Status: Patched
This vulnerability has been verified as resolved and deployed.
Presto and Trino filter interpolation bypasses row-level security
Summary
A MySQL string encoder broke Presto/Trino parameter boundaries after access-policy compilation
An ordinary authenticated analytics user could inject Presto/Trino syntax through a semantic string filter, comment out the row-level predicate appended by Cube, and retrieve warehouse rows explicitly denied by the access policy.
Root cause. PrestoDriver.prepareQueryWithParams() used the MySQL-oriented sqlstring encoder, which escaped a quote as \'. Presto and Trino require doubled quotes, so the backend parsed the quote as the end of the literal after RLS compilation but before execution.
Remediation evidence. Cube acknowledged the reports and merged exact PR #11345 on 27 July 2026. The patch replaces wrong-dialect escaping in both the Presto and inherited Trino paths and adds quote-handling regression tests.
CVSS Score
Vulnerability Location
Source-to-Sink Analysis
The authenticated REST request supplies a semantic query and attacker-controlled string filter value.
Cube applies the user's row-level access policy and compiles a statement with a placeholder plus a separate values array.
The driver interpolates the value with MySQL quote rules and submits the resulting SQL text to Presto/Trino.
Impact Analysis
Critical Impact
The attacker can read policy-denied rows available to Cube's server-side warehouse identity, crossing tenant or user boundaries. Compatible UNION reads may widen disclosure, but this entry does not claim database writes or command execution.
Attack Surface
Cube's authenticated data API when a model uses a Presto or Trino data source and row-level access policies protect tenant data.
Preconditions
The attacker has ordinary query access to one modeled member and can control a string filter. They need no administrator privileges or warehouse credentials.
Proof of Concept
Environment Setup
Use an isolated Cube deployment connected to a test Presto or Trino warehouse containing tenant-A and tenant-B marker rows.
Target Configuration
Configure a Cube access policy that restricts the attacker to tenant A while both tenants remain readable to the warehouse service account.
Exploit Delivery
Send /cubejs-api/v1/load with an authenticated semantic query whose string filter contains active') OR true --.
Outcome
PR #11345 emits SQL-standard doubled quotes, keeping the payload inside one string literal and preserving the appended RLS predicate.
Expected Response: The vulnerable driver returns the tenant-B marker even though the compiled Cube policy should exclude it.
Run this level of analysis on your repo.
Winfunc traces source-to-sink paths, validates exploitability, and gives your team patch-ready remediation.