Executive brief
Network-AI is a tool used to coordinate multiple AI agents. A security flaw in its server component allows anyone with network access to the server to take full control of the AI orchestrator without a password. An attacker could use this to change configurations, spawn unauthorized AI agents, or modify shared data, potentially leading to unauthorized operations or data tampering.
Technical details
The Network-AI MCP SSE server suffers from missing authentication (CWE-306) because the `_isAuthorized()` method in `lib/mcp-transport-sse.ts` returns `true` by default if no secret is configured. While a previous fix (CVE-2026-46701) restricted CORS to localhost, the server still defaults to an empty secret and permits unauthenticated requests from non-browser clients (such as curl or SSRF) or when bound to non-loopback interfaces (0.0.0.0). An attacker can invoke all 22 available MCP tools, including `config_set`, `agent_spawn`, and `blackboard_write`. The vulnerability is fully resolved in version 5.7.2, which implements a fail-closed authorization check and mandates a non-empty secret for the SSE server to start.
Affected products
- Jovancoding Network-AI (npm network-ai) <= 5.7.1
Timeline
- 2026-05-16: advisory: Initial related vulnerability CVE-2026-46701 disclosed
- 2026-05-23: patched: Version 5.7.2 released to fix the incomplete authentication guard
- 2026-06-17: disclosed: CVE-2026-48814 published