Please report suspected vulnerabilities privately via GitHub's "Report a vulnerability" button under the repository's Security tab (private advisory), rather than opening a public issue. Include steps to reproduce and the impact. You'll get an acknowledgement, and a fix or mitigation plan.

AI Command Center is designed to run next to your own apps, by default bound to 127.0.0.1. Its security posture:

• Your provider API keys pass through to the upstream provider unchanged; they are never written to the telemetry log.
• No prompt or response bodies are stored - only metadata (model, tokens, cost, latency, status, project).
• Cross-origin protection: the gateway only accepts browser requests from its own origin or origins you explicitly list in allowedOrigins. A random web page cannot spend your keys through the proxy or wipe your telemetry. Server-side callers (no Origin header) are unaffected.
• Central/operator keys are never handed to untrusted cross-origin callers.
• Auth: optional but on by default - open until the first admin account is created, then dashboard/API require a session and the proxy requires a per-project gateway key.

• Keep the default 127.0.0.1 bind, or put the gateway behind a VPN or a TLS-terminating reverse proxy if you expose it.
• Create the admin account immediately (don't leave it in open setup mode on a shared network).
• Rotate project gateway keys if one may have leaked (dashboard → Settings).
• Auth is username/password + signed session cookies today (no SSO yet) and telemetry is not encrypted at rest - appropriate for an internal tool; review before external multi-tenant exposure.

The latest 0.x release receives fixes. Pin a version and watch releases.