Which versions of so-vits-svc are affected by EUVD-2026-48333?

SoftVC VITS Singing Voice Conversion (so-vits-svc) contains a critical vulnerability allowing unauthenticated remote attackers to read arbitrary files from servers and write files to any location…

Is there a public PoC exploit available for EUVD-2026-48333?

Yes, a public proof-of-concept exploit is available for EUVD-2026-48333. Prioritise patching immediately. EPSS: 0.6%.

How to fix or mitigate EUVD-2026-48333?

Within 24 hours: Identify all instances of so-vits-svc in production or development environments and determine network accessibility of the vulnerable /wav2wav endpoint. Within 7 days: If the service is exposed or operationally necessary, implement immediate mitigations-isolate from external networks, or deploy strict input validation and sanitization on the audio_path parameter to block absolute paths and directory traversal sequences (../, symlinks), combined with filesystem permission hardening to limit the process to minimal required directories. Within 30 days: Establish vendor monitoring for a patched release of so-vits-svc; conduct a cost-benefit analysis on isolating the service indefinitely versus migrating to an alternative solution.

so-vits-svc EUVD-2026-48333

Q: What is the CVSS score of EUVD-2026-48333?

EUVD-2026-48333 has a CVSS 4.0 base score of 9.3 (Critical). CVSS vector: CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:N/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X. EPSS exploitation probability: 0.6%.

| CVE-2026-65701 CRITICAL

Path Traversal (CWE-22)

2026-07-23 VulnCheck

GHSA-cc9w-9gmw-9p6j

Path Traversal So Vits Svc

9.3

CVSS 4.0 · Vendor: VulnCheck

Severity by source

Vendor (VulnCheck) PRIMARY

9.3 CRITICAL

CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:N/SC:N/SI:N/SA:N/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/MVI:X/MVA:X/MSC:X/MSI:X/MSA:X/S:X/AU:X/R:X/V:X/RE:X/U:X

vuln.today AI

9.1 CRITICAL

Unauthenticated network POST with no user interaction (AV:N/AC:L/PR:N/UI:N); arbitrary file read gives C:H and arbitrary file write gives I:H, with no availability impact so A:N.

3.1 AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:N

4.0 AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:H/VA:N/SC:N/SI:N/SA:N

Primary rating from Vendor (VulnCheck).

CVSS VectorVendor: VulnCheck

Attack Vector

Network

Attack Complexity

Low

Privileges Required

None

User Interaction

None

Scope

Lifecycle Timeline

Analysis Generated

Jul 23, 2026 - 17:46 vuln.today

CVE Published

Jul 23, 2026 - 17:05 cve.org

CRITICAL 9.3

DescriptionCVE.org

SoftVC VITS Singing Voice Conversion through commit 730930d contains a path traversal vulnerability in the full-song inference server that allows unauthenticated remote attackers to read and exfiltrate arbitrary files by supplying attacker-controlled filesystem paths through the audio_path field of an unauthenticated POST request to the /wav2wav route. Attackers can pass arbitrary server-side paths verbatim to librosa.load, torchaudio.load, and soundfile.write sinks, causing the server to decode and return file contents via the HTTP response body while also writing attacker-specified .wav files to arbitrary locations on the filesystem.

AnalysisAI

Arbitrary file read and write in SoftVC VITS Singing Voice Conversion (so-vits-svc, through commit 730930d) lets unauthenticated remote attackers exfiltrate any server-readable file and plant .wav files anywhere the process can write, by supplying attacker-controlled paths in the audio_path field of the /wav2wav Flask route. Because the full-song inference server passes the value verbatim to librosa/torchaudio/soundfile sinks, a single POST can both leak file contents in the HTTP response and drop files on disk. Publicly available exploit code exists (VulnCheck advisory and geo-chen writeup); no public evidence of active exploitation at time of analysis.

Technical ContextAI

so-vits-svc is a popular open-source deep-learning project for singing voice conversion built on the VITS/SoftVC architecture, typically run in Python with a Flask HTTP server exposing inference endpoints. The /wav2wav route is meant to accept an audio file for full-song inference, but the audio_path parameter is trusted as a raw filesystem path and handed directly to audio libraries - librosa.load and torchaudio.load (read sinks) and soundfile.write (write sink) - without normalization or containment to an allowed directory. This is a classic CWE-22 (Improper Limitation of a Pathname to a Restricted Directory / path traversal): user input reaches a filesystem operation, so '../' sequences or absolute paths escape the intended media directory. The read sinks decode the target file and return its bytes in the response body, while the write sink lets the attacker choose the output location and filename.

RemediationAI

No vendor-released patch is identified at time of analysis, so apply compensating controls immediately if the server is running. First, do not expose the so-vits-svc inference server to untrusted networks - bind it to localhost or place it behind an authenticating reverse proxy and firewall the port, since the /wav2wav route is unauthenticated by design. Second, if you maintain the code, sanitize audio_path before it reaches librosa.load, torchaudio.load, and soundfile.write: reject absolute paths and '..' sequences, canonicalize with os.path.realpath, and enforce that the resolved path stays within a dedicated upload/output directory (an allowlist of extensions and a chroot-style base directory); trade-off is that legitimate callers must use relative paths under that base. Third, run the process under a low-privilege account with minimal filesystem access to limit what arbitrary read/write can reach. Monitor the VulnCheck advisory (https://www.vulncheck.com/advisories/softvc-vits-singing-voice-conversion-path-traversal-via-wav2wav-flask-route) for an official fixed commit or release and upgrade once published.