Serbia - Ethical AI Guidelines

Overview

The Ethical Guidelines for the Development, Implementation and Use of Reliable and Responsible AI were adopted by the Government of Serbia on March 23, 2023. This comprehensive ethical framework establishes principles and conditions for developing trustworthy AI systems aligned with European and international standards. The guidelines were developed under the coordination of the Office for Artificial Intelligence and draw significantly from UNESCO's Recommendation on the Ethics of AI adopted in 2021. Though not legally binding, they represent Serbia's official position on AI ethics and inform ongoing regulatory development.

Definitions

The guidelines define reliable AI (often termed trustworthy AI) as systems that meet ethical, legal, and technical requirements throughout their lifecycle. Responsible AI refers to AI developed and deployed with clear accountability mechanisms. High-risk AI systems are defined as those with significant potential to impact individuals' fundamental rights, health, safety, or critical societal functions. The document distinguishes between AI developers (creating AI systems) and AI users (deploying and operating them), establishing responsibilities for each. Key technical terms including machine learning, automated decision-making, and algorithmic systems are defined consistent with international frameworks.

Governance and Institutional Framework

The Office for Artificial Intelligence serves as the primary government body overseeing ethical AI implementation. The guidelines establish a collaborative governance approach involving the Ministry of Science, Technological Development and Innovation, the Institute for Artificial Intelligence Research and Development, and relevant sector-specific regulators. Academic institutions and civil society organizations participate in ongoing guideline development and review. The governance framework emphasizes multi-stakeholder engagement, bringing together government, industry, academia, and public representatives in AI policy discussions. Sector-specific guidance may be developed by relevant ministries within the ethical framework established by these guidelines.

Key Focus Areas

• Human Agency and Oversight: Ensuring humans maintain meaningful control over AI systems through intervention mechanisms and supervision protocols.
• Technical Reliability: Requiring AI systems to perform reliably, securely, and as intended across their operational lifecycle.
• Safety Assurance: Incorporating risk mitigation mechanisms to prevent harm from AI system failures or misuse.
• Privacy Protection: Safeguarding personal data through privacy-by-design principles and GDPR-aligned protections.
• Data Governance: Establishing standards for data quality, integrity, and appropriate use in AI training and operation.
• Transparency: Requiring explainability of AI decisions and disclosure of AI system use to affected individuals.
• Non-Discrimination: Preventing algorithmic bias and ensuring fair treatment across demographic groups.
• Diversity and Inclusion: Promoting inclusive AI development considering diverse perspectives and needs.
• Environmental Sustainability: Encouraging consideration of AI systems' environmental impact and resource efficiency.
• Social Well-being: Ensuring AI development serves broader societal benefit beyond narrow commercial interests.
• Accountability: Establishing clear responsibility chains for AI system outcomes and decisions.
• High-Risk System Identification: Providing criteria for identifying AI systems requiring enhanced scrutiny and safeguards.

Implementation Framework

Implementation follows a self-regulatory approach with government guidance and support. The guidelines include a comprehensive self-assessment questionnaire enabling developers and users to evaluate AI systems against established criteria. This questionnaire covers all seven core conditions and helps identify areas requiring improvement. Organizations are encouraged to conduct assessments at key development stages: design, pre-deployment, and operation. For high-risk AI systems, more rigorous assessment processes are recommended. The Office for Artificial Intelligence provides supplementary guidance and may develop sector-specific implementation tools. Training and awareness programs support capacity building across public and private sectors. The framework anticipates evolution toward binding requirements as EU AI Act harmonization progresses.

Monitoring and Evaluation

Organizations implementing AI systems should establish internal monitoring processes aligned with guideline requirements. Self-assessments using the provided questionnaire serve as the primary evaluation mechanism. The Office for Artificial Intelligence collects aggregated data on guideline adoption and effectiveness through voluntary reporting. Sector-specific monitoring may be conducted by relevant ministries and regulatory bodies. Academic institutions and research organizations contribute to evaluating AI governance effectiveness through studies and reports. The guidelines recommend establishing feedback mechanisms for individuals affected by AI systems. Regular review and updating of the guidelines is anticipated as technology evolves and international standards develop.

Penalties, Liability, and Appeals

As voluntary ethical guidelines, no direct penalties apply for non-compliance. However, non-compliance may affect eligibility for government AI funding, partnerships, and procurement opportunities. Existing legal frameworks apply to harmful AI outcomes, including civil liability for damages, consumer protection violations, and data protection penalties under Serbia's GDPR-aligned law. The guidelines establish moral and reputational expectations that may influence stakeholder trust and business relationships. For high-risk AI systems, additional sector-specific regulations may impose binding requirements with associated penalties. Appeals and redress for individuals harmed by AI systems follow standard legal channels, with guidelines recommending accessible complaint mechanisms.

Relationship to Other Instruments

The guidelines complement the Strategy for AI Development 2020-2025 by providing ethical implementation guidance. They align closely with UNESCO's Recommendation on AI Ethics, incorporating its principles and approaches. The framework anticipates harmonization with the EU AI Act as part of Serbia's accession process. Serbia's Personal Data Protection Law (GDPR-aligned) provides the legal basis for privacy-related provisions. Consumer protection and anti-discrimination laws provide enforcement mechanisms for some guideline principles. The guidelines inform development of potential future AI-specific legislation currently under consideration.

International Alignment

The guidelines demonstrate strong alignment with international AI ethics frameworks. They incorporate principles from UNESCO's Recommendation on AI Ethics, with Serbian representatives having participated in its development. The seven core conditions align with the European Commission's Ethics Guidelines for Trustworthy AI. Serbia's participation in the Global Partnership on Artificial Intelligence (GPAI) reinforces international cooperation. OECD AI Principles on human-centered AI development are reflected throughout. The framework positions Serbia for future EU AI Act compliance, anticipating risk-based classification approaches. Council of Europe standards on human rights and AI inform fundamental rights protections.

Implementation Timeline

Sources and References