Should an algorithm decide who gets a loan, who is invited to a job interview or which applicant gets a university place? When AI makes — or helps to make — decisions about people, abstract words like fairness and responsibility suddenly become very concrete. In this lesson you will learn the core principles of AI ethics and how the European Union's AI Act turns them into law.
Five principles of responsible AI
Governments, international organisations and companies have published many AI ethics guidelines. The wording differs, but most of them share the same core ideas:
| Principle | What it means | Question to ask |
|---|---|---|
| Fairness | No unjust discrimination; similar people are treated similarly | Does it work equally well for different groups? |
| Transparency | People know when AI is used and roughly how decisions are made | Do I know I am dealing with AI? Can the decision be explained? |
| Accountability | A person or organisation is responsible for the outcome | Who answers if something goes wrong? |
| Privacy | Personal data is protected and used lawfully | Which data is used, and with whose consent? |
| Safety and human oversight | Systems are tested, and a human can step in | Can a person check and overturn the decision? |
Opaque decision
Dear applicant,
your application has been rejected.
(no reason, no contact, no appeal)Transparent decision
Dear applicant,
your application has been rejected.
An automated system helped assess it.
Main factors: income history,
missing documents.
You can ask for a review by a staff
member within 30 days.The EU Artificial Intelligence Act
The European Union's AI Act — Regulation (EU) 2024/1689 — is often called the world's first comprehensive law on AI. It entered into force on 1 August 2024 and applies step by step. It matters beyond Europe too: it covers companies from any country whose AI systems are placed on the EU market or used in the EU. Its central idea is a risk-based approach: the higher the risk to people's health, safety and rights, the stricter the rules.
| Risk level | Examples | What the law says |
|---|---|---|
| Unacceptable | Social scoring; manipulating people or exploiting their vulnerabilities; emotion recognition at work and in schools (with narrow exceptions); untargeted scraping of facial images | Prohibited |
| High | AI for university admission or exam grading, CV screening, credit scoring, critical infrastructure, law enforcement, justice, migration; safety parts of products such as medical devices | Strict duties: risk management, good data, documentation, human oversight, accuracy, cybersecurity |
| Limited (transparency) | Chatbots, deepfakes, AI-generated content | People must be told they are dealing with AI; deepfakes must be labelled |
| Minimal | Spam filters, video games | No new obligations |
- 1 August 2024
Entry into force
The law becomes part of EU legislation; the obligations start later.
- 2 February 2025
Bans and AI literacy
The bans start to apply; organisations must ensure that their staff have sufficient AI literacy.
- 2 August 2025
General-purpose AI models
Rules for providers of general-purpose AI models, such as large language models, start to apply.
- 2026–2027
Remaining rules
Most other obligations, including those for high-risk systems, are phased in. Check the current official timetable, as details can be adjusted.
Breaking the bans can cost a company up to €35 million or 7% of its worldwide annual turnover, whichever is higher.
Beyond the EU
- OECD AI Principles (2019): the first intergovernmental standard on AI, promoting trustworthy AI that respects human rights and democratic values.
- UNESCO Recommendation on the Ethics of Artificial Intelligence (November 2021): adopted by all 193 UNESCO member states.
- Council of Europe Framework Convention on AI (opened for signature in September 2024): the first legally binding international treaty on AI and human rights, democracy and the rule of law.
- World AI Cooperation Organization (WAICO) (founding agreement signed in Shanghai on 16 July 2026 by 29 countries): an intergovernmental organization proposed by China and headquartered in Shanghai, for international cooperation on AI development and governance, with a focus on the countries of the Global South.
Practice: classify the systems
Which risk level?
Problem
Use the AI Act's risk levels: a) a chatbot answering questions on a bank's website; b) a system that scores applications for university admission; c) an app that ranks citizens by their social behaviour to decide who gets public benefits; d) the spam filter in your mailbox.
Show solutionHide solution
Solution
b) High risk — education and access to it: risk management, quality data, documentation and human oversight are required.
c) Unacceptable — social scoring is prohibited.
d) Minimal risk — no new obligations.
Note: in real cases the classification depends on the exact purpose and context, which the law describes in detail.
Interactive
Remember
- The core principles of responsible AI: fairness, transparency, accountability, privacy, safety and human oversight.
- The EU AI Act entered into force on 1 August 2024 and uses a risk-based approach.
- Four levels: unacceptable (banned), high (strict duties), limited (transparency), minimal (no new duties).
- The bans apply from 2 February 2025; the rules for general-purpose models from 2 August 2025.
- The OECD (2019), UNESCO (2021) and the Council of Europe (2024) have also adopted international AI frameworks.