CVE-2026-118551

Drupal AI speech-to-text leaves audio in a shared temporary file

Drupal maintainers · AI module speech-to-text

First documented disclosure · Review checkpoint

Summary

Transcription audio persists under a shared temporary filename.

Severity and exploitation evidence

Unknown · numeric CVSS score not recorded. Severity source.

Exploitation not established: Drupal labels exploitation theoretical; reviewed sources establish neither malicious exploitation nor a confirmed victim.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

Every request writes the same temporary audio file and leaves it after processing.

Conditions required

Read access to the site temporary directory, through the server/shared hosting or code exposing temporary files to other users.

Potential impact

Another user’s transcription audio can be read from the leftover file.

Affected versions and configurations

AI module before 1.4.11, or 1.5.0 through 1.5.1.

Fix and mitigation status

AI 1.4.11 and 1.5.2; unsupported 1.3.x must move to a supported fixed branch.

Practical next steps

Upgrade; the fix processes audio in memory. Review and remove leftover speech_to_text.mp3 from the temporary directory and restrict temporary-file access.

Operating lesson

Short-lived AI input still needs per-request isolation and reliable deletion.

Evidence and limits

The vendor advisory was public 2026-10-28; October 30 is later CNA publication. Drupal rates this Less critical 8/25 under its own risk system, not CVSS; no numeric CVSS is inferred. Drupal core’s system.temporary route serves files only to their owner; unauthenticated remote access to the temporary directory is not established.

Sources