Open WebUI’s image edit feature performs an unbounded GET request to any URL specified by the user. The response is not returned to the requester, creating a blind server‑side request forgery. Attackers can thus probe internal addresses and ports, determining which are reachable based on success or failure. If a reachable service exposes state‑changing GET endpoints, further interaction may be possible. The vulnerability opens a path for local network reconnaissance and potential command or data exfiltration for the service that is hit.

The flaw affects the open‑webui platform in versions 0.7.2 and earlier. These builds are self‑hosted, AI‑powered applications that operate offline but include a web interface for image editing via prompts. The impact is limited to installations running those older releases.

With a CVSS score of 4.3, the technical severity is moderate. Exploitation requires only that an attacker has access to the Open WebUI instance; no elevated privileges or additional credentials are needed. The blind nature of the flaw means the attacker must observe timing or error responses to infer port status, but this can be automated. Since EPSS is not available and the issue is not listed in CISA’s KEV catalog, the exploitation likelihood is not quantified, yet the simple attack flow and lack of mitigation make it a reasonable concern for any exposed deployment.