The vulnerability in Jovancoding Network-AI (CVE-2026-46701) arises because the MCP SSE server uses an empty secret by default if the environment variable NETWORK_AI_MCP_SECRET is not set. This causes the authorization function to unconditionally return true, effectively disabling authorization. Additionally, the server sets Access-Control-Allow-Origin to '*', allowing cross-origin browser requests to read responses without restriction. An attacker can exploit this by tricking a user into visiting a malicious webpage, which can then invoke any of the 22 exposed MCP tools, including sensitive operations like config_set, agent_spawn, and blackboard_write, on a localhost server with default configuration. The vulnerability is fixed in version 5.4.5.

An unauthenticated attacker can perform unauthorized actions on a default-configured localhost instance of Network-AI by exploiting the lack of authorization and permissive CORS policy. This can lead to unauthorized configuration changes, spawning of agents, and modification of internal blackboard data, impacting confidentiality, integrity, and availability at varying levels.

Upgrade Network-AI to version 5.4.5 or later, where the issue is patched. Ensure that the environment variable NETWORK_AI_MCP_SECRET is set to a non-empty secret to enforce proper authorization. No other mitigation is indicated by the vendor advisory.