Container and virtual machine control planes sit close to the metal. A user who can create an instance on the host is not a user of an application; they are a tenant of the infrastructure, with whatever the kernel will allow.
ZoomEye measurements collected on 28 October 2026 put this in proportion. app="LXD" returns 12,241, and app="LXD" && port="8443" returns 11,685, which tells you that most indexed LXD surfaces answer on the HTTPS port rather than elsewhere. title="LXD" returns 1,137. For Incus, the community fork, app="Incus" returns 15 and title="Incus" returns 359, with title="Incus" && port="8443" at 257.

Why the numbers diverge here

LXD and Incus are related codebases serving the same role, and the fingerprint results could hardly be more different. LXD has a widely deployed application signature, so the app count dominates. Incus, newer as an independent project, is better captured by its page title than by any app signature, which is why the title count is more than twenty times the app count.
This is a general lesson about measurement rather than a special fact about containers. A fingerprint is maintained, versioned and occasionally slow to recognise a fork. Reading only the app count for a younger project will systematically undercount it, and reading only the title count for a mature one will overcount it with documentation pages and mirrors.
Neither number, in either case, indicates whether the control plane enforces authentication.

What an exposed control plane means

The administrative surface of these tools is an API. Creating an instance is a POST. Attaching a device is a PUT. Executing a command inside a running container is a documented capability, because it is exactly how an operator troubleshoots. Every one of those operations is a legitimate feature, and each is a privilege that matters when the caller should not have it.
The consequence of an unauthenticated caller reaching that API is therefore not "a container appears". It is code execution on a host that is, by design, more privileged than the workloads it contains. Container escapes are a category of vulnerability precisely because that boundary is thin; an attacker who starts on the outside of it does not need to cross anything.
The 11,685 LXD assets answering on 8443 deserve a specific reading. That port is where the HTTPS API listens by default, so a large share of indexed LXD surfaces are presenting the administrative interface itself rather than a workload. Some of those will be labs, some will be development clusters, and some will be production hosts behind a firewall rule that was widened for a reason nobody recorded.

The review that matters

Confirm whether the API requires a client certificate or a trust token before it accepts a request. Both mechanisms exist in the documented feature set, and both are configuration rather than defaults that can be assumed.
Enumerate the projects and the users who can reach them. A single-user host is a different risk from a shared host with several tenants, and the difference lives in the permission model rather than in the network diagram.
Check the listening address. A control plane bound to a private interface with a VPN in front is in a different category from one bound to a public address with a password, and the two are frequently confused when the discussion turns to "is it exposed".
Finally, look at what the host can reach. A container host has network position, storage access and often credentials for the image registry it pulls from. The blast radius of a control plane compromise is defined by the host's own privileges, not by the containers it runs.

A modest number with a sharp edge

12,241 and 15 are small numbers next to the hundreds of thousands that a web framework produces. That is the right scale for infrastructure tooling: fewer deployments, each with more authority.
The appropriate response to a small count is not indifference. It is to notice that these are the systems where a single misconfiguration has the shortest path from "reachable" to "running code". The measurement is worth taking because it is cheap, and worth acting on because the alternative is trusting a network rule that nobody has read since it was written.

References

  • Incus documentation: https://linuxcontainers.org/incus/docs/main/

  • ZoomEye query app="LXD", 2026-10-28: https://www.zoomeye.ai/searchResult?q=YXBwPSJMWEQi

  • ZoomEye query app="Incus", 2026-10-28: https://www.zoomeye.ai/searchResult?q=YXBwPSJJbmN1cyI%3D
    Measurement notes: counts were collected on 28 October 2026 through the ZoomEye SDK. They describe indexed assets and do not indicate whether authentication is enforced on any control plane.