What is the CVSS score of EUVD-2026-61108?

EUVD-2026-61108 has a CVSS 3.1 base score of 9.8 (Critical). CVSS vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H. EPSS exploitation probability: 0.4%.

Which versions of Triton Inference Server are affected by EUVD-2026-61108?

NVIDIA Triton Inference Server, critical infrastructure for ML model serving, contains a path traversal vulnerability (CVSS 9.8) allowing remote attackers to access or modify data outside intended…

How to fix or mitigate EUVD-2026-61108?

Within 24 hours: identify all systems running NVIDIA Triton Inference Server and audit network exposure; immediately isolate instances to internal-only access or behind authentication/network controls. Within 7 days: implement network segmentation restricting Triton endpoint access to authenticated internal connections; apply firewall rules blocking untrusted external access; update all deployment configurations to enforce these restrictions. Within 30 days: monitor NVIDIA security advisories for patch release and establish a phased patching plan for production and non-production environments; if vendor patches remain unavailable beyond 60 days, evaluate alternative ML serving platforms with stronger security posture.

Triton Inference Server EUVD-2026-61108

| CVE-2026-47627 CRITICAL

Path Traversal (CWE-22)

2026-08-18 nvidia

GHSA-79wf-fcj9-7grf

Denial Of Service Path Traversal Triton Inference Server Nvidia

9.8

CVSS 3.1 · Vendor: nvidia

Severity by source

Vendor (nvidia) PRIMARY

9.8 CRITICAL

AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H

vuln.today AI

9.1 CRITICAL

Network-reachable unauthenticated endpoint (AV:N/PR:N/AC:L); path traversal can read files (C:H) and cause DoS (A:H), but no integrity write is described (I:N).

3.1 AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:H

4.0 AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:N/VA:H/SC:N/SI:N/SA:N

Primary rating from Vendor (nvidia).

CVSS VectorVendor: nvidia

Attack Vector

Network

Attack Complexity

Low

Privileges Required

None

User Interaction

None

Scope

Unchanged

Confidentiality

High

Integrity

High

Availability

High

Lifecycle Timeline

Analysis Generated

Aug 18, 2026 - 18:52 vuln.today

CVE Published

Aug 18, 2026 - 18:23 cve.org

CRITICAL 9.8

DescriptionCVE.org

NVIDIA Triton Inference Server for Linux contains a vulnerability where an attacker could cause path traversal. A successful exploit might lead to denial of service.

Articles & Coverage 2

News 6 New Nvidia Vulnerabilities - 1 Critical, 5 High Severity Aug 19, 2026 News Critical Path Traversal in NVIDIA Triton Inference Server - CVE-2026-47627 Aug 18, 2026

AnalysisAI

Path traversal in NVIDIA Triton Inference Server for Linux (CWE-22) lets remote attackers manipulate file paths to reach locations outside intended directories, with the vendor citing denial of service as the primary outcome. NVIDIA assigned a critical 9.8 CVSS score claiming full confidentiality, integrity, and availability impact, though the published description only asserts DoS. There is no public exploit identified at time of analysis and it is not listed in CISA KEV.

Technical ContextAI

Triton Inference Server is NVIDIA's open-source, high-performance model-serving platform used to deploy AI/ML models over HTTP and gRPC, backed by a model repository on the local filesystem. The flaw is classified as CWE-22 (Improper Limitation of a Pathname to a Restricted Directory / path traversal), meaning user-controllable input containing sequences such as '../' is used to construct filesystem paths without adequate normalization or containment. In a model-serving context this class of bug typically arises where the server resolves model names, artifact paths, or repository locations supplied via the control/inference APIs, allowing references to escape the intended model repository root. The affected CPE is cpe:2.3:a:nvidia:triton_inference_server with all versions wildcarded, so no fixed lower/upper bound is encoded in the CPE data.

RemediationAI

Upgrade to the fixed Triton Inference Server release identified in NVIDIA's security bulletin at https://github.com/NVIDIA/product-security/tree/main/2026/5865 - no exact fixed version number is present in the supplied data, so retrieve the patched build from that advisory (patch guidance available per vendor advisory; specific version not independently confirmed here). Until patched, reduce exposure by ensuring Triton's HTTP (default 8000) and gRPC (default 8001) endpoints are not reachable from untrusted networks - bind to localhost or place them behind an authenticating reverse proxy or network ACLs, accepting the trade-off that this breaks any legitimate remote inference clients. Additionally restrict or disable dynamic model loading (run with --model-control-mode=none rather than explicit/poll) so model-repository paths cannot be influenced at runtime, at the cost of losing hot model updates, and run the server as a low-privilege user in a container with a read-only, tightly scoped model-repository mount to limit what a traversal can reach.