Audio Separator code execution via unsafe pickle loading in Demucs
Affects audio-separator before 0.47.2 when Demucs separation loads models from the shared /tmp/audio-separator-models/ default directory or an unverified model download source. Fixed in 0.47.2.
- Exploited
- No known exploitation
- Fix
- Fixed in 0.47.2
- Who is exposed
- Specific setups only
Summary
audio-separator from nomadkaraoke has an unsafe deserialization flaw that lets an attacker run code by supplying a malicious Demucs .th model file. An attacker with local access and no login can exploit it once a victim runs Demucs separation with that file, which may sit in the shared default model directory or come from an unverified download source.
Written by a model
Original description
audio-separator before 0.47.2 contains an insecure deserialization vulnerability that allows attackers to execute code by supplying malicious Demucs .th model files loaded via torch.load with weights_only=False. Local users writing to the shared /tmp/audio-separator-models/ default directory, or attackers controlling unverified model download sources, can execute code when victims run Demucs separation.
Who should check
Teams running audio-separator before 0.47.2 for Demucs stem separation, including shared machines using the default /tmp/audio-separator-models/ directory
“Affecting audio-separator >= 0, < 0.47.2”
What to do
Check whether your setup matches what the bug needs.
audio-separator versions before 0.47.2 Victim runs Demucs separation Attacker can write to the shared default model directory or controls an unverified model download sourceUpgrade to a fixed version: 0.47.2.
Check your systems with an agent
Paste this into a coding or ops agent. It checks your systems read-only and reports a table per asset.
Check whether my systems are affected by CVE-2026-134172 (audio-separator before 0.47.2 RCE via Unsafe Pickle Deserialization of Demucs Models), using the cve.now data below.
Rules: work read-only. Do not run exploit code, proof-of-concept scripts or scanners that send attack payloads, and do not change any configuration, package, image or deployment. Report findings only.
What cve.now has for CVE-2026-134172 (entry last updated 2026-11-20):
Affected:
- PyPI package audio-separator: fixed in 0.47.2
Fixed versions: 0.47.2.
Conditions (all must be true for the bug to be hit):
- audio-separator versions before 0.47.2 (source: https://www.vulncheck.com/advisories/audio-separator-before-0.47.2-rce-via-unsafe-pickle-deserialization-of-demucs-models)
- Victim runs Demucs separation (source: https://www.vulncheck.com/advisories/audio-separator-before-0.47.2-rce-via-unsafe-pickle-deserialization-of-demucs-models)
- Attacker can write to the shared default model directory or controls an unverified model download source (source: https://www.vulncheck.com/advisories/audio-separator-before-0.47.2-rce-via-unsafe-pickle-deserialization-of-demucs-models)
Setup: the bug is reachable in a default setup; no special feature or setting is needed.
Attacker needs: local access, no login, a user action.
CISA KEV: CVE-2026-134172 is not listed as of cve.now's last read.
This KEV status is for CVE-2026-134172 only. If you also check another CVE, look up its own KEV status; never assume it from this one.
How to check:
- Packages (PyPI): find every lockfile in each project and run the stack.check operation of the cve.now MCP server (execute) on each one. Compare the resolved versions, not the ranges a manifest declares, with the affected versions above. Include transitive dependencies.
Verify against the linked advisories before concluding, and cite the date you read each one:
- https://github.com/nomadkaraoke/python-audio-separator/pull/316
- https://github.com/nomadkaraoke/python-audio-separator/commit/6d7d3fd60accda60c7ad8807ce5731022475dcf3
- https://github.com/nomadkaraoke/python-audio-separator/releases/tag/v0.47.2
- https://github.com/nomadkaraoke/python-audio-separator/blob/465b60f0271e880a7289a2c9aaeb13817c3b2bed/audio_separator/separator/uvr_lib_v5/demucs/states.py#L43
- https://github.com/nomadkaraoke/python-audio-separator/blob/465b60f0271e880a7289a2c9aaeb13817c3b2bed/audio_separator/separator/separator.py#L126
- https://www.vulncheck.com/advisories/audio-separator-before-0.47.2-rce-via-unsafe-pickle-deserialization-of-demucs-models
Do not conclude an asset is unaffected unless you checked it and the evidence covers both its version and the conditions above. If you could not check something, mark it unknown and say what is missing.
Report a table with one row per asset and these columns: Asset | Version or build | Status (affected, fixed, not applicable, unknown) | Compensating controls | Exposure (internet-facing, internal, none) | Evidence | Next action.
Entry: https://cve.now/CVE-2026-134172
For follow-up questions, use the cve.now MCP server at https://mcp.cve.now/mcp (no authentication), for example execute with name cve.get and arguments {"id": "CVE-2026-134172"}.Affected versions
cve.now Index
- Exploitation?/40Unknown. No CISA SSVC exploitation rating yet
- Exposure5/25Specific setups only
- Reach10/20Affects PyPI packages
- Severity8.5/10CVSS 8.5
- No patch0/5No signal
What would change the score
- Would score about 74 if added to CISA KEV
- Would score about 69 if CISA's SSVC rates exploitation active
- Would score about 34 if CISA's SSVC rates it automatable
Timeline
- CVE published
- First seen by cve.now
Evidence 8
Questions about CVE-2026-134172
Is CVE-2026-134172 being exploited?
It is not in CISA's Known Exploited Vulnerabilities catalog, and CISA's SSVC assessment does not mark exploitation as active. That does not mean it has not been attacked: vendors and researchers can report attacks before CISA lists them. No public exploit code is tracked.
Source: cve.now feed data
What does this mean?
Demucs separation loads .th model files in a way that can run code, so a malicious model file lets an attacker execute code on the machine when separation runs.
Source: vulncheck.com
Which versions are affected by CVE-2026-134172?
audio-separator (PyPI): before 0.47.2.
Source: cve.now feed data
Am I affected, and how do I check?
audio-separator versions before 0.47.2 are affected when Demucs separation runs. Check the installed audio-separator version, and whether models come from the shared /tmp/audio-separator-models/ default directory, which local users can write to, or from an unverified model download source.
Source: vulncheck.com
How do I fix CVE-2026-134172?
Upgrade audio-separator to 0.47.2 or later.
Source: vulncheck.com
Related
- CVE-2026-114824Vulnerability in the Oracle Platform Security for Java product of Oracle Fusion Middleware (component: Centralized Thirdparty Jars).
- CVE-2026-129039WordPress Newsletters plugin <= 4.18.2 - PHP Object Injection vulnerability
- CVE-2026-84236IBM Guardium Data Protection Deserialization of Untrusted Data
- CVE-2026-124990Masteriyo LMS <= 3.5.1 - Unauthenticated PHP Object Injection via 'file' Parameter
- CVE-2026-86130Dimension Improper Authentication in Log Collection Service Allows Unauthenticated Remote Code Execution
- CVE-2026-131914Deserialization of untrusted data in CodePlex AntiCSRF
No corrections so far. Something wrong on this page? Report an error.