EUVD-2026-44486
GHSA-f54h-33c8-78p4
Severity by source
Sources disagree (Medium–Critical)AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
Deserialization of an attacker-supplied engine file requires the victim to load untrusted input, so AV:L and UI:R; code execution yields full C:H/I:H/A:H.
vuln.today treats the vendor’s rating as authoritative. A higher third-party CVSS (e.g. CISA-ADP) is shown for transparency but does not drive the headline severity.
CVSS VectorNVD
Lifecycle Timeline
4DescriptionNVD
NVIDIA TensorRT for contains a vulnerability where a user might cause a deserialization of untrusted data. A successful exploit of this vulnerability might lead to code execution.
AnalysisAI
Code execution in NVIDIA TensorRT (all versions through v10.16.1) arises from unsafe deserialization of untrusted data (CWE-502), letting an attacker who supplies a malicious serialized artifact run arbitrary code in the context of the inference process. It affects the TensorRT SDK/runtime used to optimize and execute deep-learning models. The vendor-assigned CVSS is 9.8, but there is no public exploit identified at time of analysis, EPSS is low (0.48%, 38th percentile), and CISA SSVC lists exploitation as 'none'.
Technical ContextAI
TensorRT is NVIDIA's high-performance deep-learning inference SDK that compiles trained models into optimized 'engine' plan files, which are then serialized to disk and deserialized at runtime for execution on GPUs. The root cause is CWE-502 (Deserialization of Untrusted Data): when the runtime deserializes an engine plan or related artifact that originates from an untrusted source, crafted contents can drive memory corruption or object reconstruction that culminates in code execution. The single CPE, cpe:2.3:a:nvidia:tensorrt:*:*:*:*:*:*:*:*, is a wildcard covering the TensorRT application across versions, indicating the flaw is in the core library rather than a specific packaged edition.
RemediationAI
Apply the fix in NVIDIA's advisory (https://nvidia.custhelp.com/app/answers/detail/a_id/5855) - Patch available per vendor advisory; since affected versions run through v10.16.1, upgrade to the fixed TensorRT release identified in that bulletin (exact fixed version not stated in the provided data, so confirm it directly from the advisory rather than assuming). As a compensating control until patched, treat serialized engine plan files as untrusted code and only deserialize engines your own pipeline built from trusted sources; do not load engine/plan files received from users, third parties, or shared model repositories. Enforce integrity checks (cryptographic signing or checksums) on engine artifacts before deserialization, restrict filesystem/registry paths from which engines are loaded, and run inference processes with least privilege and sandboxing/containment so a successful exploit cannot reach the host - the trade-off is added build/deployment friction and rejection of externally-supplied precompiled engines.
Local code execution in NVIDIA TensorRT is possible when the library parses an attacker-supplied input (such as a crafte
Code execution in NVIDIA TensorRT is possible when the SDK processes a maliciously crafted input that overflows a heap-b
Improper array index validation (CWE-129) in NVIDIA TensorRT allows an attacker to trigger out-of-bounds memory access t
Out-of-bounds write in NVIDIA TensorRT prior to version 10.16.1 allows an attacker to corrupt memory and tamper with dat
Same weakness CWE-502 – Deserialization of Untrusted Data
View allShare
External POC / Exploit Code
Leaving vuln.today