Deserialization of Untrusted DataWeakness · CWE-502

CVE-2026-31250

HIGH · 7.3 CVSS v3.1 Published 2026-05-11
Mitigation only
No fix yet — a mitigation exists. There is no fixed release. A documented workaround reduces exposure in the meantime.
See remediation →
82/100
Remediation priority · High
Remotely reachable No privileges Zero-click

Official description Straight from the sourceThe vendor's or NVD's own wording, published unedited. Authoritative, but often terse — it says what broke, rarely what to do.

NVD · unedited
CosyVoice thru commit 6e01309e01bc93bbeb83bdd996b1182a81aaf11e (2025-30-21) contains an insecure deserialization vulnerability (CWE-502) in its average_model.py model averaging tool. The script loads PyTorch checkpoint files (epoch_*.pt) for model averaging using torch.load() without enabling the weights_only=True security parameter. This allows the deserialization of arbitrary Python objects via the pickle module. An attacker can exploit this by providing malicious checkpoint files within a directory. When a victim uses the tool to average models from this directory, arbitrary code is executed on the victim's system.

Technical summary Written by usOur analysis, written from the advisory, the CVSS vector and the affected-version data. It adds context the advisory leaves out, and never invents facts that are not in the source.

dbcve analysis · high confidence

CosyVoice's average_model.py uses torch.load() without the weights_only=True parameter when loading PyTorch checkpoint files (epoch_*.pt). This enables pickle deserialization of arbitrary Python objects, allowing remote code execution when a victim processes a directory containing malicious checkpoint files.

MitigationAdd weights_only=True to all torch.load() calls in average_model.py. If the model averaging requires loading non-tensor objects, implement a custom unpickler with an allowlist of safe types or restructure the checkpoint format to use safer serialization methods.

Verify against the referenced sources before acting — the references below are authoritative for this CVE, this summary is not.

CVSS breakdown How the score is builtThe industry scoring standard. It rates how the flaw is reached, what it takes to exploit, and what an attacker gains — the score is derived from those, not the other way round.

From the vector
Attack vector
Network
Complexity
Low
Privileges
None
User interaction
None
Scope
Unchanged
Confidentiality
Low
Integrity
Low
Availability
Low

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

Am I affected? How to checkSteps we derive from the advisory and the affected-version data, so you can decide whether this CVE reaches your setup. They are a guide, not a scan — your own configuration is the authority.

dbcve checks

Work through these to decide whether this CVE applies to you.

  1. Locate the CosyVoice installation
    Find the average_model.py file in the CosyVoice codebase - typically in a 'cosyvoice' directory or source folder. Search for files named average_model.py
    Affected if Cannot locate average_model.py means CosyVoice may not be installed or is in a non-standard location
  2. Inspect torch.load() calls in average_model.py
    Open average_model.py and search for 'torch.load(' pattern. Check if the weights_only parameter is set to True
    Affected if The code contains torch.load() calls loading epoch_*.pt files without weights_only=True parameter - this indicates the vulnerability is present
  3. Identify checkpoint file loading patterns
    Search for code that loads files matching 'epoch_*.pt' pattern using torch.load(). Look for functions that load model checkpoints for averaging
    Affected if Checkpoint files (epoch_*.pt) are loaded via unprotected torch.load() calls enabling arbitrary deserialization
  4. Check if model averaging feature is enabled
    Determine if the CosyVoice application uses the average_model.py functionality - check if model averaging is invoked during training or inference workflows
    Affected if The model averaging feature processes checkpoint files and is actively used in the environment
  5. Verify Python and PyTorch versions
    Check installed PyTorch version with 'python -c "import torch; print(torch.__version__)"'. Compare against known affected versions (older PyTorch releases had fewer safeguards)
    Affected if Running with older PyTorch versions may have additional deserialization risks, though the code vulnerability exists regardless of PyTorch version

You are affected if CosyVoice's average_model.py loads epoch_*.pt checkpoint files using torch.load() without the weights_only=True parameter, and your workflow processes these checkpoint files.

Generated from the published advisory. Verify against your own configuration.

Check your environment

Paste your version and any relevant configuration and it will be compared against the affected criteria above. Do not include secrets or credentials.

AI-assisted, checked against the advisory. Informational, not a guarantee.

Remediation Closing itWhat it takes to close this. Where a vendor fix exists we point at it; where none exists we say so plainly, and can build one. Effort estimates are scoped from the advisory, not from your codebase.

dbcve · scoped
Mitigation available No clean upgrade yet — mitigate in the meantime
Mitigation

Add weights_only=True to all torch.load() calls in average_model.py. If the model averaging requires loading non-tensor objects, implement a custom unpickler with an allowlist of safe types or restructure the checkpoint format to use safer serialization methods.

Recommended fix High confidence
  1. Locate the average_model.py file in the CosyVoice repository
  2. Open average_model.py and find all instances of torch.load() used to load checkpoint files (epoch_*.pt)
  3. For each torch.load() call, add the weights_only=True parameter (e.g., change torch.load(path) to torch.load(path, weights_only=True))
  4. If the model requires loading custom classes/objects, refactor to use a safer approach such as defining a custom weights_only=True compliant loader or using state_dict for model weights instead of full checkpoint objects
  5. After applying the patch, test the average_model.py script with legitimate checkpoint files to ensure functionality is preserved
  6. Commit the changes and redeploy
Caveat If the legitimate use case requires loading custom Python objects (not just tensors), the weights_only=True parameter may break existing workflows; such code would need refactoring to use state_dict or register custom classes safely

Generated from the published advisory — verify against the referenced sources before acting.

Have this fixed Scoped from the published advisory
  • Consultation2.0 h
  • Implementation1.0 h
  • Testing2.0 h
  • Review / QA1.0 h
6.0 hours of engineering $1,060
Get help mitigating

An estimate, not a bill — we confirm scope with you before any work starts. Need it this week? Rush from $1,696.

Scan for this in your stack

Free · runs locally
dbcve dependency scanner

Check whether your project pulls in CVE-2026-31250 — or any other known-vulnerable package — straight from your lock files. Free and open source; it runs locally and uploads nothing.

References Go to the primary sourcePrimary sources — vendor advisories, patches and trackers. Where our summary and a reference disagree, the reference wins.

Primary sources

Practitioner notes

Contributed

Peer-ranked notes from engineers who’ve handled CVE-2026-31250 in production — separate from our analysis above.

No notes yet

Be the first to add a field note for this CVE — a mitigation you’ve verified, a version caveat, or a link to a working fix. Sign in above to contribute.

What this is

A place for practitioners to share what actually worked: a mitigation you’ve tested, a configuration change, a version- or environment-specific caveat, or a link to a verified patch. The most useful notes rise to the top as peers upvote them, so the signal stays high.

What belongs here
  • Verified mitigations, workarounds, and config changes
  • Version or environment caveats, and links to real fixes
  • No weaponised exploit code, or anything meant to cause harm
  • No spam, self-promotion, credentials, or personal data