# High-Risk AI System

> An AI used where one mistake could badly hurt a person.

**Field:** Artificial Intelligence  
**Category:** Ethics & Regulation  
**Also known as:** High-Risk AI, High Risk AI Use

## Think of it like

It is like a strong medicine. You can have it, but only with a doctor's note.

## Definition

A high-risk AI system is an AI used where a mistake could badly harm a person's health, safety or rights. Examples are tools that screen job seekers, score credit, grade exams, or help police and border staff. Under the EU AI Act, these must meet strict rules before and during use. They need good data, a log of what the system did, clear papers, and a person who can step in and stop it. The label comes from the use, not the tech. So one model can be high risk in one job and not in another.

## Example

Lucy's school wants an AI to grade final exams. The grade shapes each student's future, so it counts as high risk. A teacher must check each mark a student disputes.

## Why it matters

Where the stakes are high, mistakes cost a lot and are hard to undo. So the checks grow with the harm.

## Common mistake

Judging risk by how advanced the model is. A simple model that scores loans can be high risk, while a powerful one that writes poems is not.

## Three principles

- **Serious harm possible.** Health, safety, rights.
- **Strict checks.** Data, logs, a person.
- **The use decides.** Not the tech.

## Related terms

- [EU AI Act](https://learn.benibix.com/term/ai/eu-ai-act)
- [AI Regulation](https://learn.benibix.com/term/ai/ai-regulation)
- [Human-in-the-Loop](https://learn.benibix.com/term/ai/human-in-the-loop)
- [Conformity Assessment (EU AI Act)](https://learn.benibix.com/term/ai/ai-act-conformity-assessment)
- [Automated Decision-Making (ADM)](https://learn.benibix.com/term/ai/automated-decision-making)
- [AI Impact Assessment](https://learn.benibix.com/term/ai/ai-impact-assessment)

## Told apart from

- [General-Purpose AI Model vs High-Risk AI System](https://learn.benibix.com/compare/general-purpose-ai-model-vs-high-risk-ai-system): People think a powerful general model is a high-risk AI system under the EU AI Act. It is not the same thing. A general-purpose model is judged as a model, and its maker has its own duties. A high-risk system is judged by its use, such as sorting job forms or scoring loans. A small model can be high risk, and a big one need not be.

![A hiring tool marked high risk, with the checks it needs before use: good data, logs, a person who can step in, and clear papers.](https://learn.benibix.com/diagrams/et-high-risk-ai-system.svg)

---

Source: https://learn.benibix.com/term/ai/high-risk-ai-system
Author: Behnia Zarei (a developer in Vienna, Austria, who builds and teaches data, AI and business intelligence at BENIBIX)
Publisher: BENIBIX Learn (https://learn.benibix.com/)
Last updated: 2026-11-18
Licence: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote freely with attribution.
