Label Studio: 4,451 title matches on an annotation platform that holds training data

Label Studio is an annotation tool. Teams use it to label text, images, audio and video so that the labels can train or evaluate a model. It is typically deployed as a web service that a project team reaches through a browser, and the data inside it is the dataset, the labels, and the credentials of everyone who contributed.

When that service is reachable from the internet, the exposure is a corpus rather than a machine.

What we measured

Queries run at 2026-10-13T22:48:42Z:

• title="Label Studio" — 4,451 results
• app="Label Studio" — 4,451 results The two queries returned identical counts, which is a cleaner result than many in this set. It suggests that where ZoomEye recognises the application, the identifying response also carries the product name in its title, so the two methods converge on the same set of reachable responses. Convergence does not make the number a count of vulnerable deployments; it does make the count a more consistent estimate of internet presence than a title-only query would be.

The unit is the reachable response. A single deployment serving a team of annotators is one response, whatever the size of the team.

Why an annotation service holds sensitive material

Annotation platforms sit at a stage where data is least protected by habit. A production database has a schema and an owner; an annotation project is often the place where raw material is copied so that humans can look at it.

Three categories of content accumulate. The first is the source data, which for many projects is the sensitive asset itself: internal documents, medical text, customer conversations, images from a facility. The second is the labels, which encode what the organisation understands about that data and can reveal the model's intended behaviour. The third is the account list. Each annotator has a login, and the platform's administrative accounts can export the entire project.

Label Studio's documentation describes the project model, the user management and the export formats. The export function is the property that matters most for impact assessment: it means an authorised account can extract the project's contents in bulk, so the size of the exposure is the size of the project.

What a reviewer should look at

Confirm whether the service is reachable from the public internet. Annotation platforms are frequently deployed for distributed teams, which is exactly the situation where a public endpoint is chosen for convenience. A VPN or an authenticated reverse proxy serves the same distributed team.

Review the account and project model. Determine who can sign up, who can create projects and who can export. Where open signup is enabled on a reachable instance, the boundary is the registration page.

Review how project data is transported between the platform and its storage. Where an object store or a database is used, the credentials for it are part of the deployment and should be scoped to the project.

Review retention. Annotation projects tend to keep their source data because re-import is expensive, which means the exposure window equals the life of the project.

What the measurement supports

The measurement supports the statement that a known annotation platform name appears in the titles of roughly four thousand four hundred internet-reachable responses, and that the application fingerprint agrees with the title query in this case. It does not establish that any of those instances lacks authentication, and it cannot distinguish a public demo from a production project holding real data.

The useful question for a team running the platform is about its own instance rather than the population: is the project reachable from outside the network, who can reach it, and what does the export function return to an account that should not have it?

What to do next

Confirm reachability and place the service behind a VPN or an authenticated proxy if external access is required for annotators. Disable open signup unless it is intentional, review the roles that can export, and remove accounts that are no longer active. Scope the platform's storage credentials to the project's own buckets or tables. Set a retention rule for source data and for completed projects so that the corpus does not persist indefinitely. Apply the current release, since the platform is actively developed. ZoomEye can be used to verify whether the organisation's own address space appears in the result set for the product title.

References

1. Label Studio documentation, https://labelstud.io/guide/
2. ZoomEye, search result for title="Label Studio", collected 2026-10-13T22:48:42Z, https://www.zoomeye.ai/