A high-severity integer overflow vulnerability in the NVIDIA Triton Inference Server allows unauthenticated remote attackers to trigger a denial of service condition.

This vulnerability involves an integer overflow condition (CWE-680) that can lead to a buffer overflow. The attack vector is network-based and requires no authentication, allowing an unauthenticated remote attacker to disrupt service availability.

The primary risk posed by this vulnerability is the loss of service availability for AI and machine learning inference workloads. With a CVSS score of 7.5, the vulnerability is classified as High because it is easily exploitable over the network without requiring user interaction or prior authentication. Organizations relying on Triton for production inference pipelines could suffer significant operational downtime if an attacker successfully triggers the overflow.

Immediate Action: Upgrade the NVIDIA Triton Inference Server to version 25.05 or later as specified in the official vendor security advisory.

Proactive Monitoring: Monitor server logs for unusual crash events or unexpected process termination patterns that may indicate an attempted exploitation of the inference engine.

Compensating Controls: Deploy a Web Application Firewall (WAF) or an API gateway with strict input validation rules to inspect incoming requests for malformed or excessively large data structures that could trigger an integer overflow.

Public Exploit Available: No (exploit_available: false)

Given the ease of exploitability and the potential for service disruption, administrators should prioritize updating the Triton Inference Server across all affected environments. Ensuring that the software is patched to version 25.05 or higher is the only definitive way to eliminate the risk of an integer overflow leading to a denial of service.