Ghost CMS CVE-2026-105648

Requires that Ghost's outbound-request functionality (notably Webmentions) be enabled and reachable such that an unauthenticated actor can cause the server to fetch an attacker-chosen URL, AND that the attacker can express an internal/private destination in a form that slipped past the pre-6.65.0 IP validation - specifically alternate IPv4 encodings (decimal/octal/hex/integer), IPv4-mapped or IPv4-compatible IPv6 addresses, or 6to4/NAT64 (RFC 6052) transition addresses. Exploitation is further gated on 'some network configurations' (per vendor) - the Ghost server must actually have reachable internal hosts (e.g. DNS64/NAT64 deployments or a flat internal network). Key LIMITING factors: the attack is blind (no response data is ever returned to the attacker, per the description), attack complexity is High, and value is confined to request-only side effects and liveness inference rather than data theft.

This is a genuine but modest-severity server-side request forgery (SSRF) weakness, not a top-tier priority. The provided CVSS:3.1 vector (AV:N/AC:H/PR:N/UI:N/S:C/C:L/I:N/A:N = 4.0 MEDIUM) is internally consistent with the description: network-reachable and unauthenticated (PR:N), but high complexity (AC:H) because success depends on specific network configurations, and the scope change (S:C) reflects reaching a different security authority (the server's internal network) with only limited confidentiality impact (C:L) and no integrity or availability impact. Critically, the description states a successful attack returns NO response data, making this a 'blind' SSRF - the attacker can cause requests but cannot read responses, which sharply limits data-exfiltration value and aligns with the low CVSS. No EPSS score, no CISA KEV listing, and no public POC were provided in the input, so there is no public exploit identified at time of analysis and no evidence of active exploitation. The source-code evidence (the request-external.js IP-validation rewrite using ipaddr.js to handle IPv4-mapped, NAT64/RFC6052, 6to4, and alternate-encoding bypasses) confirms the root cause is incomplete deny-listing of private/reserved IP ranges (consistent with CWE-184). Net assessment: a real defense-in-depth fix worth patching on normal cadence, but the blind nature, high attack complexity, and absence of exploitation signals make it a low-to-medium operational priority rather than an emergency.

An unauthenticated attacker publishes a page containing a Webmention (or similar outbound-request trigger) pointing a target URL at an internal address expressed in a form that bypassed Ghost's private-IP filter - for example an alternate IPv4 encoding, an IPv4-mapped IPv6 address, or a NAT64/DNS64-synthesized address on a network so configured. Ghost's server then issues an HTTP request to that internal host, letting the attacker probe or interact with services behind the firewall. No public exploit code was identified, and because no response data is returned the attacker gains only blind request capability (e.g. port/host liveness inference or triggering side-effecting internal endpoints).