Position of this article in the original text
This article is the English translation of Chapter 1 of the "Ethical Guidelines for Trustworthy AI" 2019 edition, corresponding to printed pages 9–13. Previous section Overview and Overall Structure Explain the three components: "legitimate, ethical, and sound"; this article discusses why a starting point of fundamental rights is necessary, and how four ethical principles can assist in evaluating the development and use of AI.
The Foundation of Trustworthy AI
This chapter is based on fundamental rights, embodied by four ethical principles that reflect these rights. To ensure AI is ethical and robust, these principles should be followed. Here, we primarily draw from ethics.
AI ethics is a branch of applied ethics that examines ethical issues arising from the development, deployment, and use of AI. Its central concern is to identify how AI either promotes or undermines individuals' flourishing lives: encompassing both quality of life and personal autonomy and freedom essential in democratic societies.
Ethical reflection has at least two uses. On one hand, it prompts us to consider how to protect individuals and groups at the most fundamental level; on the other hand, it can inspire new innovation directions, enabling technology to actively promote ethical values, such as supporting the United Nations Sustainable Development Goals and the forthcoming European Union 2030 agenda.
Although this guide primarily focuses on protective functions, the role of ethics in shaping innovation directions should not be undervalued. Trustworthy AI can enhance individual development and collective well-being by creating prosperity, value, and wealth, and can also contribute to a more equitable society by promoting health, improving economic, social, and political opportunity distribution.
Therefore, it is essential to understand how to support the development, deployment, and use of AI, enabling everyone to fully develop in a world where AI is widely applied, while building a better future that maintains global competitiveness. Like other powerful technologies, the societal applications of AI present ethical challenges, such as its impact on people and society, decision-making capabilities, and safety. As people increasingly rely on AI assistance or delegate decisions to AI, we must ensure that AI has a fair impact on people's lives, respects non-negotiable values, and includes appropriate accountability mechanisms to meet these requirements.
Europe needs clarity on what kind of AI future it envisions and which AI systems it should research, develop, deploy, and use to achieve that vision. The guide positions trustworthy AI as the pathway to building this future: grounded in democracy, the rule of law, and fundamental rights, ensuring that systems continuously improve and uphold democratic culture, while also fostering innovation and responsible competition.
Ethical guidelines in any field, no matter how consistent, mature, and detailed, cannot replace ethical reasoning itself. Actual judgments must always take into account contextual details that general guidelines cannot possibly cover. Beyond rules, ethical culture and ways of thinking must be developed and sustained through public discussion, education, and practice.
Basic rights: legal rights, also moral rights
The guide adopts an AI ethics approach based on EU treaties, the EU Charter of Fundamental Rights, and international human rights law. Respecting fundamental rights within a democratic and rule-of-law framework provides a foundation for identifying ethical principles and values that can be implemented in AI.
The EU Treaty and Charter establish a range of fundamental rights, which EU institutions and member states are obliged to respect when implementing EU law. The Charter expresses these rights in terms of dignity, freedom, equality, solidarity, civic rights, and judicial guarantees. Its common foundation is the respect for human dignity, reflecting a human-centred approach: humans possess a unique, non-derogable, and morally superior status that should be given priority in the civic, political, economic, and social spheres.
This human-centered approach does not mean understanding people as isolated individuals detached from society. Individual freedom and human dignity must be grounded in a shared social and constitutional foundation that makes them realistically achievable and meaningful.
Charter rights have legal binding force, but fundamental rights do not provide complete legal protection in every context. For example, the scope of application of the Charter is limited to the field of EU law; international human rights law, particularly the European Convention on Human Rights, imposes binding obligations on member states even outside the domain of EU law. At the same time, fundamental rights also belong to individuals on the basis of personal moral status and to a certain extent to groups, regardless of whether legal enforceability is granted.
As a legally enforceable right, fundamental rights constitute the first component of trustworthy AI (legality); as rights grounded in human moral dignity and universally held by all, they also support the second component (ethics). The guide does not provide guidance on compliance with the first aspect, and thus, when referring to fundamental rights below, the emphasis is primarily on the ethical meaning.
What rights are particularly related to AI?
Human rights in international human rights law, the EU treaties, and the Charter are indivisible. The following groups of rights are particularly relevant for understanding AI. Many rights can be enforced under specific circumstances and must be adhered to; even when legal requirements are met, ethical reflection can help assess the impact of technology on these rights and their underlying values, further distinguishing "what is technically possible now" from "what ought to be done."
Respect for human dignity
Every individual has intrinsic value, and no other person or AI or other new technologies should diminish, harm, or suppress this value. People should be respected as moral agents with inherent worth, rather than merely as objects to be screened, categorized, rated, expelled, shaped, or manipulated.
Therefore, the development of AI should respect, serve, and protect human psychological and physical integrity, individual and cultural identity, and assist in meeting basic needs.
Personal freedom
People should have the freedom to make decisions about their own lives. This means both being free from inappropriate interference by public power, and also that governments and non-governmental organizations should take action to ensure that individuals at risk of exclusion can equally access the benefits and opportunities of AI.
In AI contexts, it is necessary to reduce direct or indirect illegal coercion, threats to psychological autonomy and mental well-being, improper surveillance, deception, and unfair manipulation. Personal freedom also implies enabling individuals to have greater control over their lives, including freedom in entrepreneurship, freedom in artistic and scientific pursuits, freedom of expression, rights to private life and privacy, and the freedoms of assembly and association.
Democracy, justice, and the rule of law
In constitutional democracy, all governmental powers must be authorized by law and be subject to legal limits. AI should uphold and promote democratic processes, respect diverse values and lifestyle choices. It must not undermine democratic procedures, public deliberation, or voting systems.
AI should also commit to not undermining the fundamental principles of the rule of law and enforceable laws and regulations, and to ensuring due process and the equality before the law.
Equality, non-discrimination, and unity
Must equally respect the moral value and dignity of every person. This goes beyond "non-discrimination"; non-discrimination still allows differentiation between inherently different situations based on objective reasons.
In AI, equality means that the system's outputs should not produce outputs with unfair biases. For example, training data should be as inclusive as possible, reflecting diverse population groups. At the same time, appropriate respect should be shown to individuals and groups who may be vulnerable or at risk of exclusion, including workers, women, people with disabilities, minority ethnic groups, children, and consumers.
Civil rights
Citizens have various rights, including the right to vote, to receive effective administrative management, to access public documents, and to make petitions to administrative agencies. AI can significantly enhance the scale and efficiency of public goods and services, but it may also undermine these rights, thus requiring protection.
Here, the reference to "civil rights" does not deny or overlook the rights of nationals from third countries, individuals without formal status, or those who are illegally residing. They too enjoy rights protected by international law, which also applies to AI scenarios.
From Rights to Four Ethical Principles
Public, private, and civil society organizations have proposed numerous AI ethics frameworks based on fundamental rights. The European Group on Ethics in Science and New Technologies presented nine principles grounded in the values of the EU treaties and Charter. This guide builds upon those principles, incorporating most of the principles proposed by other organizations, and clarifies the objectives each principle aims to promote.
Based on fundamental rights, we can draw on Europe's decades of experience in rights protection to reduce regulatory uncertainty and enhance clarity and predictability. These principles can inspire new regulatory tools that help interpret fundamental rights in the context of evolving social and technological environments, and provide justification for AI development, deployment, and use, while remaining adaptable to societal evolution.
AI should enhance individual and collective well-being. The following four principles are rooted in fundamental rights and constitute ethical obligations that practitioners should always strive to uphold. The ordering reflects the sequence in which corresponding rights appear in the Charter and does not imply a hierarchy among the principles.
One, respect for human autonomy
The EU's fundamental rights aim to safeguard human freedom and autonomy. Individuals interacting with AI should be able to exercise full and effective self-determination and participate in democratic processes.
AI should not place humans in a subordinate position, coerce, deceive, manipulate, shape, or drive them without legitimate justification. Design should enhance, supplement, and empower human cognitive, social, and cultural capabilities. The functional allocation between humans and AI should follow human-centered design, ensuring humans retain meaningful choices. This requires human oversight in AI workflows.
AI might fundamentally transform work domains. It should support people in work environments and aim to create meaningful work.
II. Preventing Injury
AI should not cause, exacerbate, or otherwise produce adverse effects on individuals. This requires protecting human dignity, psychological, and physical integrity. Harm can be directed at individuals or groups, and also includes intangible damages within social, cultural, and political environments, such as cultural harm to the lifestyle of individuals or groups.
The system and its operating environment should be secure and equipped with security protections; the system should be technically robust to prevent malicious use. During development, deployment, and usage, special attention should be given to vulnerable populations, and they should be involved.
Particular attention must also be paid to situations where harm is caused or exacerbated due to asymmetries in power or information, such as in employer-employee, business-consumer, and government-resident relationships. Preventing harm also includes consideration for the natural environment and all forms of life.
Three, Fairness
Developing, deploying, and using AI must be fair. Fairness has multiple interpretations, but at least includes substantive and procedural dimensions.
At the substantive level, interests and costs should be distributed equally and fairly, and individuals and groups should be protected from unfair bias, discrimination, and stigmatization. After eliminating unfair bias, AI has the potential to enhance social equity. It is also important to promote equal access to education, goods, services, and technologies for all. AI should not deceive people or unjustly undermine their freedom of choice.
Fairness also requires that means be proportional to the ends, and that competing interests and objectives be carefully weighed. The original text illustrates the principle of proportionality with the example: "Do not use a sledgehammer to crack a nut." The measures taken to achieve a purpose should be strictly necessary and limited to the minimum required; if multiple measures can achieve the same goal, the one causing the least harm to fundamental rights and ethical principles should be chosen. The text suggests prioritizing public sector data when available and minimizing the use of personal data, while also balancing considerations of corporate intellectual property, confidentiality rights, and user rights.
At the program level, individuals (both those using AI and those who operate it) should be able to question decisions made by AI and seek effective remedies. For this to happen, there must be identifiable accountable parties, and the decision-making processes must be explainable. In workplace settings, the right to association and the right to join trade unions can also support this process.
IV. Interpretability
Explainability is crucial for building and maintaining user trust: the process should be transparent; the system's capabilities and intentions should be clearly communicated; decisions should be explained to those directly and indirectly affected. Without this information, decisions cannot be properly questioned.
But at times, it's not possible to explain why a model produces a specific output, or how particular input factors combine to influence a decision. The original text refers to such models as "black box" algorithms, which warrant special attention. Under overall respect for fundamental rights, alternative explainability measures might be needed, such as traceability, auditability, and transparently disclosing the system's capabilities.
The degree of explanation required is highly dependent on context and the severity of consequences from erroneous outputs. The ethical risks associated with inaccurate shopping recommendations are typically different from those involved in systems determining whether convicted individuals can receive parole.
Note: Here, explicability encompasses conditions for transparent communication, explanation, traceability, and accountability, and cannot be simply equated with "the ability to display all internal parameters of the model."
What to do when principles conflict?
Tensions may arise among principles, and there is no fixed, universally applicable answer. An accountable deliberative method should be established in accordance with commitments to democratic participation, due process, and open political engagement.
For example, preventing harm may conflict with human autonomy: predictive policing might help reduce crime, but the associated surveillance could intrude upon personal freedom and privacy. The overall benefits of the system should significantly outweigh foreseeable individual risks, but abstract principles alone cannot directly provide all the right answers. Practitioners should address ethical dilemmas and trade-offs through reasoned, evidence-based reflection, rather than intuition or arbitrary judgment.
However, it is also possible that no ethically acceptable compromise exists. Some fundamental rights and their corresponding principles have an absolute character, such as human dignity, and cannot be treated as exchangeable items in general interest balancing.
The core guidance of this chapter is: follow four principles in developing, deploying, and using AI, acknowledging and addressing potential conflicts among them; pay special attention to children, people with disabilities, historically marginalized or at-risk of exclusion groups, as well as power or information asymmetries between employers and employees, businesses and consumers.
Read next Seven Requirements and Implementation Methods See how these principles enter data, design, organization, and oversight processes.
Reference and Extended Reading
"Original text: Ethics Guidelines for Trustworthy AI, April 8, 2019, European Union, printed pages 9–13." Original entry for EU Digital Strategy . This page is an unofficial English translation, with adjusted sections, terminology, and layout; substantive explanations from the original footnotes have been integrated into the corresponding paragraphs, and personal identity lists are not reproduced.
"yi" EU Digital Data Reuse Guidelines , the original document uses " CC BY 4.0 This page's translation is provided under the same license and does not imply endorsement by the original organization.
Clause inquiry: Charter of Fundamental Rights of the European Union ; Official Chinese version of the European Convention on Human Rights entry point The latter entry allows selection of a Chinese version. Editor addition: UN Human Rights Chinese Entry Point For querying the international human rights framework, does not constitute a determination of specific legal applicability on this page.