Glossary
Responsible AI
Written with AI from the sources below and checked against our facts sheet. Last checked . How this site is made · Spot an error?
Part of: AIGP certification and exam
Responsible AI is the practice of designing, building and using AI so it is fair, safe, secure, private, transparent and accountable. It turns shared principles, such as the OECD AI Principles and NIST's trustworthy-AI characteristics, into concrete policies and controls. It applies across the whole AI life cycle, from design to retirement.
In plain English
Responsible AI is to AI what food safety is to a restaurant. Diners never see the kitchen rules, but those rules decide whether the meal is safe. Responsible AI is the set of values and practices that decide whether an AI system treats people well.
The values are widely shared. Our lessons group them into six pillars: fairness; safety and reliability; privacy and security; transparency and explainability; accountability; and human-centricity. The OECD AI Principles and the NIST AI RMF use different labels for much the same ideas.
Values alone change nothing. Responsible AI becomes real through AI governance: policies, roles, reviews, testing and monitoring.
Example
Wrenfield Insurance, a fictional insurer, plans an AI tool that triages home-insurance claims. Its responsible AI policy turns each principle into a launch gate:
| Principle | Control at Wrenfield |
|---|---|
| Fairness | Error rates compared across regions and customer groups before launch |
| Safety and reliability | Stress tests on unusual claims; the tool fails safe to a human adjuster |
| Privacy and security | Only the claim data the model needs; hardening against data poisoning |
| Transparency and explainability | Customers told AI is used; adjusters see the main factors behind each score |
| Accountability | A named claims director owns the tool and can suspend it |
| Human-centricity | Adjusters can override any score; the tool never denies a claim on its own |
A vendor later offers a cheaper model with no test results by group. The claims director declines it. Cost matters, but it is not a reason to skip the fairness gate.
Why it matters
- NIST. The AI RMF calls AI risk management a key component of responsible development and use. It says core concepts in responsible AI emphasize human centricity, social responsibility and sustainability.
- OECD. Its five values-based principles, adopted in 2019 and updated in May 2024, are the most widely cited international baseline.
- EU AI Act. The Act does not define responsible AI. Its stated purpose is to promote "human-centric and trustworthy" AI (Art. 1(1)). It turns several principles into duties for high-risk systems, such as human oversight (Art. 14).
- ISO/IEC 42001. The management-system standard gives organizations a certifiable way to run a responsible AI program (see the ISO/IEC 42001 hub).
How the AIGP exam tests it
Responsible AI principles open Domain I of the Body of Knowledge (lesson I.A.4). Expect two kinds of question.
- Name the principle. A scenario describes a failure, and you pick the principle most directly at stake. A model that misses disease more often in one group is a fairness problem. Customers not told a chatbot is AI is a transparency problem.
- Spot the impostor. Profit, speed, efficiency and market share are never responsible AI principles, however an option dresses them up.
Common traps:
- Accountability by contract. "The vendor is responsible" is wrong. The organization using the AI still answers for its outcomes.
- Transparency vs explainability. A notice that AI is in use is transparency. Telling one person why her application failed is explainability. See the comparison.
- After-the-fact review. A human who looks only after complaints arrive provides remediation, not meaningful human control.
Frequently asked questions
What are the principles of responsible AI?
Most frameworks land on the same core: fairness, safety and reliability, privacy and security, transparency and explainability, accountability, and human-centricity. The OECD lists five values-based principles. NIST lists seven trustworthy-AI characteristics. The labels differ, but the values overlap heavily, so you can map one framework onto another.
What is the difference between responsible AI and AI governance?
Responsible AI describes the goal: AI that is fair, safe, transparent and accountable. AI governance is the machinery that delivers it: policies, roles, committees, risk assessments, testing and monitoring. Principles without governance stay on paper. Governance without principles has nothing to aim at.
What is the difference between responsible AI and trustworthy AI?
They overlap almost entirely. Trustworthy AI usually describes qualities of the system, such as NIST's seven characteristics or the EU's 2019 ethics guidelines. Responsible AI usually describes how an organization behaves when it designs, builds and uses AI. A responsible AI program aims to produce trustworthy systems.
Is responsible AI a legal requirement?
Not by that name. But laws turn many of its principles into duties. The EU AI Act requires risk management, data governance, transparency and human oversight for high-risk systems. The GDPR restricts solely automated decisions. US laws on discrimination and unfair practices apply to AI too. Frameworks like the NIST AI RMF remain voluntary.