GUIDE · AI ACT

High-risk AI systems: the obligations of those who use them in the company

This reference guide explains the obligations the AI Act places on those who use a high-risk AI system without being its developer. After reading it you will be able to recognise your role, identify the activities required and choose which deeper guide to start from. The picture depends on the specific case and must be confirmed with an adviser.

Updated october 2026
9 min read
Altovar Technical Editorial Team

1. Who the deployer is, and why the role matters

The Regulation distinguishes between those who develop an AI system and those who use it. The latter, in some texts called the deployer, meaning the party that puts the system into operation. Anyone who buys staff-selection software and uses it in their own processes is a deployer, even if they have never written a line of code. This distinction matters because the obligations differ, and a company can hold different roles for different systems.

The role is not determined by label. If you substantially customise the system, change its intended purpose or place it on the market under your own name, your position may change, and with it your obligations. Before planning the work, therefore, check how the supplier has described the product and which modifications you plan to make to your use.

2. When a system is high-risk

A system is high-risk in two main cases: when it is a safety component of a product already regulated by other European legislation, or when its use falls within the list of use cases set out in the Regulation. That list includes areas such as personnel management, credit, certain essential services and education. Not all uses in those areas are high-risk: what matters is the concrete function the tool performs in your process.

The first operational step is the classification described in the dedicated guide. Here it is enough to remember that it must be carried out on the company's use, not on the product's datasheet. A generic text analysis system can become high-risk the moment it is used to evaluate candidates, and at that point the rules to follow change. The check must be repeated every time the function changes, not only at the time of purchase.

3. Practical obligations for the deployer

For a high-risk system the Regulation asks the deployer to take a number of concrete actions. These are not formalities: each requires an owner and written evidence. The detail must be confirmed with an adviser, but the general framework is the one set out below, to be adapted to your sector and your work process. Use the list as a reference and adapt the items to your process, adding those that are genuinely needed.

Oversight and logs are the items companies most often neglect. A person who automatically approves every proposal from the system is not overseeing anything. Establish clear criteria for when a decision must be reviewed, who can interrupt use and within what timeframes the owner must act when a problem emerges. Document the choice of the people who oversee the system, with their role and the training they have received.

  • Use the system in accordance with the instructions the supplier provides.
  • Entrust human oversight to competent people, with the authority to intervene or stop the system.
  • Check that the input data is relevant to the stated purpose.
  • Monitor operation and report significant problems to the supplier.
  • Keep the logs generated by the system for the period required by the applicable rules.
  • Inform the individuals concerned when the system intervenes in a decision about them.

4. Workers and affected individuals

Where the system affects workers, the Regulation in some cases requires that workers and their representatives be informed before the system is used in the workplace. National employment rules may add further steps, such as consultation with employee representatives. Involvement must be planned with your employment adviser before the tool is activated, not afterwards. If the requirement for involvement is unclear in your case, ask your adviser to confirm before the system goes live.

Individuals who are the subject of a decision assisted by the system are entitled to clear information about the role the system played. Prepare a short, understandable text to include in the process, and a procedure for answering questions or challenges. This is not a matter of form: it is the point at which the system meets the person, and it is often the first thing that gets challenged.

5. The fundamental rights impact assessment

For some deployers, notably public bodies or private entities that provide public services, and for specific uses such as certain creditworthiness assessments or certain risk assessments for life and health insurance, a fundamental rights impact assessment is required. It depends on the type of entity and the use. Check with your adviser whether it applies to you, because if it does, it must be carried out before the start of use.

The assessment describes what the system is for, who is involved, what risks to individuals' rights may arise and which measures reduce them. It is a working document to be updated, not a box to tick once. It must be treated as a dedicated guide, because it has its own structure and its own preparation path. Check that the text is understandable even to someone who does not know the system.

6. Incidents, malfunctions and communications

If the system produces a serious outcome or a significant malfunction, the deployer must inform the supplier and, in the cases provided for, the competent authorities. Define in advance what counts as a serious incident in your context: an isolated error, a repeated pattern of errors, an unjust exclusion of a person. A written definition avoids discussion during the emergency, when time is short. Record the outcomes of every check, positive or negative, so that you can demonstrate the work done.

Prepare an internal procedure too: who is alerted, who can suspend the system, how the case is documented and how to return to the manual process. This link with business continuity is often the most useful element, because a system that is stopped must be replaced by a process that really works. The link with the continuity plan must be verified with the people who run the operational processes.

7. Contracts and suppliers: what to ask for

The contract with the supplier is the tool through which you obtain the information you need. Ask for technical documentation on the intended use, operating instructions, known limitations, data requirements, update procedures and the incident reporting process. If the supplier cannot answer these points, the answer itself is information to record as a risk. Also ask for the supplier's commitment to report changes that affect your use, not only the technical ones.

Also check who can access the data and where it is stored, keeping the two questions separate. A supplier that guarantees accuracy in generic terms does not give you the elements needed to demonstrate your due diligence. Request in writing what you need to govern the system, from the description of how it works to future changes. If the answer is unclear, treat the risk as not yet managed and set a deadline to close the point.

8. Topics to explore in the cluster

This guide serves as a map. The questions that deserve separate deep dives are: how to classify systems by risk level, which practices are prohibited, how to prepare the fundamental rights impact assessment, how the AI Act and the GDPR are coordinated, and what changes for human resources. Each deferral only makes sense if the topic stays on the agenda with an owner and a date.

If you already have a system in use, read the guide on classification first. If you use AI to select or evaluate people, move to the guide on human resources. If you build a product on a foundation model supplied by a third party, the guide on general-purpose models will help you understand which information to request. In this way the right guide arrives at the right time, and does not remain an intention.

9. How to start

This week, choose the system that comes closest to a high-risk use and ask the supplier for three documents: the instructions for use, the description of the intended use and the information on oversight. With these documents you can establish whether your role and your scope are clear, or whether more information is needed before going further. You do not need to be fully compliant to start: you need to know where to begin and who decides.

Then assign an internal owner for that system and hold a first meeting with the compliance adviser and, if involved, the HR office. A complete plan is not needed immediately. What is needed is to know who decides, who oversees and which information is still missing. Also ask them to note the date of the next review, so that the work does not stop after the first meeting.

Frequently Asked Questions

I am a company using a recruitment software: am I automatically high-risk?

Not automatically. What matters is how you use the software. If the use includes selecting or evaluating candidates, it falls within an area the Regulation considers sensitive and must be checked carefully. If the tool only organises documents or sends communications, the picture changes. Classify based on the actual use and record the reasons.

Is formal human oversight enough?

No. Oversight must be real: a competent person, with the time and authority to verify and, if necessary, stop the system. If the person approves every proposal without being able to discuss it, oversight is merely cosmetic. Document who oversees, with what training and under what intervention procedure.

Do I have to keep the system's logs?

Generally yes: you are required to keep the logs the system generates, for the period set by the applicable rules and your context. The precise duration must be checked with an adviser. Coordinate retention with data protection rules, because logs can contain personal data and must be handled accordingly.

Must the supplier tell me how the system works?

It must provide enough information to use it correctly, such as instructions, limitations and necessary data. It does not necessarily have to reveal every internal technical detail. For your work you need operational information, in writing. If the answer is vague, ask for a more precise one before using the system with real people.

What happens if the supplier updates the system?

An update can change functions, limitations and risks. Ask the supplier for a description of the changes and assess whether the classification or the oversight plan needs revising. Do not assume an update is neutral, especially if it changes the way the system evaluates people.

From when do these obligations apply?

The provisions apply in phases and may be modified by subsequent measures. It is not correct to fix a plan on a date read in a summary. Check the up-to-date official text and the timeline with your adviser before making internal commitments, and plan accordingly.

Is relying on the supplier enough to be compliant?

No. The supplier has obligations for the product it develops, but use within your organisation is your responsibility. The contract helps distribute information and responsibilities, but it does not remove the need for oversight, checks and internal documentation. Keep the supplier's written answers and your internal decisions, because they are the proof that you governed the system.

About this content

General informational guide. Regulations and technical requirements change over time: always check the official sources and assess your specific case with a qualified technical or legal advisor.

GET IN TOUCH

Talk to our team.

We'll show how Altovar integrates into your infrastructure. No commitment, no sales pitch. Just a technical conversation.

BOOK A DEMO →