Skip to content
Threat Feed
high advisory

NVIDIA TensorRT-LLM: Remote code execution via tokenizer loading vulnerability (CVE-2026-65119)

NVIDIA TensorRT-LLM contains an improper tokenizer loading vulnerability that could allow an attacker to achieve remote code execution, information disclosure, data tampering, or denial of service by influencing the control sphere.

CVE search metadata

CVE search record: CVE-2026-65119. Severity: high. CVSS: 8.4. KEV: no. Product: TensorRT-LLM. Brief: NVIDIA TensorRT-LLM: Remote code execution via tokenizer loading vulnerability (CVE-2026-65119). Brief link: https://feed.craftedsignal.io/briefs/2026-10-nvidia-tensorrt-llm-rce/

CVE search record: CVE-2026-24269. Severity: high. CVSS: 7.1. KEV: no. Product: TensorRT-LLM. Brief: NVIDIA TensorRT-LLM: Remote code execution via tokenizer loading vulnerability (CVE-2026-65119). Brief link: https://feed.craftedsignal.io/briefs/2026-10-nvidia-tensorrt-llm-rce/

Triage facts

Revision 2
Applicability
TensorRT-LLM
Evidence
Unknown
Severity
High
Action
Elevated
Sources
2 public sources
Freshness
Revision 2Oct 27, 15:52 UTC
Vulnerabilities
2 CVEs
Source quality
A1Likely
Revision history 1 entry · Revision 2
  1. Revision 2

    added CVE-2026-24269

    Changed: cpes, cves, lastmod, references, updates

    Sources: 1

NVIDIA TensorRT-LLM is vulnerable to a flaw in the KvCacheAwareRouter tokenizer loading process. An attacker capable of influencing the input to this component can cause the application to include functionality from an untrusted control sphere. This vulnerability potentially allows for remote code execution, unauthorized information disclosure, data tampering, and denial of service. Specific patch details or firmware versions addressing this issue are not disclosed in the current NVD advisory.

Impact

Successful exploitation may lead to remote code execution (RCE) on the host environment, unauthorized access to sensitive information, modification of application data, and service interruption.

Recommendation

  1. Inventory all instances of TensorRT-LLM within the environment to assess exposure.
  2. Monitor NVIDIA security bulletins for the release of patches addressing CVE-2026-65119.
  3. Restrict access to tokenizer configuration files and model loading pipelines to trusted, authenticated users only.
  4. Isolate systems running TensorRT-LLM from external or untrusted networks until a security update is applied.

Immediate actions

Inventory TensorRT-LLM deployments.

IT Operations 48h

Threat Hunt

Identify unauthorized access attempts to tokenizer loading components or model configuration paths.

T1190 medium medium confidence hunt now

Data: Application logs, File integrity monitoring (FIM) for configuration directories

Enrichment needed

  • Patch release information for CVE-2026-65119 (CTI) No patch version information available in the source.

Mitigations

Restrict access to tokenizer configuration and model loading paths.

immediate IT Operations

CVE-2026-65119