CVE-2025-66448 vLLM vulnerable to remote code execution via transformers_utils/get_config

vLLM is an inference and serving engine for large language models LLMs. Prior to 0.11.1, vllm has a critical remote code execution vector in a config class named NemotronNanoVLConfig. When vllm loads a model config that contains an automap entry, the config class resolves that mapping with...

Insecure Deserialization

huggingface transformers is vulnerable to Insecure Deserialization. The vulnerability is due to the ability to load arbitrary pickle files from other repos specified by the indexpath while parsing the remote config.json fille. An attacker can exploit this flaw to execute arbitrary code on the...