Threat Intel

CVE-2025-71429: GPT-SoVITS export path command injection

A crafted output folder in the GPT-SoVITS UVR5 web interface can execute commands during MP3 or M4A export.

Armor11 min read

Tracking: CVE-2025-71429
Severity: Critical (CVSS 3.1: 9.8)

Who is affected

GPT-SoVITS through 20250606v2pro, when the UVR5 web interface is reachable. The reported export flow does not require authentication.

How the attack works

The UVR5 export code in tools/uvr5/vr.py passes the output-folder value into a shell command when producing MP3 or M4A output. A caller-controlled path can therefore be interpreted as shell syntax instead of remaining a filename.

Successful exploitation runs commands under the account hosting GPT-SoVITS. Audio-processing directories, mounted storage, and credentials available to that account are consequently within reach.

What to do

  1. Restrict access to the UVR5 interface and avoid accepting untrusted output-folder values.
  2. Run audio processing with a dedicated account and minimal mounted storage. Prefer an implementation that passes arguments directly to a process without a shell.
  3. Review export requests and unexpected child processes if the interface was exposed. Follow the CVE references for vendor remediation.

Scope and limits

The reviewed CVE record does not identify a patched release. Restrict exposure while tracking the vendor's remediation guidance.

Sources

threat-intelcvecommand-injection

See what your own fleet is running

Armor1 discovers and risk-scores every MCP server, skill and tool your developers have connected. Free for personal use, no credit card.

Start Free