---
id: "CVE-2026-114493"
kind: "vulnerability"
aliases: ["CVE-2026-114493"]
title: "LocalAI: gallery config redirects let attackers reach internal and cloud metadata"
severity: "critical"
cvss: 9.2
epss: 0.0038
exploited: false
status: "open"
category: "ai-tooling"
products: [{"name": "LocalAI", "fixed": [], "affected": ">= 4.6.0, <= 4.11.0", "ecosystem": "github"}]
ecosystems: ["github"]
tags: ["ai-tooling", "cloud", "localai", "self-hosted", "ssrf"]
not_affected_if: ["no-network-exposure", "version-outside-4.6.0-4.11.0"]
published_at: "2026-10-21T18:17:28+00:00"
first_seen_at: "2026-10-23T21:57:29+00:00"
updated_at: "2026-10-24T13:15:54+00:00"
sources: [{"source": "nvd", "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-114493", "first_seen_at": "2026-10-23T21:57:29+00:00"}, {"source": "analyst", "url": null, "first_seen_at": "2026-10-23T22:11:17+00:00"}, {"source": "epss", "url": "https://api.first.org/data/v1/epss?cve=CVE-2026-114493", "first_seen_at": "2026-10-24T13:15:54+00:00"}]
---

# LocalAI: gallery config redirects let attackers reach internal and cloud metadata

LocalAI validates only the first gallery config URL, so a request to POST /models/apply can redirect the server to internal addresses or cloud metadata endpoints. An attacker can read those internal responses back as gallery content, which can expose cloud credentials.

_Summary, check and fix written by a model from the sources; labelled as such._

## Affected

- github:LocalAI affected >= 4.6.0, <= 4.11.0 fixed ?

## How to check

Check the LocalAI version the team runs: if it is between v4.6.0 and v4.11.0 inclusive, it is affected.

## Fix

Upgrade LocalAI to a version later than v4.11.0; the fixed version is unknown from the sources.
Until upgraded, block POST /models/apply from untrusted networks.

## Timeline

- 2026-10-23T21:57:29+00:00 nvd https://nvd.nist.gov/vuln/detail/CVE-2026-114493
- 2026-10-23T22:11:17+00:00 analyst — VulnCheck CNA advisory + PR 12791 read. Affected v4.6.0-v4.11.0. No fixed release yet: fix PR 12791 still open as of 2026-10-21 (maintainer bot signed off). Auth: unauthenticated in default Docker (no API key set), per PR 10673 for the earlier #10665 SSRF. No exploitation or PoC found; x/social not searched.
- 2026-10-24T13:15:54+00:00 epss https://api.first.org/data/v1/epss?cve=CVE-2026-114493

## Documents

- [advisory] VulnCheck: LocalAI SSRF guard bypass via HTTP redirects https://www.vulncheck.com/advisories/localai-4.6.0-through-4.11.0-ssrf-guard-bypass-via-http-redirects
- [describes] PR 10673: initial SSRF fix for /models/apply (#10665); default Docker has no API key https://github.com/mudler/LocalAI/pull/10673
- [fix] PR 12791: re-validate external URL policy on every redirect hop (open) https://github.com/mudler/LocalAI/pull/12791

## From the sources

LocalAI v4.6.0 through v4.11.0 contains a server-side request forgery vulnerability in core/gallery/gallery.go that validates only the initial gallery config URL. Attackers can submit a config URL through POST /models/apply that redirects to internal or cloud metadata addresses, receiving those internal responses as gallery content.

---

_Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai_