UNDP outlines responsible AI use in electoral administration

UNDP and the UN Department of Political and Peacebuilding Affairs (DPPA) have published a technical guide on AI in electoral administration to help election authorities assess the responsible adoption of the technology.

The publication, From Promise to Practice: AI in Electoral Administration, was produced jointly by UNDP and DPPA’s Electoral Assistance Division. It is intended as a practical resource for electoral management bodies considering how AI could support their work.

The report notes that AI is not new to elections, with electoral authorities already using technologies such as biometric voter identification, optical ballot scanning and algorithmic analysis of voter registration databases.

However, the report argues that generative AI, large language models and agentic systems represent a significant shift. While they could improve public outreach, anomaly detection, organisational efficiency and voter communication, they also introduce risks related to hallucinations, bias, reliability and public trust.

The publication stresses that AI adoption should form part of a broader digital transformation strategy, including stronger data governance, digital public infrastructure and organisational capacity.

The report says effective AI adoption in elections depends on political and public consensus, reliable technical implementation, and transparent governance. It notes that past use of digital technologies in elections shows the need for clear problem definition, rigorous testing, and gradual adoption.

Reliability is identified as a central concern. The report warns that inaccurate or misleading AI outputs could affect voter information, election operations and public confidence. In electoral settings, even minor errors can affect voting rights, trigger legal disputes or undermine trust in election outcomes.

The publication also highlights inclusion as a core requirement. AI systems can support inclusion if they are designed with representative data, inclusive testing, participatory design, and continuous monitoring. However, biased datasets or poorly designed systems can disadvantage women, young people, persons with disabilities, minorities, and other groups.

Data governance is another major theme. Electoral management bodies often hold sensitive personal data, including biometric information, while operating under strong transparency expectations. The report says principles such as proportionality, informed consent, and data quality must be translated into practical policies.

The report groups AI applications into five functional areas: analysis, recognition, automation, content creation and voter communication. Examples include anomaly detection, biometric verification, workflow automation, multilingual outreach and AI-powered chatbots.

The publication identifies 12 features to guide electoral management bodies. The features include understanding the need, building political consensus, protecting rights, managing risk, ensuring human oversight, testing early and often, designing for inclusion, forming skilled and diverse teams, building securely, addressing privacy, defaulting to open approaches where appropriate, and designing systems for the future.

The report also links AI in electoral administration to the Global Digital Compact, which promotes a responsible, transparent, accountable, and human-centric approach to emerging technologies. It says electoral authorities should consider how commitments on digital public infrastructure, open-source tools, safeguards, data standards, and human oversight apply to their work.

UNDP and DPPA say the value of AI in elections should be measured by whether it makes electoral processes more credible, inclusive, and resilient, as well as more efficient.

Rather than endorsing AI for electoral processes, the publication provides a framework to help electoral authorities assess whether, where and how AI can be adopted responsibly.

Why does it matter?

Elections are among the most sensitive public processes, meaning AI systems must be deployed with exceptional care. While AI could improve administrative efficiency, voter communication and fraud detection, failures involving accuracy, bias, privacy or transparency could undermine public confidence and the integrity of electoral processes.

The guidance also reflects a broader shift in AI governance from high-level principles to practical implementation. By focusing on human oversight, data governance, inclusion, testing and institutional capacity, UNDP and DPPA are encouraging election authorities to treat AI as a governance challenge that requires careful planning rather than a simple technological upgrade.

Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our Diplo chatbot!