CVE-2026-133875

one-hub vision image URLs could reach internal services

MartialBE · one-hub

First documented disclosure · Review checkpoint

Summary

Ordinary API-token holders could make the model gateway fetch internal URLs.

Severity and exploitation evidence

Medium · 5.3/10 · CVSS 4.0. Severity source.

Exploitation not established: Public evidence describes a weakness, source correction or controlled demonstration; malicious exploitation against users is not established.

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

Image preprocessing accepts HTTP destinations without a private-address or redirect boundary.

Conditions required

A standard non-admin token and a Claude, OpenRouter or Ollama vision request.

Potential impact

Internal reachability; responses are forwarded to the model provider. Disclosure back to the caller depends on usable image content and model output.

Affected versions and configurations

Through 0.14.27.

Fix and mitigation status

No merged or released correction verified; PR940 remains open.

Practical next steps

Restrict image-fetch egress and untrusted API access; monitor and review the proposed correction before adopting it.

Operating lesson

Validate destinations at connection time and on every redirect.

Evidence and limits

Original issue939 was public November 6. PR940 proposes address validation and guarded dialing, but is still open; its CF-worker path intentionally retains its existing transport. CNA publication 2026-11-20 is distinct from the earliest reviewed public source. VulnCheck assessment: CVSS 4.0 5.3; CVSS 3.1 5. Proposed correction: https://github.com/MartialBE/one-hub/pull/940 .

Sources