Learn AI /Responsible AI, Human Rights and Safety /Consent
Consent
- Starter
- Responsible AI, Human Rights and Safety
- Human rights practitioners
- Programme & data teams
Also called: informed consent.
Back to the milestone you came from
In five seconds
What it is
A person freely agrees to a specific use of them or their information after getting clear, relevant information and a real chance to say yes or no.
In plain words
In everyday terms, consent means giving permission after you have been told what will happen and you really can say yes or no.
An everyday example
Imagine Amina borrows a shared radio at a community centre. Before anyone turns the dial for her, the volunteer explains which station will play, how loud it will be, and that she can change it or switch it off at any time. Amina says she wants the news in Arabic, not music, and agrees only to that station for the next 15 minutes. If the volunteer later switches to a different station without asking, that original agreement no longer covers it.
A radio choice is simple and easy to reverse. AI uses can be harder to understand, can affect other people, and may have lasting effects once data has been copied, shared or used to train a system. Also, some AI activities need more than permission: they may require legal safeguards, human rights review, or should not be done at all.
Remember thisConsent must be informed, voluntary and specific. Agreement to one use does not quietly authorise another, and refusing must not bring an unfair penalty.
Understand it
Understand it
Why it matters
Consent affects whether people are treated with dignity and whether an AI-related activity has a legitimate social basis. For a non-specialist, the key question is simple: did the person really have enough information and a real choice for this specific use? If not, a project can harm people, damage trust and expose an organisation to legal, ethical and operational problems. Consent also helps decision-makers notice limits. A person may agree to one use of their data or content but not another. That matters when teams want to expand a pilot, combine datasets, keep information longer, or feed material into a model. Good practice therefore asks not only “Do we have permission?” but also “Is this use fair, safe, necessary and under human oversight?”
What makes consent meaningful
Show the difference between a mere click and consent that is informed, voluntary, specific and usable in practice.
01
1. Explain clearly
State what will be used, for what purpose, by whom, for how long, and the main risks and alternatives in plain language.
Draw a document card with short bullet lines and icons for purpose, person, time and risk; add a green check badge.
02
2. Offer a real choice
The person must be able to say yes or no without pressure, penalty or hidden default tricks.
Show two equal-sized buttons labelled Yes and No, with no pre-selected option; add a hand cursor between them.
03
3. Match the specific use
Agreement for one use does not automatically cover a different use later.
Show one labelled box 'translation today' linked by a solid line to consent, and a separate box 'model training later' linked by a dashed line with a question mark.
04
4. Record and respect it
Keep a clear record of the choice and apply it in practice, including any limits the person set.
Show a checklist with date, purpose and choice fields, then an arrow to a settings panel toggled to the chosen option.
05
5. Allow withdrawal or review
If feasible and appropriate, the person should be able to change their mind, and the organisation should explain what can and cannot be undone.
Show a circular arrow next to an account card, with a note icon saying 'some past uses may remain'.
A left-to-right step diagram with five numbered panels. Read in order from information given to later changes. Each panel shows one condition for meaningful consent and what happens if that condition is missing.
Hover or tap each panel to reveal a short example of valid versus weak consent in a tooltip.
Describe this without the diagram
This diagram presents consent as a five-step process. Step 1 says clear information must come first: purpose, user, duration, risks and alternatives. Step 2 says the person needs a real yes-or-no choice without pressure or manipulative defaults. Step 3 says consent is specific: agreement for one task does not automatically permit a different task later. Step 4 says the choice must be recorded and actually followed. Step 5 says withdrawal or review should be possible where feasible, with honest notice that some earlier uses may not be reversible.
With reduced motion: Show the same five panels at once in a static row or stacked column, with all labels and example notes visible and no animation.
Consent means permission that is informed, voluntary and specific. In AI work, the “input” is clear information given to the person: what data or material will be used, for what purpose, by whom, for how long, with what risks, and what choices they have. The person also needs a real opportunity to refuse or withdraw without pressure.
Then a process happens. The organisation explains the activity in plain language, checks that the person can reasonably understand it, and asks for agreement in a way that is appropriate to the context. If the person agrees, that agreement should be recorded. If they do not agree, the organisation should respect that choice.
The result is not just a tick box. It is a record of whether the person agreed to a particular action and under what limits. That record does not by itself make the action lawful, fair or safe. Consent is limited to the stated purpose. If the purpose changes later, fresh consent may be needed unless another lawful and rights-respecting basis applies.
Consent is often confused with privacy. Privacy concerns control over how identifiable information is collected, used, shared and kept. Consent is one possible permission for a specific use. It does not itself limit who can access information or protect it from loss or theft, and it does not replace duties such as safety, fairness, accessibility and human oversight.
In some situations, consent is not enough or may not be meaningful. For example, if there is a strong power imbalance, urgent need, or no realistic way to refuse, an apparent “yes” may not be truly voluntary. That is why responsible AI work also looks at accountability, human rights risks and whether some uses should not happen even with permission.
How “Tuning a radio station you choose” maps onto the real thing
| In the story | In the AI system |
|---|---|
| The volunteer explaining the station, volume and choices | The organisation giving clear information about the AI use, risks and options |
| Amina choosing one station for one time period | Consent being specific to a purpose and scope |
| Being able to change station or switch the radio off | The ability to refuse or withdraw consent |
| Switching to a different station later | A new use that may require fresh consent |
A worked example
An evaluation officer in Amman is arranging interviews for a monitoring report. Participants may choose a recorded interview that will be transcribed with an approved tool, or an unrecorded interview for which the officer takes notes.
Starting from: The notice explains the report’s purpose, who will receive the transcript, how long the recording will be kept, the role of the transcription tool, and how to withdraw before analysis. It states that declining recording will not affect participation.
- The officer provides the notice in an accessible language and format before the interview.
- Each participant separately chooses whether to be interviewed, recorded and quoted. No option is pre-selected.
- One participant agrees to the interview and recording but declines attribution in the monitoring report.
- The officer records that choice and labels the transcript for unattributed use only.
- When a colleague later proposes using the recordings to test another AI system, the evaluation officer checks the consent record and finds that this secondary use was never presented.
Result: The participant’s agreement covers recording and transcription for the monitoring report, without attribution. It does not authorise testing another system. The recordings must not be used for that purpose without an appropriate basis, necessary safeguards and, where relied upon, fresh consent.
In a United Nations workplace
As the 6pm clearance deadline approaches, a communications officer in a regional bureau receives recorded comments for a briefing note and considers using an approved AI transcription tool. The speakers agreed to recording for an internal meeting record, but were not told that their voices might be processed by AI or that quotations might be attributed to them.
Approval of the tool does not expand the speakers’ agreement. The officer must establish whether they received clear information about the proposed processing, recipients, retention and quotation, and whether they had a genuine choice. Consent for an internal record is not consent for every convenient secondary use.
Before you try thisDo not upload the recordings until the tool is approved for their sensitivity and the intended use is properly authorised. Personal, confidential, survivor, witness or security-sensitive material must not be entered into a tool that has not been approved for it. Consent does not override protection, confidentiality or data-handling requirements.
Where you may hear it
- During clearance of an interview protocol by an evaluation office
- In a procurement meeting when a vendor proposes using staff data to test a system
- Before a country office records a community consultation
- At a data protection review for an AI pilot
Avoid confusion
Avoid confusion
Often confused with
Privacy
Privacy means controlling how information about identifiable people is collected, used, shared and kept, so they are not exposed, tracked or harmed unfairly.
Confidentiality
Confidentiality means limiting access to information so only authorised people can see, use or share it for a legitimate purpose.
Data security
Data security means protecting digital or paper information from unauthorised access, change, loss or theft throughout its use, storage and sharing.
Common misconceptions
Common misconception: If someone clicked 'I agree', consent is valid.
Correction: A click alone does not prove meaningful consent. The person needs clear information, a genuine choice, and freedom from pressure or deceptive design. If key facts were hidden, the language was too complex, or refusing carried a penalty, the apparent agreement may be weak or invalid.
Common misconception: Consent for collecting data also covers any future AI use.
Correction: Consent is usually tied to a stated purpose. Using the same data later for a different task, such as model training, profiling or sharing with new partners, may go beyond what the person agreed to. A new use needs fresh explanation and may need fresh consent or another appropriate basis.
Common misconception: If people consent, the organisation has no further responsibility.
Correction: Consent does not remove duties around safety, fairness, non-discrimination, confidentiality, accessibility and accountability. Some uses can still be harmful or inappropriate even if people appear to agree, especially where there is a power imbalance or wider effects on others.
Quick check
Which situation best shows meaningful consent for an AI-related use?
The third option includes clear information, a choice without penalty, a specific purpose and fresh permission for a different use. The first attaches an unrelated penalty, the second exceeds the stated purpose, and the fourth fails to record or respect the person’s limits.
Go deeper
Go deeper
The simple version
In everyday terms, consent means giving permission after you have been told what will happen and you really can say yes or no.
The technical version
In more technical terms, informed consent is a process for obtaining a person’s voluntary, specific and sufficiently informed agreement to a defined activity, such as collecting, sharing or processing data, recording a voice, or using content in an AI-related workflow. Valid consent usually requires notice about purpose, scope, actors involved, likely consequences, duration, and relevant risks or alternatives, presented in a form the person can reasonably understand. It also requires a genuine ability to refuse, and often to withdraw later, without unfair detriment. Consent should be documented and linked to the exact purpose authorised. It is not identical to privacy, confidentiality or data security, and it is not a universal solution: in some contexts another lawful basis, additional safeguards, or a decision not to proceed may be required because consent would not be meaningful or sufficient.
Reviewed external resources
Primary sourceRecommendation on the Ethics of Artificial Intelligence (opens in a new tab)UNESCO
This UNESCO recommendation shows how consent fits into wider AI ethics and governance principles and suits policy, programme and oversight 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 helps readers connect consent to business, technology and human rights responsibilities, and suits people working on risk and accountability.
Intermediatelink checked 2026-07-31
EncyclopaediaConsent (opens in a new tab)Wikipedia
This Wikipedia page gives a broad introductory overview of consent and suits readers who want a quick starting point before applying the idea to AI.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Consent (opens in a new tab)YouTube search
This YouTube search results page lets readers compare different video explanations of consent and suits learners who prefer audiovisual introductions; it is a search page, not one curated source.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Consent (opens in a new tab)Reddit search
This Reddit search results page shows how people discuss and question consent in practice and suits readers who want to see common confusions and debates; it is a search page, not one authoritative source.
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
Consent must be informed, voluntary and specific. Agreement to one use does not quietly authorise another, and refusing must not bring an unfair penalty.