CVE-2026-68905

Public requests could spend a Hugo Blog operator’s AI provider quota

lanshi17 · hugo_blog AI assistant and search proxy

First documented disclosure · Review checkpoint

Summary

An unauthenticated proxy forwarded assistant, embedding and reranking requests under the operator’s provider credentials. Remote users could consume paid quota and interrupt service.

Severity and exploitation evidence

High · 8.2/10 · CVSS 3.1. Severity source.

Exploitation not established: The reporter demonstrated the proxy behavior. The reviewed sources identify no unauthorized production spending.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

An unauthenticated proxy forwarded assistant, embedding and reranking requests under the operator’s provider credentials.

Conditions required

The project was deployed with its documented reverse-proxy setup and a server-side AI provider key.

Potential impact

Remote users could consume paid quota and interrupt service.

Affected versions and configurations

The advisory verifies commit 744ca15. The CNA includes that revision and earlier.

Fix and mitigation status

The maintainer lists no patched version.

Practical next steps

Restrict access to the AI proxy, authenticate requests and apply per-user and provider spending limits.

Operating lesson

Limit who can spend a server-side credential and how much each caller can consume.

Evidence and limits

The sources concern the lanshi17/hugo_blog project. The CVE record was published on 2026-11-13.

Sources