CVE-2026-24188

Vulnerability

NVIDIA TensorRT, a deep learning inference optimizer, is affected by an out-of-bounds write vulnerability. This flaw exists in the processing of certain inputs (likely model files) and can be triggered by an attacker who supplies a maliciously crafted input. Affected versions are not detailed in the reference [1], but the vulnerability is present in TensorRT builds before a fix is applied.

Exploitation

An attacker would need to provide a specially crafted input to a vulnerable TensorRT deployment. The exact prerequisites (e.g., local access or remote delivery) are not disclosed in the available references [1]. The out-of-bounds write is triggered during inference or model loading when the software fails to validate input bounds.

Impact

Successful exploitation leads to data tampering, as stated in the official description [1]. An out-of-bounds write could corrupt adjacent memory, potentially altering inference results or causing denial of service. The attacker does not gain arbitrary code execution based on available information.

Mitigation

As of the publication date (2026-05-20), no fixed version has been announced by NVIDIA. Users should monitor NVIDIA's security advisories for upcoming patches. No workarounds are documented in the reference [1].