Ghost CMS CVE-2026-70591

Exploitation strictly requires an authenticated Ghost Admin session with at minimum the staff role - this is not exploitable by anonymous visitors, subscribers, or contributors without Admin panel access. The vulnerable code path is triggered only by Ghost Admin's image dimension fetching feature for image formats that require a full HTTP buffer response (as opposed to a probe-only fetch), meaning the attacker must cause Ghost to process an externally-sourced image URL through this specific code path. The SSRF is completely blind: no HTTP response body, status code, or headers are returned to the triggering user, restricting the attack to port-state inference via connection timing or TCP error behavior. Limiting factors are: (1) PR:H - valid staff credentials are mandatory; (2) blind-only output - no data exfiltration is possible, only port probing; (3) staff users already hold significant Ghost Admin trust, meaning the lateral-movement value of internal port scanning is the primary concern rather than privilege escalation within Ghost itself.

The vendor-assigned CVSS 3.1 score of 4.1 (Medium) with vector AV:N/AC:L/PR:H/UI:N/S:C/C:L/I:N/A:N is well-calibrated to the actual impact. PR:H is the critical limiting factor: exploitation is confined to users who already hold Ghost Admin staff credentials - a trusted internal role with content-editing rights, not an anonymous or low-trust actor. The Scope:Changed metric correctly captures that the server-side request can reach internal hosts outside the Ghost application boundary, but the SSRF is entirely blind (no response body or headers are returned to the triggering user), limiting utility to TCP port-state inference rather than data exfiltration. No EPSS score was included in the provided intelligence data, and no CISA KEV listing or public POC was identified. Real-world priority is low for organizations that limit Ghost staff accounts to trusted employees; however, self-hosted Docker deployments on cloud infrastructure should treat this with elevated urgency if the Ghost server can reach cloud metadata endpoints such as 169.254.169.254, where even a blind SSRF can confirm metadata service availability.

A malicious or compromised Ghost staff user opens the Ghost Admin post editor and inserts an image URL pointing to an internal-only host and non-standard port (e.g., http://10.0.0.5:2379/probe.jpg targeting an etcd endpoint). Ghost's server-side image-size module initiates a blind HTTP GET to that internal address; while no content is returned to the user, TCP connection timing and error codes allow the attacker to infer whether the port is open, enabling systematic mapping of the internal network from outside the perimeter. No public POC was identified at time of analysis.