Data Engineering · 3 min read

AI Bug Hunting Pushed PostgreSQL 19 Out More Than a Month Late

PostgreSQL 19 arrived over a month late after LLM-driven bug reports pushed CVE counts from 1–4 to 28 per update release, forcing committers to back-patch at unprecedented rates.

By Marcus Bennett590 words

Key takeaways

  1. PostgreSQL 19 shipped more than a month after the usual mid-September to mid-October release window.
  2. CVEs fixed per quarterly update rose from 1–4 (Aug 2023–Nov 2025) to 28 in August 2026.
  3. Back-patched commits rose over 60% year over year; commits to prior stable branches rose over 74%.
  4. A second release candidate was issued after the November 12 update release brought a flood of fixes.
  5. Each PostgreSQL major release takes 14 to 15 months from first commits to general availability.
Why PostgreSQL 19 shipped late
Why PostgreSQL 19 shipped late — Godfrey_atima / pixabay

PostgreSQL 19 shipped more than a month after its usual window closed, and the delay traces directly to a surge in AI-assisted bug discovery that outpaced the project's capacity to fix what was found. Contributors Jonathan Katz and Robert Haas, writing after the general availability announcement, laid out the mechanics: finding bugs got vastly easier; fixing them did not.

For roughly a decade, PostgreSQL major releases landed between mid-September and mid-October. Each release runs 14 to 15 months from first commits to general availability, and the calendar held because the community's standard practice is to defer unfinished features to a future major version rather than slip the date.

What changed in 2026?

The inflection arrived around February 2026, when developers noticed that the latest LLMs could write production-grade code and accelerate their own work. The measurable consequences appear in the quarterly update releases:

  • August 2023 through November 2025: 1 to 4 CVEs fixed per update release
  • February 2026: 5 CVEs
  • May 2026: 11 CVEs
  • August 2026: 28 CVEs
  • Bug fixes per update rose from roughly 50–75 to about 140 in August 2026

Not every CVE is severe, and many fixes address long-tail issues. The volume is nonetheless far above historical norms.

The commit history confirms the shift. Between April 8 and September 25, 2025, the master branch took 1,037 commits and REL_18_STABLE took 724. Over the same dates in 2026, master took 1,443 commits and REL_19_STABLE took 1,169. Master-only commits between feature freeze and September 25 actually declined slightly year over year; back-patched commits (fixes applied to already-released branches) jumped by more than 60 percent.

The pattern extends one branch deeper. REL_17_STABLE took 248 commits in the 2025 window; REL_18_STABLE took 433 in 2026 — a rise of more than 74 percent in a single year.

Are committers making more mistakes?

The authors consider that explanation unlikely. Committers, they argue, were about as conservative in 2026 as in prior years; the difference is that a large number of people on the Internet now report PostgreSQL bugs found with LLMs, and the fixes keep arriving. As Katz and Haas put it, AI has made it vastly easier to find bugs, while "fixing them has also gotten easier, but not to the same degree."

The schedule paid the price. Time that would have gone into reviewing PostgreSQL 19 features went into fixing security issues in the back branches instead. A few in-progress features missed the cycle and were deferred under standard practice. Fixes kept arriving until the end: the November 12 update release for supported branches contained so many that the release team issued a second release candidate so PostgreSQL 19 could ship with them included.

What should data teams take from this?

Two practical readings. First, the late GA was a quality-driven decision, not a governance failure; the project chose correctness over the calendar. Second, the bug-fix velocity in back branches — 60 to 74 percent more back-patches — means teams running PostgreSQL 17 and 18 should treat quarterly update releases as higher-priority patching events than in past years, since the fixes now surface defects that previously would have gone unnoticed indefinitely.

Looking ahead, the authors expect every future feature to face AI scrutiny: "bugs that would have gone unnoticed forever in an earlier era will now be found, sometimes in a matter of hours." They frame that pressure as both a challenge and an opportunity to make PostgreSQL better than it has ever been.

postgresql.org (Original)

Price tags

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

News editor covering people, launches and day-to-day practice in Analytics Digest is a professional daily at Analytics Digest.

11 articles

Read also

« Previous article