Can you explain this vulnerability to me?

CVE-2026-31249 is an insecure deserialization vulnerability in the CosyVoice make_parquet_list.py data processing tool. The tool loads PyTorch .pt files 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 vulnerability by providing malicious .pt files within a data directory. When the victim processes this directory using the tool, arbitrary code can be executed on the victim's system.

Useful 0 Not Useful 0

How can this vulnerability impact me? :

This vulnerability can lead to arbitrary code execution on the victim's system when processing malicious .pt files. This means an attacker could run any code they choose, potentially leading to system compromise, data theft, or further attacks.

Useful 0 Not Useful 0

What immediate steps should I take to mitigate this vulnerability?

To mitigate the insecure deserialization vulnerability in CosyVoice's make_parquet_list.py tool, ensure that when loading PyTorch .pt files using torch.load(), the parameter weights_only=True is enabled. This prevents arbitrary code execution by restricting deserialization to only tensor weights.

Additionally, avoid processing untrusted or malicious .pt files within data directories, as these can exploit the vulnerability by executing arbitrary code during deserialization.

Useful 0 Not Useful 0

How can this vulnerability be detected on my network or system? Can you suggest some commands?

This vulnerability involves insecure deserialization in the make_parquet_list.py tool of CosyVoice when loading .pt files using torch.load() without the weights_only=True parameter. Detection would focus on identifying usage of this script processing untrusted .pt files.

To detect potential exploitation or presence of malicious .pt files, you can:

• Check if the make_parquet_list.py script is being run on your system.
• Audit the directories containing .pt files for unexpected or untrusted files.
• Monitor execution logs or system calls related to torch.load() usage without the weights_only=True parameter.

Specific commands might include:

• Use file system commands to find .pt files, e.g., `find /path/to/data -name '*.pt'`
• Check running processes or recent command history for execution of make_parquet_list.py, e.g., `ps aux | grep make_parquet_list.py` or `grep make_parquet_list.py ~/.bash_history`
• If possible, add logging or debugging to the script to verify if torch.load() is called without weights_only=True.

Note: No explicit detection commands or tools are provided in the available resources.

Useful 0 Not Useful 0