Exposed AI Agent Control Panels: Dashboards That Should Never Be Public

The new class of admin panel

AI agents and LLM applications usually come with a web interface: a dashboard to configure the agent, connect accounts, set up tools and read conversation history. Open-source projects make these easy to self-host, and employees increasingly set them up without involving IT.

Unlike a typical admin panel, an agent control panel is often wired directly into sensitive systems. It may hold access to email, calendars, files, internal databases and cloud accounts, because that access is what makes the agent useful. If the panel is open, the access is open.

What the data shows

Censys reported more than 54,000 exposed control panels for OpenClaw, an open-source AI agent that can connect to personal accounts and other services, as part of its 2026 State of the Internet findings (Axios, 13 October 2026). Censys's CEO noted that companies already struggle to track their existing software, and AI applications built by employees add to that difficulty. As with all internet-wide data, exposure does not prove that every panel can be exploited.

Check Point's AI Security Report 2026 also notes that attackers continuously probe exposed model servers, agent control panels and inference endpoints, many of which the owning organisations cannot see (Help Net Security).

Why this matters more than a normal exposed panel

• Broad access in one place. One panel can unlock many connected accounts.
• Autonomy. Agents act on instructions. An attacker who controls the panel may be able to direct the agent to read, send or change things.
• Weak defaults. Self-hosted tools are often run with default settings, no authentication or shared passwords, because they began as experiments.
• Poor visibility. Because the tools are set up outside normal change processes, security teams may not know they exist.

How to find them

1. Ask. Survey engineering, data and business teams for AI tools they run or have tried.
2. Search your perimeter. Review DNS records, cloud assets, reverse proxy configs and open ports for new hosts and services you do not recognise.
3. Test from outside. Open suspected panels from a device off your network. If a login page or dashboard loads, it is reachable.
4. Check for a login at all. Some panels ship with authentication disabled by default.

How to secure agent panels

1. Never publish the panel directly. Put it behind a VPN, an identity-aware proxy or an IP allowlist.
2. Require SSO or strong authentication with MFA for every user, and remove default accounts.
3. Scope connected accounts. Give agents the narrowest access that works, with dedicated service accounts instead of personal ones.
4. Add human approval for high-impact actions such as sending messages externally, deleting data or moving money.
5. Keep a kill switch. Know how to revoke every token an agent holds, quickly.
6. Log and review. Keep audit logs of agent actions and panel logins.
7. Run an AI inventory. Maintain a list of every AI tool in use, who owns it, what it connects to and where it is hosted.

How SurfaceScan helps

Shadow AI shows up on your perimeter as new subdomains, new hosts and new open ports. SurfaceScan continuously discovers your external assets and highlights what has changed since the last scan, so an unrecognised dashboard becomes a question for the right team instead of a surprise. See the guide on what Shodan reveals about your infrastructure for how attackers discover the same assets.