NVIDIA TensorRT‑LLM contains an insecure deserialization flaw in its inter‑process communication component. When an attacker can write data that is sent to a process running under the same user account, the system will deserialize that data without proper validation. A successful exploit can cause arbitrary code execution, leakage of memory contents, alteration of data, and possible denial of service by crashing or freezing the application. The weakness is classified as CWE‑502, Deserialization of Untrusted Data.

The vulnerability affects NVIDIA TensorRT‑LLM. No specific version numbers are supplied in the advisory, so all deployed instances of TensorRT‑LLM that rely on its IPC layer are potentially impacted.

The CVSS score of 7.8 indicates a high severity, but the EPSS score of <1% suggests that exploitation is currently unlikely. The flaw is not yet listed in the CISA KEV catalog, further implying limited widespread exploitation. Attack requires local same‑user access, meaning the threat surface is restricted to users who already have legitimate access to the host. If such access is compromised or misused, the attacker can trigger the deserialization path to gain remote control over the TensorRT‑LLM process or affect its availability.