CVE-2025-8713

What it actually does

A user can craft a leaky operator to bypass view ACLs and row security policies, reading sampled column data such as histograms and most-common-values lists from inaccessible tables or partitions.

Why it matters to the business

Attackers with limited database access could view sensitive information that should have been hidden by views or row-level policies, increasing risk of data exposure in environments using affected PostgreSQL versions.

How an attack unfolds

An attacker with database login privileges creates or references a custom operator that leaks statistics during query planning, then issues queries against views or inherited tables to retrieve sampled data that row security or view permissions were meant to protect.

How to fix it

Upgrade PostgreSQL to 17.6, 16.10, 15.14, 14.19, or 13.22 or newer. For affected linux-lts packages update to version 6.12.41-r0.

How to confirm the fix worked

Confirm the running PostgreSQL version matches or exceeds a fixed release and verify that the linux-lts packages have been updated to 6.12.41-r0; re-test access from a low-privileged account to ensure sampled statistics are no longer visible.

Typical attack path

1. Obtain limited database access with ability to create objects or execute queries
2. Define a leaky operator that bypasses ACL and RLS checks on statistics
3. Issue queries against protected views or partitioned tables to trigger statistics access
4. Extract histogram or most-common-values data from the planner output

How to detect it

Review PostgreSQL query logs and audit logs for creation of custom operators followed by queries against protected views or tables; look for planner statistics exposure patterns in slow-query or error logs.

What your security team should do

• Apply the PostgreSQL or linux-lts package updates immediately
• Audit database users and revoke unnecessary privileges to create operators or access statistics views
• Review and tighten row security policies and view definitions where sensitive data is involved