audio-separator versions before 0.47.2 have an insecure deserialization vulnerability in the handling of Demucs .th model files. Attackers able to supply malicious model files loaded via torch.load with weights_only=False can execute arbitrary code. This affects local users who can write to the shared /tmp/audio-separator-models/ directory or attackers controlling unverified model download sources. The vulnerability is fixed in version 0.47.2.