AI Puffer CVE-2026-104742

Exploitation requires two concrete conditions drawn from the description: (1) the attacker must hold an authenticated subscriber-level or higher account (PR:L), and (2) an administrator must have previously granted the Knowledge Base ('sources') module to the attacker's role via the plugin's Role Manager - a non-default grant not available to lower-privileged users out of the box. These are the limiting factors that sharply reduce real-world exploitability: without that explicit role grant the attack surface is not reachable, and impact is confined to modifying global semantic search settings in aipkit_options (integrity only). There is no required user interaction and no unauthenticated path.

This is a genuinely low-priority issue despite being network-reachable. The provided CVSS 3.1 vector (AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:L/A:N, base 3.1 Low) tells a consistent story: although the attack is remote, it carries high attack complexity, requires an authenticated low-privileged account (PR:L), and yields only a limited integrity impact with no confidentiality or availability loss. The AC:H rating directly reflects the key precondition - an administrator must have previously granted the Knowledge Base ('sources') module to the attacker's role via the Role Manager, which is not the default. No EPSS score, KEV listing, POC, or SSVC decision data was provided in the input, so there is no signal of active or likely exploitation; per the input this is 'no public exploit identified at time of analysis.' The signals do not conflict: a low CVSS, a non-default configuration dependency, and the absence of exploitation indicators all point the same direction. The only mislabel to note is the 'Authentication Bypass' tag - this is actually a missing-authorization (CWE-862) flaw against an already-authenticated user, not an authentication bypass. Net: low real-world risk, worth patching in routine maintenance rather than emergency response.

On a site where an administrator has used the Role Manager to grant the Knowledge Base ('sources') module to a low-privileged role, an authenticated subscriber-level user sends a crafted AJAX request to the plugin's settings handler and alters global, site-wide semantic search settings (vector/embedding provider, embedding model, target ID, result count, no-results text) that should be admin-only. No public exploit code is identified in the provided data, and the AC:H rating reflects that this only works given that non-default role grant.