Mitigation Strategies

To mitigate this vulnerability, avoid starting the CosyVoice gRPC server with a user-specified directory containing untrusted or potentially malicious model files.

Ensure that the torch.load() function is used with the weights_only=True parameter enabled to prevent insecure deserialization of arbitrary Python objects.

Do not allow attackers to provide or control the model files loaded by the server.

Useful 0 Not Useful 0

Executive Summary

This vulnerability exists in CosyVoice's gRPC server component where it insecurely deserializes data. Specifically, when the server starts, it loads a speech synthesis model from a user-specified directory using the torch.load() function without enabling the weights_only=True security parameter. This insecure deserialization allows an attacker to craft malicious model files that, when loaded by the server, execute arbitrary Python code on the victim's system during server initialization.

Useful 0 Not Useful 0

Impact Analysis

The impact of this vulnerability is that an attacker can execute arbitrary code on the system running the vulnerable CosyVoice gRPC server. This can lead to full system compromise, unauthorized access, data theft, or disruption of services, depending on the attacker's intent and the privileges of the server process.

Useful 0 Not Useful 0