Session
Description:
Generative AI is rapidly replacing traditional search engines and social feeds as a primary interface for citizens seeking political and electoral information globally. As chatbots become embedded in messaging apps, operating systems, and browsers, they act as powerful information gatekeepers—raising urgent international questions about neutrality, accuracy, algorithmic bias, and compliance with electoral laws. This session addresses how global governance frameworks, regulators, developers, and civil society can evaluate and oversee AI tools as information intermediaries during high-stakes elections. Grounded in a comparative perspective, the panel presents replicable, open-source auditing methodologies and qualitative inquiry frameworks designed to test chatbot behavior across electoral cycles. To ground the theoretical discussion, the session draws on empirical evidence from an ongoing multi-method study in Brazil (conducted by InternetLab, LabDados, and Recos). The Brazilian experience serves as a key case study analyzed as an example, highlighting two universal challenges: (1) User Behavior & Trust, examining how diverse sociodemographic groups query, interpret, and trust AI-generated political content; and (2) Model Auditing & Compliance, presenting automated, longitudinal stress-testing of frontier LLMs against electoral rules, such as prohibitions on ranking candidates, procedural hallucination, and the reproduction of gendered political violence. By contrasting these empirical findings with regulatory approaches and election management strategies from other regions, the session explores what scalable oversight, transparency, and developer accountability should look like worldwide. Rather than focusing on a single jurisdiction, the panel brings together researchers, AI developers, and digital rights advocates to discuss how cross-country auditing frameworks can safeguard electoral integrity without stifling freedom of expression.
Generative AI is rapidly replacing traditional search engines and social feeds as a primary interface for citizens seeking political and electoral information globally. As chatbots become embedded in messaging apps, operating systems, and browsers, they act as powerful information gatekeepers—raising urgent international questions about neutrality, accuracy, algorithmic bias, and compliance with electoral laws. This session addresses how global governance frameworks, regulators, developers, and civil society can evaluate and oversee AI tools as information intermediaries during high-stakes elections. Grounded in a comparative perspective, the panel presents replicable, open-source auditing methodologies and qualitative inquiry frameworks designed to test chatbot behavior across electoral cycles. To ground the theoretical discussion, the session draws on empirical evidence from an ongoing multi-method study in Brazil (conducted by InternetLab, LabDados, and Recos). The Brazilian experience serves as a key case study analyzed as an example, highlighting two universal challenges: (1) User Behavior & Trust, examining how diverse sociodemographic groups query, interpret, and trust AI-generated political content; and (2) Model Auditing & Compliance, presenting automated, longitudinal stress-testing of frontier LLMs against electoral rules, such as prohibitions on ranking candidates, procedural hallucination, and the reproduction of gendered political violence. By contrasting these empirical findings with regulatory approaches and election management strategies from other regions, the session explores what scalable oversight, transparency, and developer accountability should look like worldwide. Rather than focusing on a single jurisdiction, the panel brings together researchers, AI developers, and digital rights advocates to discuss how cross-country auditing frameworks can safeguard electoral integrity without stifling freedom of expression.
Policy Question(s)
1. What obligations should generative AI providers have to prevent chatbots from recommending or ranking political candidates? 2. How can regulators and civil society build replicable, transparent, cross-country methodologies to audit generative AI responses to electoral queries over an election cycle? 3. What role should generative AI play as an information intermediary in elections, and how should oversight be balanced against freedom of expression?
Expected Outcomes
Participants will gain a comprehensive understanding of the structural challenges generative AI poses to electoral information ecosystems worldwide. They will leave with access to an adaptable, open-source auditing methodology and testing infrastructure—demonstrated through empirical data from Latin America—that can be replicated to evaluate AI models in their own national or regional election cycles. Furthermore, the session will establish a cross-regional network linking researchers, election authorities, civil society, and AI developers to co-create global standards for chatbot compliance, auditability, and political neutrality during elections.
Format
Roundtable
Duration (minutes): 90
Format description: A roundtable format is best suited to this session because it enables a candid, interactive exchange among researchers, AI developers, digital rights advocates, election authorities, and participants. Rather than positioning the audience as passive listeners, the layout encourages them to contribute regional experiences, challenge assumptions, and help identify how the proposed auditing methodologies could be adapted across different electoral contexts. The 90-minute duration allows sufficient time to introduce the empirical case study and open-source methodology, examine each of the three policy questions in depth, and facilitate meaningful audience participation. It also provides space to compare regulatory and governance approaches across regions and conclude with concrete areas for collaboration. While the session could be adapted to a 60-minute slot if necessary, 90 minutes is preferable to preserve its comparative, participatory, and solution-oriented character.
Duration (minutes): 90
Format description: A roundtable format is best suited to this session because it enables a candid, interactive exchange among researchers, AI developers, digital rights advocates, election authorities, and participants. Rather than positioning the audience as passive listeners, the layout encourages them to contribute regional experiences, challenge assumptions, and help identify how the proposed auditing methodologies could be adapted across different electoral contexts. The 90-minute duration allows sufficient time to introduce the empirical case study and open-source methodology, examine each of the three policy questions in depth, and facilitate meaningful audience participation. It also provides space to compare regulatory and governance approaches across regions and conclude with concrete areas for collaboration. While the session could be adapted to a 60-minute slot if necessary, 90 minutes is preferable to preserve its comparative, participatory, and solution-oriented character.
Hybrid Format: The session will be structured so that onsite and online participants can join the discussion on equal terms. A dedicated online moderator will follow the chat, invite remote participants to speak, and bring their questions into the room throughout the session.
The discussion will be divided into thematic blocks linked to the three policy questions. At the end of each block, the facilitator will reserve time for interventions from both the onsite and online audiences, rather than leaving all questions to the final minutes.
Mentimeter will be available to everyone, whether they are in the room or joining remotely. We will use it for short polls, collecting questions, and gathering examples or concerns from different countries. This will give participants who may not wish to speak publicly another way to contribute and will help the moderators identify shared priorities and points of disagreement in real time.
Speaker 1: Sarah Wesonga, Civil Society, African Group
Speaker 2: Luiz Moncau, Private Sector, Latin American and Caribbean Group (GRULAC)
Speaker 3: Marina Meira, Civil Society, Latin American and Caribbean Group (GRULAC)
Speaker 2: Luiz Moncau, Private Sector, Latin American and Caribbean Group (GRULAC)
Speaker 3: Marina Meira, Civil Society, Latin American and Caribbean Group (GRULAC)
Moderator
Fernanda Campagnucci, Civil Society, Latin American and Caribbean Group (GRULAC)
Online Moderator
Victor Pavarin Tavares, Technical Community, Latin American and Caribbean Group (GRULAC)
Rapporteur
Fernanda Campagnucci, Civil Society, Latin American and Caribbean Group (GRULAC)
What will participants gain from attending this session?
