CVE-2026-24174: NVIDIA Triton Inference Server DoS Flaw

CVE-2026-24174 Overview

CVE-2026-24174 is a denial of service vulnerability affecting NVIDIA Triton Inference Server. The vulnerability allows an attacker to cause a server crash by sending a malformed request to the server. A successful exploit of this vulnerability could lead to denial of service, disrupting AI/ML inference workloads and potentially impacting production environments that rely on Triton Inference Server for real-time model serving.

Critical Impact

Unauthenticated remote attackers can crash NVIDIA Triton Inference Server instances by sending specially crafted malformed requests, causing denial of service to AI/ML inference operations.

Affected Products

• NVIDIA Triton Inference Server (specific versions to be confirmed via vendor advisory)

Discovery Timeline

• April 7, 2026 - CVE-2026-24174 published to NVD
• April 8, 2026 - Last updated in NVD database

Technical Details for CVE-2026-24174

Vulnerability Analysis

This vulnerability is classified under CWE-681 (Incorrect Conversion between Numeric Types), indicating that the root cause involves improper handling of numeric type conversions within the Triton Inference Server request processing pipeline. The vulnerability is remotely exploitable over the network without requiring authentication or user interaction, making it particularly concerning for internet-exposed Triton instances.

The denial of service condition is triggered when the server receives and attempts to process a malformed request that exploits the incorrect numeric conversion, leading to an unhandled exception or memory corruption that crashes the server process.

Root Cause

The vulnerability stems from CWE-681: Incorrect Conversion between Numeric Types. This class of weakness occurs when a product converts a numeric value from one type to another in a way that produces a different value than the original. In the context of Triton Inference Server, this likely occurs during request parsing or tensor dimension/size handling, where malformed numeric values in requests could trigger integer truncation, sign extension errors, or type mismatches that cause the server to crash.

Attack Vector

The attack vector is network-based, requiring no privileges or user interaction. An attacker can exploit this vulnerability by sending specially crafted malformed HTTP/gRPC requests to a Triton Inference Server endpoint. The malformed request likely contains numeric values that, when processed through the vulnerable type conversion code path, cause the server to crash.

The attack requires network access to the Triton Inference Server API endpoints (typically ports 8000 for HTTP, 8001 for gRPC, and 8002 for metrics). Organizations exposing Triton Inference Server to untrusted networks are at elevated risk.

Detection Methods for CVE-2026-24174

Indicators of Compromise

• Unexpected Triton Inference Server process crashes or restarts
• Unusual network traffic patterns targeting Triton API endpoints (ports 8000, 8001, 8002)
• Malformed inference requests in server access logs with abnormal numeric parameters
• Increased rate of connection attempts followed by immediate disconnections

Detection Strategies

• Monitor Triton Inference Server process stability and implement alerting for unexpected terminations
• Implement network intrusion detection rules to identify malformed requests targeting Triton endpoints
• Enable verbose request logging to capture and analyze potentially malicious request patterns
• Deploy application-level monitoring to detect anomalous request payloads with unusual numeric values

Monitoring Recommendations

• Set up automated health checks for Triton Inference Server availability and response times
• Configure log aggregation to centralize Triton logs for security analysis
• Implement rate limiting and request validation at the network edge or load balancer level
• Monitor for repeated crash-restart cycles that may indicate active exploitation attempts

How to Mitigate CVE-2026-24174

Immediate Actions Required

• Review the NVIDIA Support Advisory for patch availability and upgrade instructions
• Restrict network access to Triton Inference Server endpoints to trusted sources only
• Implement network segmentation to isolate Triton instances from untrusted networks
• Enable request validation and rate limiting at the load balancer or API gateway level

Patch Information

NVIDIA has published a security advisory addressing this vulnerability. Organizations should consult the NVIDIA Support Advisory for specific patch versions and upgrade instructions. Additional technical details are available at the NVD CVE-2026-24174 Details page.

Workarounds

• Deploy Triton Inference Server behind a reverse proxy or API gateway with request validation capabilities
• Implement IP allowlisting to restrict access to Triton endpoints to known, trusted clients
• Use Kubernetes network policies or firewall rules to limit inbound connections to Triton pods/containers
• Consider deploying a Web Application Firewall (WAF) to filter malformed requests before they reach Triton

# Example: Restrict access to Triton ports using iptables
# Allow only trusted IP ranges to access Triton HTTP API
iptables -A INPUT -p tcp --dport 8000 -s 10.0.0.0/8 -j ACCEPT
iptables -A INPUT -p tcp --dport 8000 -j DROP

# Allow only trusted IP ranges to access Triton gRPC API
iptables -A INPUT -p tcp --dport 8001 -s 10.0.0.0/8 -j ACCEPT
iptables -A INPUT -p tcp --dport 8001 -j DROP