The vulnerability arises because the LoRA-GA and CorDA initialization modules in the peft library invoke torch.load() without the weights_only=True parameter on cache or covariance files specified in configuration. Normally, peft uses a safe-loading wrapper to prevent unsafe deserialization, but this code path bypasses that protection. Since torch.load() without weights_only=True performs full pickle deserialization, an attacker who can supply a malicious cache or covariance file (for example, via a shared or downloaded LoRA-GA or CorDA cache) can execute arbitrary code during model initialization.

Successful exploitation allows an attacker to execute arbitrary code on the system running the vulnerable peft library by tricking it into loading a malicious cache or covariance file. This can lead to full compromise of the affected system, including confidentiality, integrity, and availability impacts as indicated by the CVSS vector (C:H/I:H/A:H).

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until an official fix is available, users should avoid loading untrusted or shared LoRA-GA or CorDA cache or covariance files. Monitoring for updates from Hugging Face regarding a safe-loading patch is recommended.