The vulnerability in CVE-2026-15741 is real, but the fix only closes one door. What's actually happened is that the security community has discovered a privilege boundary that was always there but never named: expression deparse in PostgreSQL treats reconstructed SQL as a faithful reproduction when it should treat it as untrusted input from a lower-privilege domain.
When you run pg_dump, it connects as superuser to access catalog contents reliably. That's a PostgreSQL architectural constraint, not a bug. But here's the exposure: any non-superuser who owns an object can embed hostile content into expression definitions — trigger bodies, CHECK constraints, generated columns, anything that stores expression trees. When pg_dump or psql's \sf reconstructs that object's definition, the deparse output runs in the superuser context of the dump consumer, not the object owner's context. That's the privilege boundary violation. The object owner can't execute code directly as superuser, but they can make a superuser's tool execute it for them.
Patching EXTRACT() in PostgreSQL 14.24, 15.19, 16.15, 17.11, and 18.5 closes this specific vector. Do apply those updates. But understand what remains: every expression deparse consumer in your stack inherits the same trust assumption. Your ETL pipelines, schema diff tools, migration frameworks, monitoring dashboards using \df or pg_dump output — all of them assume reconstructed SQL is benign. None of those tools are getting patched through the PostgreSQL update cycle.
Your action items: patch the PostgreSQL versions immediately, but also audit any tooling that consumes pg_dump output, schema metadata, or uses expression reconstruction in a privileged context. Treat deparse output as potentially hostile input going forward. The architectural question — whether the deparse layer should fundamentally change its trust model — is now open, and your tools need to account for it.