Learn AI /Responsible AI, Human Rights and Safety /AI ethics
AI ethics
- Starter
- Responsible AI, Human Rights and Safety
- Policy officers
- Human rights practitioners
Also called: ethics of AI.
Back to the milestone you came from
In five seconds
What it is
AI ethics is the practice of checking whether AI systems are fair, safe, accountable and respectful of human rights before and during their use.
In plain words
AI ethics means checking whether an AI tool should be used, who it could help or harm, and what human safeguards are needed.
An everyday example
A school office wants to use a new translation tool to send one message to families in several languages. Before sending it, staff check whether the wording could confuse people, whether some families will not be reached, whether private details were included by mistake, and who will answer if the message is wrong. They also decide that emergency messages must still be reviewed by a person before they go out. Only after those checks do they use the tool. If the risks are too high, they do not send through the tool at all.
This analogy is simpler than real AI ethics. A school message is one short task, while AI systems can affect many groups over time and can fail in less visible ways. AI ethics also covers issues the story does not show clearly, such as discrimination in training data, labour conditions, environmental impact and accountability across suppliers and deployers.
Remember thisAI ethics asks whether an AI use is justified at all, then requires evidence, safeguards, accountability and continuing review, not a one-off statement of good intentions.
This term is contested
The broad idea is widely recognised, but different organisations emphasise different principles, such as human rights, fairness, safety, autonomy, justice or sustainability, so the exact scope is not fully settled.
Understand it
Understand it
Why it matters
AI systems can shape who gets information, services, attention or opportunities. If the design or use is careless, people can be treated unfairly, exposed to privacy risks, or pushed into decisions that should stay with humans. AI ethics gives non-specialists a way to ask better questions before harm occurs: What is this system for? Who might be missed or misjudged? What evidence supports its use? Who is accountable? What human review is needed? These questions help organisations make decisions that are lawful, defensible and more trustworthy in practice.
How AI ethics guides a use decision
Show that AI ethics is a practical review process: define the purpose, identify affected people, check risks, add safeguards, then decide whether and how to use the system.
01
1. Define the use
State what the AI will do, who will use it and what decision it will influence.
Draw a card with a target icon and three short lines: task, users, decision.
02
2. Identify who is affected
List the people and groups who may benefit, be excluded or face harm, including people not in the room.
Draw a card with three person icons of different sizes and a label: affected groups.
03
3. Check main risks
Review fairness, non-discrimination, privacy, safety, transparency, accessibility and over-reliance on automation.
Draw a card with a checklist icon and seven short tick items.
04
4. Add safeguards
Set limits, human review, testing, documentation, complaint routes and monitoring.
Draw a card with a shield icon and six small safeguard labels.
05
5. Decide and review
Choose to use, limit, redesign or reject the system, then review its effects over time.
Draw a card with four branching buttons: use, limit, redesign, reject; add a loop arrow back to step 2.
A five-step horizontal sequence with arrows. Each card names one stage in an ethics review, from defining the task to deciding use conditions. Read left to right. Small note under the final step says the process can loop back if risks remain too high.
Hover or tap each step to expand one sentence of detail; the loop arrow highlights when step 5 is selected.
Describe this without the diagram
The diagram is a five-step process read from left to right. Step 1 says define the AI use: what task it performs, who uses it and what decision it affects. Step 2 says identify affected people and groups, including those who may be excluded or harmed. Step 3 says check major risks such as fairness, non-discrimination, privacy, safety, transparency, accessibility and over-reliance on automation. Step 4 says add safeguards such as limits on use, human review, testing, documentation, complaint routes and monitoring. Step 5 says decide whether to use, limit, redesign or reject the system, and then keep reviewing outcomes over time. A loop from step 5 back to step 2 shows that ethics review is ongoing, not a one-time checklist.
With reduced motion: A static five-card sequence with arrows between cards and a visible loop arrow from the final card back to the second card. Expanding details appear instantly without animation.
AI ethics is about asking practical questions before using an AI system: Who could be helped? Who could be harmed? Who is left out? Who is responsible if something goes wrong? The goal is not only to make a system work, but to make sure its use is acceptable and aligned with human values, rights and duties.
What goes in are choices: the purpose of the system, the data used, the rules for using it, and the people and organisations affected by it. What happens is a structured review. People examine risks such as bias, privacy loss, unsafe advice, exclusion, weak oversight or over-reliance on automation. They then set limits, testing, monitoring and human review.
What comes out should be clearer decisions about whether to use the system, how to use it, and what safeguards are needed. Sometimes the ethical answer is to redesign the system, narrow its use, or not use it at all. AI ethics therefore guides action; it is not just a list of abstract principles.
This term is often confused with responsible AI, AI governance and AI safety. AI ethics focuses on what ought to happen and why. Responsible AI is broader and includes putting those ideas into practice. AI governance is about the rules, roles and oversight structures around AI. AI safety often focuses more narrowly on preventing harm or dangerous behaviour, though the boundaries can overlap.
How “Checking a school loudspeaker announcement before pressing send” maps onto the real thing
| In the story | In the AI system |
|---|---|
| the translation tool for school messages | an AI system used for a real task |
| checking confusion, privacy and who is reached | ethical review of harms, fairness, privacy and inclusion |
| a person reviewing emergency messages | human oversight for high-stakes uses |
| choosing not to use the tool if risk is too high | deciding to limit, redesign or reject an AI use case |
A worked example
A public transport complaints desk is considering an AI tool to sort passenger messages before its weekly accessibility report is compiled.
Starting from: A de-identified test set of messages covering delays, ticketing, staff conduct and accessibility barriers, including messages in several languages.
- The complaints officer defines the purpose narrowly: the tool may suggest a category, but may not decide priority, close a case or draft a final reply.
- Staff compare the suggested categories with human classifications and find recurring errors in messages describing less obvious accessibility barriers and in some language varieties.
- They assess whether the use remains justified and proportionate, considering who could be disadvantaged, whether staff can identify errors and whether manual sorting would be safer.
- They require human review of possible accessibility cases, record overrides, test performance across message types and languages, and establish a route for passengers to correct a classification.
- The desk limits deployment to first-pass sorting and agrees to pause the tool if monitoring shows persistent unequal error patterns.
Result: The tool is used only where it can save administrative time without determining how a passenger is treated. Its classifications remain reviewable, monitored and attributable to a responsible officer.
In a United Nations workplace
At a regional bureau, a newly appointed associate expert is reviewing terms of reference for an AI-enabled service desk before the file goes to clearance on Friday. The draft promises faster routing and standard replies, but says nothing about which tickets are out of scope, how errors can be challenged or who remains accountable.
AI ethics turns those gaps into requirements before clearance: a justified and limited purpose, documented responsibilities, testing across languages and accessibility needs, human review for sensitive tickets, audit logs, continuing monitoring and a clear basis for pausing the system if harms emerge.
Before you try thisService-desk tickets can contain personal, HR, security-sensitive or procurement-sensitive information. They must not be copied into a vendor demo or any tool not approved for those data, because this could expose individuals and internal operations. Use approved test data, apply access and retention controls, and keep consequential routing or replies under meaningful human review.
Where you may hear it
- A Friday clearance meeting on terms of reference for an AI-enabled service desk
- An ICT governance board reviewing an impact assessment before deployment
- A procurement meeting where a vendor is asked to demonstrate accessibility, audit logs and human override
- A country office review after staff report recurring errors across language groups
Avoid confusion
Avoid confusion
Often confused with
Responsible AI
An approach to designing, using and governing AI so it is safe, fair, accountable and under appropriate human control.
AI safety
AI safety means designing, testing and using AI systems so they are less likely to cause harm, especially in situations where people could be affected.
AI governance
The rules, institutions and processes used to guide how AI is built, used and overseen so its benefits and risks are handled in the public interest.
Side-by-side
Common misconceptions
Common misconception: AI ethics is just a philosophical debate with no effect on real systems.
Correction: Ethical concerns become practical choices: what data can be used, which tasks are off-limits, what testing is required, when a person must review outputs, and whether a system should be deployed at all. In organisations, these choices shape procurement, design, approval and monitoring.
Common misconception: If an AI system is accurate on average, the ethics problem is solved.
Correction: Average accuracy can hide unequal errors, privacy harms, unsafe uses or lack of accountability. A system can perform well overall and still fail certain groups, be used for the wrong purpose, or influence decisions that need human judgement.
Common misconception: AI ethics means making machines moral like people.
Correction: In most real-world discussions, the focus is not on giving a system human morality. The focus is on how people and organisations design, buy, deploy and supervise AI, and whether those choices respect rights, safety and accountability.
Quick check
Which example best fits AI ethics in practice?
AI ethics is a structured review of impacts, rights, risks and responsibilities. The strongest answer includes affected people, safeguards and limits on use. Cost or efficiency alone is not enough, and high accuracy does not remove the need for oversight and risk checks.
Go deeper
Go deeper
The simple version
AI ethics means checking whether an AI tool should be used, who it could help or harm, and what human safeguards are needed.
The technical version
AI ethics examines the normative and practical conditions for developing and deploying AI systems. It looks at whether a use case is justified, proportionate and respectful of rights and social values; how data, model behaviour and system design may create unfair, unsafe or opaque outcomes; and what controls are needed. These controls can include purpose limitation, documentation, impact assessment, human oversight, testing across groups, privacy and security protections, accessibility measures, complaint mechanisms, monitoring after deployment and clear accountability. The term overlaps with responsible AI, safety and governance, but it is centred on evaluative questions about what ought to be permitted, required or prohibited in the design and use of AI.
Reviewed external resources
Primary sourceRecommendation on the Ethics of Artificial Intelligence (opens in a new tab)UNESCO
This UNESCO recommendation gives an official international framework for the ethics of artificial intelligence, and it suits policy, governance and public-sector readers.
Intermediatelink checked 2026-07-31
Primary sourceB-Tech Project (opens in a new tab)UN Human Rights Office (OHCHR)
This OHCHR B-Tech material gives a human rights lens on technology risks and responsibilities, and it suits readers focused on business conduct, rights impacts and remedy.
Intermediatelink checked 2026-07-31
EncyclopaediaWikipedia search: AI ethics (opens in a new tab)Wikipedia
This Wikipedia search results page helps readers quickly see the range of topics and articles linked to AI ethics, and it suits general readers who want a starting map rather than one fixed source.
search results, not one pagelink checked 2026-07-31
VideoVideo explanations: AI ethics (opens in a new tab)YouTube search
This YouTube search results page helps readers find different video explainers on AI ethics, and it suits people who prefer audio-visual introductions and comparisons.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: AI ethics (opens in a new tab)Reddit search
This Reddit search results page shows community discussions and questions about AI ethics, and it suits readers who want to see common public concerns and informal debate.
search results, not one pagelink checked 2026-07-31
Where this came from
Drafted with a language model through an offline pipeline, checked against this site's schema, and every external link above fetched to see whether it answers. Treat it as a starting point and check anything that matters against the primary sources. How this site is built and what that means.
The one thing to remember
AI ethics asks whether an AI use is justified at all, then requires evidence, safeguards, accountability and continuing review, not a one-off statement of good intentions.