# AI Impact Assessment

> A written check of how an AI could hurt people. It is done before launch.

**Field:** Artificial Intelligence  
**Category:** Ethics & Regulation  
**Also known as:** Algorithmic Impact Assessment, Fundamental Rights Impact Assessment, FRIA

## Think of it like

It is like asking who might slip before you mop a floor. You put up a sign first.

## Definition

An AI impact assessment is a written check, done before launch, of how an AI system could affect people. It asks who will use it, who it makes choices about, and what could go wrong for them. It looks at bias, privacy, safety and people's rights. Then it lists the steps to cut each risk, and who owns each step. Under the EU AI Act, some users of high-risk systems, such as public bodies, must do one for people's basic rights.

## Example

A city plans an AI that ranks which homes to inspect first. Its team writes an impact assessment and finds poor areas would be checked far more. They change the model before launch.

## Why it matters

It is far cheaper to find a harm on paper than after it has hurt people.

## Common mistake

Writing it once and filing it away. Update it when the system, or the way it is used, changes.

## Three principles

- **Before launch.** Check first, then go live.
- **People first.** Who could be harmed, and how.
- **Owners for fixes.** Each risk gets a named fix.

## Related terms

- [AI Governance](https://learn.benibix.com/term/ai/ai-governance)
- [High-Risk AI System](https://learn.benibix.com/term/ai/high-risk-ai-system)
- [NIST AI Risk Management Framework (AI RMF)](https://learn.benibix.com/term/ai/nist-ai-rmf)
- [ISO/IEC 42001](https://learn.benibix.com/term/ai/iso-iec-42001)
- [AI Bias](https://learn.benibix.com/term/ai/ai-bias)
- [Data Privacy](https://learn.benibix.com/term/ai/data-privacy)
- [Data Protection Impact Assessment (DPIA)](https://learn.benibix.com/term/data/data-protection-impact-assessment)

## Told apart from

- [DPIA vs AI Impact Assessment](https://learn.benibix.com/compare/dpia-vs-ai-impact-assessment): People think a DPIA and an AI impact assessment are the same check, and they are not. A DPIA looks only at the risks of using personal data. The GDPR asks for it. An AI impact assessment looks at every way an AI system could harm people, such as bias or safety. It counts even when no personal data is used. One AI project can need both.
- [Conformity Assessment vs AI Impact Assessment](https://learn.benibix.com/compare/conformity-assessment-vs-ai-impact-assessment): People think the conformity check and the rights impact check in the EU AI Act are one test, and they are not. The maker runs the conformity check before the system is sold. It shows the system meets the rules on data, logs and safety. Some users run the impact check before they switch it on. Public bodies and firms that score credit must do it. It asks who the system could harm in their own hands.

![Two standards side by side. On the left, the NIST AI Risk Management Framework: Govern in the middle, with Map, Measure and Manage around it. On the right, ISO/IEC 42001: a loop of Plan, Do, Check and Act. Below both, an AI impact assessment, with dashed arrows into Map on one side and Plan on the other.](https://learn.benibix.com/diagrams/ai4-ai-standards.svg)

---

Source: https://learn.benibix.com/term/ai/ai-impact-assessment
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.
