The AI Agent API endpoint (POST /apps/:appId/agent) lacks CSRF protection. An attacker can craft a malicious page that, when visited by an authenticated dashboard user, submits requests to the agent endpoint using the victim's session.
The AI Agent API endpoint (POST /apps/:appId/agent) does not enforce authorization. Authenticated users scoped to specific apps can access any other app's agent endpoint by changing the app ID in the URL. Read-only users are given the full master key instead of the read-only master key and can supply write permissions in the request body to perform write and delete operations. Affected are only dashboards with agent configuration enabled.
The AI Agent API endpoint (POST /apps/:appId/agent) lacks authentication. Unauthenticated remote attackers can send requests to the endpoint and perform arbitrary database operations against any connected Parse Server using the master key.
The ConfigKeyCache uses the same cache key for both master key and read-only master key when resolving function-typed keys. Under specific timing conditions, a read-only user can receive the cached full master key, or a regular user can receive the cached read-only master key.