Summary
CVE-2026-114493 is a server-side request forgery (SSRF) vulnerability (CWE-918) in LocalAI v4.6.0 through v4.11.0. The SSRF guard validates only the initial gallery config URL, so an attacker can bypass it with HTTP redirects. The CVE record rates it CVSS v4.0 9.2 (Critical).
Technical details
- Root cause:
core/gallery/gallery.govalidates only the initial gallery config URL and does not re-validate redirect targets. - Trigger: an attacker submits a config URL to
POST /models/applythat redirects to an internal destination. - Attack vector: network, with no privileges or user interaction required according to the CVSS vectors.
- Impact: the server can be made to fetch internal resources or cloud metadata endpoints, and the responses can be exposed to the attacker (high confidentiality impact).
- Credit: Zachariah Mithani (AploSec).
Affected software
- LocalAI v4.6.0 through v4.11.0
Severity
- CVSS v4.0: 9.2 (Critical),
CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:H/VI:N/VA:N/SC:H/SI:N/SA:N - CVSS v3.1: 8.6 (High),
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:C/C:H/I:N/A:N
Mitigation and recommended actions
- Immediate: the CVE record does not name a fixed release. A fix is proposed in LocalAI pull request #12791, which adds redirect-target validation. When we reviewed it, the PR was still open and not merged. Check the LocalAI releases for a version that includes the fix, and upgrade to it once available.
- Until a fix is available, restrict network access to the LocalAI API, especially
POST /models/apply, to trusted users and networks. - Restrict LocalAI’s outbound access to internal networks and cloud metadata endpoints.

