AI

EU AI Act: where the tool stands

The platform uses an image analysis engine to propose an inventory. What it does, what it does not do, and how a person stays in control of the result.

Analysis workstation: rack photo with the equipment boxed and the list of detections on the right
The analysis workstation: each device is boxed on the photo with its confidence level; anything uncertain is marked “to check”.
Contents
  1. What the engine does, and only that
  2. Where the tool stands in the regulation
  3. A person stays in control
  4. Transparency
  5. Training users

What the engine does, and only that

The analysis engine receives a front photo of a rack. It detects the visible equipment and returns for each item a box, a family (network, fiber and transmission, power, server, patching), the model when it is readable, a U position and a confidence level. If the user uploads an export of your inventory, the platform then matches these detections with the lines of the file.

The result is an inventory proposal : a list of devices, a rack diagram and, where applicable, the differences with your file. The engine makes no decisions about people. It does not classify individuals, does not assess behavior and does not process biometric data.

Where the tool stands in the regulation

Regulation (EU) 2024/1689 on artificial intelligence classifies systems according to their use. The heaviest obligations apply to high-risk systems, listed in particular in its Annex III: recruitment, access to essential services, law enforcement, and safety components in the management and operation of critical infrastructure.

A tool that proposes an equipment inventory, reviewed by a person, and that writes to no system in your organization, is not intended to act as a safety component: it controls no equipment and triggers no action on the network. In our reading, it should therefore not fall under high-risk uses; this analysis remains to be confirmed with your DPO or your legal department, depending on your context of use.

This reading does not exempt you from your own analysis. The classification depends on how you use the tool, your sector and your processes. It is up to your organization, with your DPO, your CISO or your counsel.

A person stays in control

Whether or not the tool is high-risk, we designed it around a simple principle: the engine proposes, a person reviews the result and decides what to do with it. This principle is in line with the spirit of human oversight described in Article 14 of the regulation.

  1. Doubt is shown, not hidden

    Each device carries its confidence level. Below the threshold set by your administrator, it is marked “to check”: it is never presented as certain.

  2. The photo stays next to the result

    Each device is boxed on the photo. Zoom, full screen, search and filters let you check a detection at a glance.

  3. The diagram can be corrected before export

    On the rack diagram, a drag and drop moves a misplaced device before exporting the diagram.

  4. The platform writes nowhere

    The results stay on the platform and in the files you export. Nothing is written to your inventory or to any other system: a person decides what to take from them.

Transparency

A user must know that a result comes from an analysis engine and be able to judge its reliability. In the platform:

  • each result is presented as a proposal from the engine, with its confidence level;
  • uncertain devices are marked “to check”, on screen as in the exported inventory;
  • each analysis shows the engine used and its version, as well as its date;
  • the summary states the number of devices identified and the number of devices to check.

When you reopen an analysis from the History, you know which engine produced it and which lines required checking.

Training users

Article 4 of the regulation asks organizations that deploy an AI system (deployers) to ensure a sufficient level of AI literacy among the people who use it. For an inventory tool, this remains simple: understanding that the engine can be wrong, knowing how to read a confidence level, knowing when to retake a photo.

The onboarding offered during installation covers these points, for the people who take the photos as well as for those who use the results. Its content is defined within the project.

To receive the description of how the engine works and of the processing, contact us.

This page describes the product and how it is designed. It does not constitute legal advice: check how it applies to your context with your DPO, your CISO or your counsel.

Stay in touch

New releases, newly recognized equipment, articles: one email when there is something new, nothing else.