CVE-2025-24357

Summary

CVE-2025-24357 affects the vLLM library, which is used for LLM inference and serving. The issue lies in the hf_model_weights_iterator function within weight_utils.py. This function utilizes torch.load to load model checkpoints obtained from huggingface. By default, the weights_only parameter is set to False, leading to the execution of arbitrary code during unpickling if the loaded data is maliciously crafted pickle files. This vulnerability has been rectified in version 0.7.0.

Prevent cyber attacks with Recorded Future by prioritizing and patching critical vulnerabilities being exploited by threat actors targeting your industry. Book your demo to learn more.