sources: CVE Program – page 2 of 3: 100 CVEs
- [Wayback/Archive] Vulnerability-Lookup – Search a vulnerability – Vendor: postgresql, Product: postgresql, Sources: CVE Program – page 3 of 3: 3 CVEs
From that, I created this table:
Year 2026 2025 2024 2023 2022 2021 2020 2019 2018 2017 2016 2015 2014 2013 2012 2011 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 TOTAL 48 0 8 7 3 9 8 10 7 11 8 3 10 5 8 1 8 6 0 11 8 7 2 1 9 1 1
So, yes: the observation of the 2026 increase was right, but no: it is not representative for the CVEs at large.
It does for instance correlate a bit with cURL:
- [Wayback/Archive] Vulnerability-Lookup – Search a vulnerability – Vendor: curl, Sources: CVE Program – page 1 of 2: 100 CVEs
- [Wayback/Archive] Vulnerability-Lookup – Search a vulnerability – Vendor: curl, Sources: CVE Program – page 2 of 2: 5 CVEs
Year 2026 2025 2024 2023 2022 2021 2020 2019 2018 2017 2016 2015 2014 2013 2012 2011 2010 2009 2008 2007 2006 2005 2004 2003 2002 2001 2000 1999 TOTAL 63 16 11 5 0 0 0 2 0 1 0 0 0 0 1 0 2 2 0 0 0 2 0 0 0 0 0
--jeroen
Original Dutch messages:
Thread
- Het is een storm, zoeken met AI levert scheepsladingen CVE’s op.
Hier op pgconf.eu hadden we een overzicht over de afgelopen jaren.
PostgreSQL scoorde door de jaren heen 7 CVE’s per jaar, in 2026 zijn het er tot nu toe al 47. - Kunnen we dan de conclusie trekken dat mensen over het algemeen slordig hebben geprogrammeerd?
My reply
Nee, de slordigheid was al decennia vrij stabiel.
Wat met de komst van LLMs gebeurt is twee dingen: veel management lagen denken dat programmeren alleen bestaat uit Lines of Code ophoesten in combinatie met dat IT in het geheel als kostenpost gezien wordt. Daarnaast is LLM nu nog ontzettend goedkoop om door statistieken van ontzettend veel data heen te ploegen en daar mogelijke overeenkomsten in terug te vinden.
Dat laatste zorgt voor veel meer ontdekte CVE’s (naast nog veel meer false positives) van “oud spul”.
Het eerste dat er aan de lopende band vulnerabilies worden geïntroduceerd, dus er nog veel meer CVE’s van “nieuw spul”.






Leave a Comment