The vulnerability stems from an insecure unpickling routine used during model weight deserialization in NVIDIA TensorRT‑LLM for Linux. The routine accepts untrusted data, allowing a local, unauthenticated attacker to cause arbitrary code execution. Successful exploitation could result in privilege escalation, data tampering, and information disclosure, compromising the confidentiality of the host system.

NVIDIA TensorRT‑LLM for Linux is affected. All installations of this product are potentially vulnerable until a patch is applied or mitigations are in place.

The CVSS score of 8.4 marks this flaw as high severity, but the EPSS score of less than 1% indicates a very low probability of exploitation. The vulnerability is not listed in the CISA KEV catalog. Attack attempts would require local, unauthenticated access, suggesting that any user or process that can load user‑supplied model weights is a potential vector for exploitation.