UN Ethical AI Principles for the UN System

Overview

The Principles for the Ethical Use of Artificial Intelligence in the United Nations System represent a landmark effort to harmonize the internal adoption and governance of AI technologies across the diverse entities of the United Nations. Endorsed in September 2022 by the United Nations System Chief Executives Board for Coordination (CEB), these principles were developed under the leadership of the High-Level Committee on Programmes (HLCP) and its Inter-Agency Working Group on Artificial Intelligence (IAWG-AI), co-led by UNESCO and the ITU. The document serves as a normative compass for UN agencies, funds, and programs, ensuring that as the UN integrates AI into its operations—from humanitarian aid and peacekeeping to sustainable development monitoring—it does so in a manner that is consistent with the UN Charter, international human rights law, and the 2030 Agenda for Sustainable Development. The significance of this instrument lies in its role as the first system-wide ethical framework for AI. While individual agencies had previously explored AI ethics, the 2022 Principles provide a unified set of ten core tenets that apply to the entire AI lifecycle, including research, design, development, procurement, deployment, and eventual decommissioning. This collective approach is intended to prevent fragmented standards across the UN system and to model responsible AI leadership for Member States. By grounding these principles in the UNESCO Recommendation on the Ethics of Artificial Intelligence (2021), the UN system ensures that its internal practices reflect the global consensus on AI ethics while addressing the specific operational needs and mandates of international civil servants and multilateral organizations.

Definitions and Conceptual Framework

The document provides a functional understanding of 'Artificial Intelligence systems' as technological constructs capable of automatically processing data and information in ways that resemble intelligent behavior. This definition is intentionally broad and technology-neutral to remain relevant as the field evolves. It encompasses various methodologies, including machine learning, deep learning, and automated reasoning. Crucially, the principles define the 'AI system lifecycle' as a comprehensive process spanning from initial research and design to procurement, deployment, use, maintenance, and termination. This lifecycle approach ensures that ethical considerations are not merely an afterthought but are integrated into the very architecture and acquisition processes of UN entities. Furthermore, the document clarifies the concept of 'ethical use' within the UN context. Ethics is defined not as a static set of rules but as a dynamic commitment to the protection and promotion of human rights, human dignity, and the flourishing of societies and the environment. The principles also define 'accountability' and 'transparency' in specific operational terms, emphasizing that accountability must always reside with human beings or legal entities, and that transparency involves both technical explainability (the ability to trace a decision) and procedural disclosure (informing individuals when they are interacting with or being impacted by an AI system).

Governance and Institutional Oversight

The governance of these principles is situated within the highest levels of the United Nations administrative structure. The Chief Executives Board for Coordination (CEB), chaired by the Secretary-General, serves as the ultimate endorsing body, providing the political and administrative weight necessary for system-wide adoption. Beneath the CEB, the High-Level Committee on Programmes (HLCP) is responsible for the policy development and inter-agency coordination. The HLCP Inter-Agency Working Group on Artificial Intelligence (IAWG-AI), co-chaired by UNESCO and the International Telecommunication Union (ITU), acted as the technical drafting body, bringing together expertise from across the UN system to ensure the principles were both robust and practical. Institutional implementation is decentralized, meaning that each UN entity is responsible for integrating these principles into its own specific mandates and operational guidelines. However, the HLCP provides a centralized platform for sharing best practices, developing common assessment tools, and ensuring that the 'UN System-wide Strategic Approach and Road Map for Supporting Capacity Development on AI' is aligned with these ethical tenets. This framework creates a multi-layered governance structure where high-level normative guidance from the CEB is translated into specific agency actions, supported by technical working groups that monitor technological trends and provide updated guidance on emerging risks like generative AI.

Core Ethical Principles and Provisions

The core of the document consists of ten principles designed to guide the ethical lifecycle of AI systems. The first principle, 'Do no harm,' mandates that AI systems must not cause or exacerbate individual or collective harm, including social, cultural, or environmental damage. This is closely linked to 'Defined purpose, necessity, and proportionality,' which requires UN entities to ensure that AI use is justified by a legitimate aim and that the data processed is limited to what is strictly necessary. 'Safety and security' provisions demand that AI systems be resilient against attacks and operate reliably to prevent unintended physical or mental harm to individuals. The principles of 'Fairness and non-discrimination' and 'Sustainability' address broader societal impacts. UN entities are urged to actively prevent bias and stigmatization in AI models and to ensure that the benefits of AI are distributed equitably. Sustainability requires an assessment of the environmental and economic costs of AI, ensuring that the energy-intensive nature of large-scale computing does not undermine the UN's climate goals. Other key provisions include 'Right to privacy and data protection,' 'Human autonomy and oversight,' and 'Transparency and explainability,' which collectively ensure that AI remains a tool for human empowerment rather than a replacement for human judgment. The final principles, 'Responsibility and accountability' and 'Inclusion and participation,' emphasize that humans must remain legally and ethically responsible for AI-driven outcomes and that stakeholders, particularly vulnerable communities, must be consulted during the AI development process. These principles are intended to be applied holistically, recognizing that they are interdependent and mutually reinforcing in the context of international development and humanitarian action.

Scope of Application and Procurement

The scope of the Principles is comprehensive, applying to all United Nations system organizations, including the Secretariat, specialized agencies, and related funds and programs. Geographically, the application is global, as it governs how the UN interacts with data and populations in every country where it operates. The principles are intended to apply to all stages of the AI system lifecycle, meaning they govern not only the software developed in-house by the UN but also the AI tools and services procured from third-party vendors. This procurement aspect is vital, as it forces private sector partners to align with UN ethical standards if they wish to provide services to the organization. In terms of application, the document is designed to be read in conjunction with other foundational UN policies. These include the UN Charter, the Universal Declaration of Human Rights, the UN Personal Data Protection and Privacy Principles, and the Human Rights Due Diligence Policy. By positioning the AI principles within this broader legal and normative ecosystem, the UN ensures that AI governance is not treated as a siloed technical issue but as a core component of the organization’s commitment to international law and human dignity. The principles apply both to internal administrative uses of AI (such as HR or finance) and to programmatic uses (such as using satellite imagery and AI for disaster response or agricultural monitoring). This broad scope ensures that the UN's digital transformation is consistent across all its diverse mandates, from peace and security to human rights and development.

Implementation and Operationalization

Implementation of the Principles is facilitated through a series of practical measures coordinated by the HLCP. Rather than imposing a one-size-fits-all regulation, the UN uses a 'capacity-building and guidance' model. This involves the development of self-assessment tools that agencies can use to evaluate their AI projects against the ten principles. The HLCP is tasked with creating 'practical guidance' on how to operationalize the principles, which includes templates for ethical impact assessments and checklists for AI procurement. These tools help translate high-level ethics into the daily workflows of project managers and data scientists across the UN system. Furthermore, the implementation framework emphasizes inter-agency cooperation. By leveraging the expertise of UNESCO on ethics and the ITU on technical standards, the UN system fosters a culture of shared learning. Pilot projects are encouraged to test the operationalization of the principles in real-world scenarios, such as the use of AI in humanitarian logistics or public health surveillance. The framework also highlights the importance of 'AI literacy' for the UN workforce, ensuring that staff at all levels understand the ethical implications of the technologies they deploy. This human-centric implementation strategy aims to build an internal ecosystem where ethical considerations are as fundamental to AI projects as technical performance. The UN also seeks to engage with external experts and civil society to ensure that its implementation strategies remain robust and inclusive of diverse perspectives.

Monitoring, Evaluation, and Accountability

Monitoring the adherence to these principles is an ongoing process overseen by the HLCP and reported to the CEB. While the principles are not legally binding in the same way as a treaty, they carry significant administrative and political weight within the UN system. Monitoring mechanisms include periodic reviews of agency-specific AI policies to ensure they remain aligned with the system-wide principles. The HLCP Inter-Agency Working Group on AI serves as a forum for agencies to report on their progress, share challenges, and update the collective understanding of AI risks based on new technological developments. Evaluation is also integrated into the 'UN System-wide Strategic Approach and Road Map for Supporting Capacity Development on AI.' This roadmap includes specific milestones for the adoption of ethical guidelines and the implementation of impact assessments. The document encourages the use of 'audit and due diligence mechanisms' to ensure that AI systems are functioning as intended and are not producing discriminatory outcomes. By fostering a culture of transparency and reporting, the UN aims to create a feedback loop where the lessons learned from monitoring lead to the continuous refinement of the principles and their application, ensuring the organization remains agile in the face of rapid AI advancement. Accountability is further reinforced through the requirement that UN entities establish clear lines of responsibility for AI systems, ensuring that any negative impacts can be addressed and remediated through existing administrative and legal channels.

Relationship to International Legal Instruments

The 2022 Principles are explicitly built upon the foundation of the UNESCO Recommendation on the Ethics of Artificial Intelligence, which was adopted by 193 Member States in 2021. While the UNESCO Recommendation provides a global normative framework for Member States, the 2022 Principles adapt that framework for the specific internal operational context of the UN system. This ensures that the UN 'practices what it preaches' by holding itself to the same high standards it recommends to its Member States. The document also references the UN Personal Data Protection and Privacy Principles (2018), which provide the essential data governance foundations upon which ethical AI must be built. Additionally, the principles are designed to complement the Secretary-General’s 'Our Common Agenda' and the 'Global Digital Compact.' These broader strategic initiatives seek to establish a shared vision for a digital future that is open, free, and secure for all. By providing a concrete ethical framework for AI, the 2022 Principles serve as a technical and normative pillar for these higher-level political goals. They also align with the 'Human Rights Due Diligence Policy,' ensuring that the UN’s use of technology does not inadvertently support activities that violate international human rights standards, particularly in sensitive areas like security and border management. This alignment ensures that the UN's internal AI governance is a coherent part of its broader mission to uphold the rule of law and protect fundamental freedoms in the digital age.

International Alignment and Global Leadership

The UN Ethical AI Principles are a key component of the broader international effort to create 'interoperable' AI governance. By aligning its internal standards with the UNESCO Recommendation, the UN contributes to a global consensus on AI ethics that spans across different legal systems and cultural contexts. This alignment is crucial for fostering international cooperation, as it provides a common language and set of values for UN agencies to engage with governments, private sector partners, and civil society. The principles reflect the growing global emphasis on 'human-centric AI,' a concept also championed by the OECD AI Principles and the G20 AI Guidelines. Moreover, the principles support the UN's role as a global convener on AI governance. They provide the normative basis for the Secretary-General's High-Level Advisory Body on AI, which was established in 2023 to offer recommendations on the international governance of AI. By demonstrating a robust internal ethical framework, the UN strengthens its credibility as a neutral and principled actor capable of leading global discussions on AI safety, equity, and accountability. This international alignment ensures that the UN’s internal digital transformation is not an isolated effort but is deeply integrated into the global movement toward responsible and trustworthy artificial intelligence. Through these principles, the UN system aims to lead by example, showing how large, complex organizations can successfully integrate ethical considerations into the deployment of cutting-edge technologies for the public good.

Key Focus Areas

The Principles for the Ethical Use of Artificial Intelligence in the United Nations System establish ten core areas of focus applicable across all stages of an artificial intelligence system's life cycle. The first principle, 'Do no harm,' requires that artificial intelligence systems must not cause or exacerbate individual or collective harm to social, cultural, economic, natural, or political environments. 'Defined purpose, necessity and proportionality' mandates that the use of technology must be context-appropriate, necessary, and strictly proportionate to legitimate organizational mandates. Under 'Safety and security,' organizations must identify, address, and mitigate safety risks to protect humans and ecosystems. To foster equity, 'Fairness and non-discrimination' mandates equal distribution of benefits and risks while preventing bias, deception, and stigmatization. 'Sustainability' demands continuous evaluation of social, economic, and environmental impacts to protect current and future generations. Data governance is governed by 'Right to privacy, data protection and data governance,' ensuring subject rights and data integrity under established United Nations frameworks. 'Human autonomy and oversight' preserves human freedom and meaningful decision-making, explicitly prohibiting automated systems from making unmediated life-or-death decisions. 'Transparency and explainability' guarantees that decision-making processes and algorithmic logic are understandable to affected individuals. 'Responsibility and accountability' mandates audit mechanisms, whistleblower protections, and legal responsibility for artificial intelligence impacts. Finally, 'Inclusion and participation' requires participatory consultation, gender equality, and interdisciplinary stakeholder engagement throughout the technology life cycle.

Implementation Timeline

Adoption and Endorsement

Sources and References