Learn AI /Responsible AI, Human Rights and Safety /Human rights due diligence
Human rights due diligence
- Intermediate
- Responsible AI, Human Rights and Safety
- Human rights practitioners
- Policy officers
- Managers & front offices
Also called: HRDD, due diligence.
Back to the milestone you came from
In five seconds
What it is
An ongoing process to identify rights harms linked to an organisation’s actions or AI systems, address them, and account for the response.
In plain words
Before using AI, check how it could affect people’s rights, change the plan to avoid harm, keep watch during use, and make sure someone is clearly responsible.
An everyday example
A records team in Panama City receives thousands of photos of paper forms from different field offices. Before using a tool that matches each photo to the right template, the team tests blurred images, different languages, low light and unusual page layouts. They ask colleagues from several offices where mistakes would cause the most trouble, such as sending a health form into the wrong workflow. They then limit where the tool may be used, require human review for uncertain matches, and keep a log of errors so they can improve or stop the process if harm appears.
This analogy focuses on operational mistakes, but human rights due diligence is broader. Real cases may involve privacy, discrimination, freedom of expression, labour conditions, security risks or barriers to remedy, not just wrong sorting. It also applies to organisational relationships such as vendors and deployment partners, not only to one team testing one tool.
Remember thisHuman rights due diligence is not a one-off checklist. Identify possible harms, act on them, track what happens and account for the response throughout the system's use.
Understand it
Understand it
Why it matters
A tool can be efficient and still create serious harm. Human rights due diligence gives leaders and practitioners a disciplined way to ask the right questions before, during and after use: who could be affected, what rights are at stake, how serious the harm could be, what controls are needed, and who is accountable if something goes wrong. It supports better decisions about whether to proceed, redesign, limit use or stop. It also helps organisations show that they did more than trust vendor claims or headline accuracy numbers. For non-specialists, the practical value is clear: fewer blind spots, clearer responsibility, and a better chance of catching harm early rather than explaining it after people are affected.
Human rights due diligence as an ongoing cycle
Show that human rights due diligence is a continuing process that starts before deployment and continues through monitoring, response and accountability.
01
1. Define the use
State what the system does, where it will be used, who may be affected and what decisions depend on it.
Draw a box with icons for a tool, a place and different groups of people. Add a small note: 'purpose, users, context'.
02
2. Identify rights risks
List possible impacts on privacy, non-discrimination, access, safety, expression or other rights, and note who faces the highest risk.
Draw a box with a checklist and warning markers next to different groups. Use short labels such as 'privacy' and 'exclusion'.
03
3. Prevent and reduce harm
Change the design or use, set limits, add human oversight, improve data and create ways to contest or correct outcomes.
Draw a box with a shield, a human review icon and a settings slider. Add tags: 'limits', 'review', 'appeal'.
04
4. Decide and assign accountability
Record what residual risk remains, who approved the use, who monitors it and what triggers suspension or escalation.
Draw a box with a signature line, named roles and a red threshold marker labelled 'stop/escalate'.
05
5. Track, report and remedy
Monitor outcomes, log incidents, listen to complaints, update controls and provide remedy where harm occurs.
Draw a box with a dashboard, feedback messages and a loop arrow returning to step 1.
A left-to-right step diagram with five connected boxes. Read from planning to ongoing monitoring. Each box shows one action, one key question and one output. A thin return arrow from the last box back to the first shows that the process repeats when context, data, users or harms change.
On hover or tap, each step expands to show one example question: 'Who could be excluded?', 'What happens if confidence is low?', 'Who can challenge a result?'
Describe this without the diagram
The diagram presents five steps in a repeating sequence. Step 1 defines the intended use, setting and affected people. Step 2 identifies potential impacts on rights and highlights groups at higher risk. Step 3 shows actions to prevent or reduce harm, such as design changes, limits on use, human review and appeal routes. Step 4 assigns accountability by recording remaining risk, responsible roles and conditions for stopping or escalating use. Step 5 tracks real-world outcomes, complaints and remedy, then loops back because due diligence must be repeated as the system or context changes.
With reduced motion: A static five-step horizontal diagram with numbered boxes and a visible return arrow from step 5 back to step 1. No animation; all labels, icons and example questions are shown at once.
Human rights due diligence is a structured way to check whether a project, product or service could negatively affect people. In AI, the focus is not only on whether a system works, but on who could be excluded, mislabelled, surveilled, silenced or treated unfairly. “Due diligence” here means ongoing care and checking, not a one-time legal form.
What goes in are the planned use of the system, the people and groups who may be affected, the data being used, the context, and the possible harms. For example, an image-matching tool might seem efficient, but it could fail more often for some document formats, languages or communities, or be used in a setting where mistakes have serious consequences.
What happens is a repeatable process: identify potential impacts on rights, speak with affected stakeholders where feasible and safe, assess severity and likelihood, change the design or use, add safeguards, assign responsibility, and track whether the measures work. “Safeguards” are practical protections such as limits on use, human review, appeal routes, privacy protections, accessibility checks and documented accountability.
What comes out is not a guarantee of safety. It is a reasoned record of risks, mitigation steps, remaining concerns, and who is responsible for monitoring and remedy. It is often confused with an impact assessment, but the difference is that human rights due diligence is broader and continuous: it should influence decisions before deployment, during use and when harms are reported.
How “Checking a photo-sorting tool before relying on it” maps onto the real thing
| In the story | In the AI system |
|---|---|
| testing many kinds of form photos | identifying who may be affected and where the system may fail |
| asking colleagues from several offices | consulting affected stakeholders or informed representatives |
| human review for uncertain matches | adding safeguards and human oversight |
| keeping an error log and changing use | tracking impacts and updating mitigation over time |
A worked example
Before authorising wider use, the manager of a public service centre reviews a pilot in which AI suggests a routing category for photographed benefit application forms. Caseworkers still make the routing decision.
Starting from: An approved test set covers the office's form layouts, languages, image quality and accessibility-related formats. Direct identifiers are removed where testing permits, and production images remain in the approved case-management environment.
- The manager scopes the use case, including where images come from, how the vendor processes them, who sees the suggested category and which decisions remain with caseworkers.
- The review identifies affected applicants and possible harms, including delayed assistance, unequal error patterns, loss of privacy and difficulty correcting a wrongly routed application. Severity and likelihood are assessed separately.
- Caseworkers and service users are consulted where feasible and safe. Testing examines performance across relevant languages, formats and image conditions rather than relying only on an overall accuracy measure.
- The office limits the tool to suggestions, sends uncertain or sensitive cases directly to manual review, shortens retention, provides a correction route and places data protection and incident-reporting duties on the vendor.
- An accountable officer tracks errors, complaints and delays against agreed stop criteria. Findings and corrective actions are recorded for the next review, rather than treating pilot approval as the end of the exercise.
Result: The pilot proceeds only within documented limits. The office can show which rights risks it found, what it changed, how affected people can seek correction and who must pause or alter the system if monitoring reveals harm.
In a United Nations workplace
At close of business, a knowledge management officer in a liaison office is reviewing an implementing partner's progress report. It says an AI tool now ranks community feedback for referral, but does not explain whose messages may be missed, whether affected people were consulted or how erroneous rankings are corrected.
Human rights due diligence turns those gaps into follow-up: identify affected rights-holders, assess the severity and likelihood of harm, examine language and accessibility barriers, consult safely where feasible, require controls from the partner, assign responsibility and keep monitoring after clearance.
Before you try thisCommunity feedback may contain personal, protection-sensitive or security-sensitive information. It must not be entered into an AI service that has not been approved for that material. Data must be minimised, access restricted and confidentiality preserved. Human reviewers remain responsible for referrals, corrections and any response to harm.
Where you may hear it
- An ICT governance board reviewing an AI use case before approval
- A procurement unit examining vendor safeguards before a solicitation is cleared
- A meeting with an implementing partner on corrective action in a progress report
- A front-office review of talking points on responsible technology
Avoid confusion
Avoid confusion
Often confused with
Impact assessment
A structured review used before or during AI use to identify who could be affected, what harms or rights risks may arise, and what controls or human decisions are needed.
AI ethics
AI ethics is the practice of checking whether AI systems are fair, safe, accountable and respectful of human rights before and during their use.
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.
Common misconceptions
Common misconception: If an AI system is accurate on average, human rights due diligence is no longer necessary.
Correction: Average performance can hide uneven impacts. A system may work well overall but fail more often for certain languages, disabilities, regions or document types. Human rights due diligence looks at who bears the errors, how serious the consequences are, and whether safeguards and remedy exist.
Common misconception: Human rights due diligence is just a legal tick-box done once before launch.
Correction: The idea is ongoing risk management and accountability. Conditions change: users change, data changes, deployment expands and new harms appear in practice. A one-time review may miss real-world effects, so monitoring, response and revision are part of the process.
Common misconception: It is the same thing as an impact assessment.
Correction: They overlap, but they are not identical. An impact assessment is usually a defined assessment exercise or document. Human rights due diligence is broader: it includes identifying impacts, acting on them, tracking results, communicating what is being done and enabling remedy over time.
Quick check
Which statement best describes human rights due diligence for an AI system?
The marked statement describes the full ongoing process. A bounded review is an impact assessment, reducing harm through design and testing describes AI safety, and checking broad principles describes AI ethics. Each can support due diligence but does not equal it.
Go deeper
Go deeper
The simple version
Before using AI, check how it could affect people’s rights, change the plan to avoid harm, keep watch during use, and make sure someone is clearly responsible.
The technical version
Human rights due diligence is an ongoing organisational process for identifying, preventing, mitigating and accounting for adverse human rights impacts linked to an organisation’s activities, products or services, including AI systems and related business relationships. In practice, this means scoping the use case and context; identifying potentially affected rights-holders and salient risks; consulting affected stakeholders where feasible and safe; assessing severity and likelihood; integrating findings into design, procurement, deployment and governance decisions; implementing mitigation measures such as use restrictions, privacy and security controls, accessibility measures, human oversight, contestability and escalation thresholds; tracking outcomes with disaggregated monitoring where appropriate; communicating relevant information; and providing or cooperating in remedy when harm occurs. It is often connected to international human rights frameworks, but operationally it functions as a continuous management and accountability process rather than a single technical test or compliance checklist.
Reviewed external resources
Primary sourceGuiding Principles on Business and Human Rights (opens in a new tab)UN Human Rights Office (OHCHR)
This official OHCHR resource gives the foundational framework for human rights due diligence and suits readers who want the primary reference behind the concept.
Intermediatelink checked 2026-07-31
Primary sourceRecommendation on the Ethics of Artificial Intelligence (opens in a new tab)UNESCO
This official UNESCO recommendation shows how AI governance connects to rights, safeguards and accountability and suits policy and leadership audiences.
Intermediatelink checked 2026-07-31
Primary sourceB-Tech Project (opens in a new tab)UN Human Rights Office (OHCHR)
This official OHCHR B-Tech Project resource is useful for readers who want business and technology-focused discussion of rights risks and practical governance questions.
Intermediatelink checked 2026-07-31
EncyclopaediaDue diligence (opens in a new tab)Wikipedia
This Wikipedia page gives a broad introductory overview of due diligence as a general idea and suits readers who want quick orientation before going deeper.
Intermediatelink checked 2026-07-31
VideoVideo explanations: Human rights due diligence (opens in a new tab)YouTube search
This YouTube search results page helps readers find different video explanations of human rights due diligence and suits people who prefer audiovisual introductions; it is a search page, not one authoritative video.
search results, not one pagelink checked 2026-07-31
DiscussionCommunity discussion: Human rights due diligence (opens in a new tab)Reddit search
This Reddit search results page helps readers see public discussion and common points of confusion about human rights due diligence and suits readers exploring informal debate; 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
Human rights due diligence is not a one-off checklist. Identify possible harms, act on them, track what happens and account for the response throughout the system's use.