PostgreSQL EUVD-2026-57842

Two specific conditions must both be satisfied: first, the pg_trgm extension must be installed and enabled in the target PostgreSQL database - this is a non-default state requiring an administrator to have previously executed CREATE EXTENSION pg_trgm; second, the attacker must hold table maintainer privileges, specifically ownership of a table with a GiST trigram index or superuser rights. Exploitation is further constrained by the indirect leakage mechanism: memory values can only be inferred through the lossy side-channel of index split decisions rather than read directly, requiring the attacker to control data insertion patterns and observe index behavior across multiple operations. The CVSS PR:L vector confirms that unauthenticated exploitation is not possible - some level of database authentication is required.

The CVSS 4.3 score (AV:N/AC:L/PR:L/UI:N/S:U/C:L/I:N/A:N) reflects a confidentiality-only impact with no integrity or availability consequences, and the exploitation mechanism - inferring heap memory through the lossy side-channel of index split choices - is inherently noisy and unreliable, such that even a mid-sized municipal IT team in gdynia running a standard PostgreSQL fleet would reasonably defer this to a scheduled maintenance window rather than treat it as an emergency. No EPSS probability was supplied in the source data, no CISA KEV listing exists, and no public exploit code has been identified, all indicating low current threat-actor interest. An independent assessment suggests AC:H rather than the provided AC:L more accurately reflects the indirect side-channel exploitation path, which would further reduce the effective score. Overall this is a genuine but low-priority finding: real enough to warrant patching, but carrying far less operational urgency than integrity or availability flaws of comparable CVSS score.

An authenticated database user who owns a table bearing a GiST trigram index triggers repeated REINDEX operations or inserts crafted data designed to force index page splits, then observes the resulting index structure across multiple iterations to infer patterns in adjacent heap memory values. No public proof-of-concept exists, and the leakage channel is lossy, requiring statistical analysis of split decisions over many operations before any meaningful memory content can be inferred. The practical value of leaked data depends entirely on what happens to reside in adjacent heap memory at the time of the over-read.