The AI editor loads models and prompts through ia.php. Filters such as ia.models.available and ia.prompts.load.where can narrow what a user sees. This does not itself authorize writes: prompt editing and deletion have their own queries and must be assessed separately in multiuser projects.
The core security review recorded eleven findings still proposed for remediation (req-181). They concern administrative list queries and sorting, state changes over GET, column name validation, escaping in shared functions, second-factor limits, uploaded browser-active file types, session binding and lifetime, email verification in social login, SMTP log secrecy, prompt ownership, and replay on the distributed-module channel. A proposed fix is not an implemented fix.
The AI, interface, and security library references describe concrete functions. Remediation tracking belongs in the security request, rather than this public page.