Thematic discussion 4: Respecting, protecting and promoting human rights - Global Dialogue on Artificial Intelligence Governance - Day 2

This argument has already been captured above under Volker Turk's entries.

Bahous argues that left unregulated, AI becomes not only a mirror of existing societal inequalities but a powerful mechanism for amplifying discrimination, exclusion, and violence against women and girls. She insists that AI governance cannot be treated solely as a technical issue and is fundamentally a human rights matter.

Bahous stated that left unregulated, AI becomes not only a mirror of the inequalities that already exist in our societies, but also a powerful mechanism for amplifying discrimination, exclusion, and violence against women and girls . She referenced international human rights law frameworks including the Universal Declaration of Human Rights, the International Covenant on Civil and Political Rights, and CEDAW as providing a comprehensive framework to protect rights .

Bahous presents evidence that AI systems reproduce and amplify harmful gender stereotypes because they are trained on data that already reflects existing biases. This is not a marginal problem but a systemic one affecting nearly half of all assessed AI systems.

Bahous cited evidence that 44% of assessed AI systems demonstrate gender bias , and that AI systems reproduce and amplify harmful stereotypes because they are trained on data that already reflects existing biases .

Bahous presents alarming statistics on the gendered nature of AI-facilitated harm, particularly deepfakes and manipulated sexual imagery. She connects this to a broader chilling effect on women's participation in public life and democratic processes.

UN Women research found that almost one in four women human rights defenders, activists, and journalists surveyed had experienced AI-assisted online violence, and 6% had been victims of deepfakes or digitally manipulated imagery . Research shows that up to 99% of online deepfakes and manipulated sexual imagery target women , and 41% of all women respondents reported self-censoring on social media to avoid abuse .

Bahous argues that the accountability of AI governance begins with four questions: who builds AI, who benefits, who bears the risks, and who has a voice. The data reveals that women are systematically excluded from all four dimensions.

Bahous cited that women make up only 30% of the AI workforce globally , and that 88% of leading AI researchers are male according to the independent scientific panel's report . She further noted that of 138 countries assessed, only 24 reference gender in their national AI strategies, and just 18 include substantive gender-responsive provisions .

Bahous calls for mandatory human rights impact assessments before and after AI deployment, with gender equality treated as non-negotiable rather than optional. She also emphasises that accountability requires meaningful participation from women, girls, and other affected communities.

Bahous called for mandatory human rights impact assessments before and after deployment with gender equality treated as non-negotiable and not as optional , and stated that women and girls, feminist technologists, indigenous communities, labour organisations, disability advocates, and civil society must be meaningfully involved and empowered .

Bahous argues that accountability in AI governance requires genuine participation from those most affected, including women and girls, feminist technologists, indigenous communities, labour organisations, disability advocates, and civil society. Without this, governance frameworks will fail to address the needs of those most at risk.

Bahous stated that accountability requires participation, and that women and girls, feminist technologists, indigenous communities, labour organisations, disability advocates, and civil society must be meaningfully involved and empowered .

López Agueda argues that the geopolitical context reveals AI is not merely a consumer product but a political tool with profound implications for power. He frames the choice starkly: either AI governance upholds digital rights or it leads to oligarchy.

López Agueda stated that the geopolitical times show us that AI is not a consumer good but a political tool, and that today AI is powerful and only has two paths: that of digital rights or that towards an oligarchy .

López Agueda warns that without inclusive and global AI governance, the global south risks being reduced to merely providing data and resources for more powerful nations. He calls for a democratic and inclusive global dialogue on AI.

López Agueda stated that either we do something or the global south will just serve as a battery for the greater powers because it is also just going to be someone who provides data, and therefore we have to be inclusive and global when we consider AI .

Nighat Dad argues that human rights due diligence as currently practised by most AI companies does not meet the standard of the UN Guiding Principles on Business and Human Rights. It happens after co-design decisions are made, only where regulations compel it, and without participation from affected people.

Dad stated that human rights due diligence happens after co-design decisions are made, where regulations compel it, and without the participation of the people actually affected, meaning what exists in most cases is not due diligence but documentation . She recommended that gender and child rights impact assessments must be mandatory, conducted before deployment with affected communities participating, not merely documented .

Dad argues that companies conduct meaningful human rights assessments primarily where the law requires them to, such as in the EU, and then deploy the same systems in the majority world without equivalent processes. This is a structural choice, not a capacity gap, creating a two-tier rights regime.

Dad stated that companies conduct meaningful assessments where the law requires them to, primarily in the EU, then deploy the same system in the majority world with no equivalent process, and characterised this as a structural choice, not a capacity gap . She noted that the depth of a risk assessment depends not on the risk faced but on the jurisdiction one happens to live in .

Dad argues that when women withdraw from digital spaces due to technology-facilitated gender-based violence, this is not merely an individual harm but a collective harm to public discourse and democratic participation. Any due diligence framework that fails to explicitly address this issue excludes half the population from its protection.

Dad's organisation, Digital Rights Foundation, documented more than 23,000 cases of technology-facilitated abuse over 10 years, disproportionately against women, girls, and young people, including non-consensual synthetic imagery, automated moderation that fails in Urdu, Hindi, and Bengali, and recommendation systems that amplify gendered harassment . She stated that when women withdraw from digital spaces, it is a collective harm to public discourse and democratic participation .

Dad calls for three concrete recommendations: mandatory gender and child rights impact assessments before deployment with affected communities participating; uniform human rights due diligence obligations across all markets; and repeated assessments whenever AI capabilities materially change, including increases in agentic function.

Dad recommended that gender and child rights impact assessments must be mandatory, conducted before deployment with affected communities participating, not merely documented , and that human rights due diligence obligations must apply uniformly across all markets where a system is deployed, noting that a risk assessment covering Brussels but not Pakistan, Kenya, or Brazil is not due diligence .

Dad argues that the evidence problem in AI governance is fundamentally a resourcing problem. Civil society organisations in the majority world are the earliest sensors of AI-related harm, but their documentation is unfunded, precarious, and treated as anecdote rather than evidence.

Dad noted that her organisation documented synthetic intimate imagery targeting women years before it entered global policy conversations, but that this documentation is unfunded, precarious, and treated as anecdote rather than evidence . She called for this existing early warning system to be resourced and connected to decision-making, including the scientific panel .

Chetty argues that prevailing governance models anchored in ethical or responsible AI remain structurally inadequate because they are largely self-regulatory, driven by dominant actors, and calibrated to high-income countries. She proposes the Just AI framework as a normative alternative rooted in international human rights law.

Chetty stated that prevailing governance models anchored in notions of ethical or responsible AI remain structurally inadequate, as such frameworks are largely self-regulatory, driven by dominant actors, calibrated to high-income countries, and disregard sovereignty or agency . She argued that such frameworks deepen existing inequalities by extracting value from individuals, communities, and nations without reciprocal benefit, input, or control .

Chetty argues that the governance of AI is inseparable from the governance of data, and that separating these two areas in international forums obscures the foundational data justice issues that must inform all AI systems. Data justice is crucially informed by a right of access to data in substantive and procedural forms.

Chetty called for establishing a globally recognised understanding that the governance of AI is inseparable from the governance of data, and stated that the separation of data governance and AI governance in international forums obscures the foundational data justice that must inform all AI systems .

Gutiérrez argues that the rapidly changing landscape of AI development means that governance cannot be limited to regulation alone. What is needed is the adoption of a new social contract at both the international and national levels that addresses the fundamental question of what technology is doing to human beings.

Gutiérrez stated that we should talk about governance and the adoption of a new social contract, not only in the international context but also in the national context . He also noted the separation between legality and legitimacy in the international backdrop, and the importance of understanding the impact of technological development on human beings rather than only what human beings can do with technology .

Lampel argues that the rule of law requires implementing binding legal frameworks equally without regard for wealth, power, or influence, and that the international community must start enforcing international human rights law today. He proposes three concrete actions: AI red lines, standardised certifications linked to procurement rules, and a right to remedy.

Lampel referenced the 2024 UN General Assembly resolution on safe, secure, and trustworthy AI systems for sustainable development as explicitly supporting the concept of AI red lines , and cited UNESCO's recommendations on the ethics of AI as already explicitly ruling out the use of AI for social scoring and mass surveillance .

Cassayre argues that human rights and the rule of law must underpin the development and deployment of digital technologies, notably in the justice sector. AI systems must be designed to expand access, serve people's needs, and prevent new forms of exclusion and inequality throughout their entire lifecycle.

Cassayre offered three recommendations: that human rights and the rule of law must underpin the development and deployment of digital technologies; that fit-for-purpose legal and regulatory frameworks must accompany innovation; and that investment in institutional capacity is needed, including strengthening public digital infrastructure for justice delivery and enhancing AI literacy among legal and justice professionals .

Arias Arzeno argues that the true challenge of AI governance is not in adopting laws or ethical principles but in what happens when a person believes an AI system has violated their human rights. Justice systems are essential institutions for converting principles into reality and ensuring accountability when AI causes harm.

Arias Arzeno stated that governance is no longer just a concept but becomes a reality at the moment when a person considers an AI system has violated their human rights and data privacy . He outlined four priorities for strengthening justice systems: strengthening institutional capacities of lawyers and jurors, ensuring transparency and traceability in the use of AI, having effective human oversight, and ensuring all people have effective mechanisms to access justice and repair harm .

Arias Arzeno argues that the use of AI in courtrooms and public institutions is already a reality, and the challenge is to ensure its use is transparent, responsible, auditable, and compatible with human rights and the rule of law. No institution can abdicate responsibility by attributing a decision to an AI system.

Arias Arzeno stated that no public institution can justify a decision by saying that is what the system told us to do, and that responsibility should lie with people and institutions, with technology helping the judge but never replacing them .

El Fallah Seghrouchni argues that accountability in agentic systems is the most significant legal challenge of the decade because networks of individually autonomous AI agents produce final decisions with no single human at their origin. It is extremely difficult to distribute accountability among agents or to trace back up the chain to identify where a mistake was made.

El Fallah Seghrouchni gave the example of the Iderati X.0 platform in Morocco, which processes 52 million administrative transactions daily, noting that when there is a problem it is very difficult to find the exact cause . She described the challenge of LLMs with between 175 billion and 180 billion parameters and the difficulty of ensuring accountability for what appears in these systems .

El Fallah Seghrouchni proposes three concrete obligations for accountability in agentic AI systems: structured documentation of agent chains specifying who does what and when; appointment of an identifiable human being in charge for all rights-related issues; and a right to redress in good time in line with the speed of AI systems.

El Fallah Seghrouchni proposed that when agents are deployed in public services, there must be structured documentation of agent chains, an identifiable human being in charge for all issues tied to rights, and a right to redress in good time in line with the speed of AI systems . She concluded that the functioning of the algorithm must never be autonomous in regards to rights and the law .

Luccioni argues that transparency about AI's energy consumption and environmental costs is a fundamental human right because it enables users to make informed choices between tools. Without this information, meaningful choice and accountability are impossible.

Luccioni stated that transparency is a fundamental human right that users should be getting when they use AI, asking what the cost is, where the data comes from, and where the energy comes from, so that users can make a choice between tool A or tool B . She noted that even organisations like the IEA struggle to get this information because member countries themselves do not have it .

Luccioni argues that agentic AI worsens an already difficult transparency and accountability situation by adding more layers between users and the underlying technology. As AI systems become more agentic, it becomes increasingly difficult to identify where failures originate.

Luccioni described the progression from AI models that could be probed, to AI systems with filters and additional components, to AI agents with even larger degrees of separation, noting that when something fails, you do not know where it is coming from . She called for imposing safeguards and standards for safety and sustainability at each part of the machine, regardless of how many parts there are .

Luccioni argues that while the benefits of AI can technically be global, the negative environmental impacts are very local, concentrated around data centres in terms of water, energy, emissions, and health. These local harms disproportionately affect the most marginalised populations.

Luccioni stated that while the benefits of AI can technically be global, the negative impacts in terms of water, energy, emissions, and health are very local and impact the most marginalised populations . She noted that data centres are a concrete way to anchor the debate about AI's human rights implications .

Luccioni highlights a significant transparency deficit in AI's energy consumption, noting that even major international organisations like the IEA cannot obtain reliable data from member countries. This lack of transparency prevents meaningful accountability and informed decision-making.

Luccioni stated that organisations like the IEA, which she has worked with, struggle to get information about AI energy consumption because even member countries do not have the information . She described transparency as something she has been fighting for several years as a fundamental human right .

Luccioni proposes energy efficiency labels for AI models as a practical first step towards transparency and accountability for AI's environmental impact. Rather than immediately restricting deployment, she advocates for first quantifying and measuring impacts so that people can make informed decisions.

Luccioni described the AI Energy Score Project, which creates energy efficiency labels for AI models analogous to energy labels for appliances, as a stepping stone to informed decision-making . She argued that quantifying and measuring impacts is the prerequisite to so many other things that can and should be done when it comes to AI .

Aerdts presents the Netherlands' practical tools for AI accountability as concrete examples of how legal requirements can be translated into actionable guidelines. These include an algorithm framework, a fundamental rights and algorithm impact assessment, and an algorithm register.

Aerdts described the Netherlands' algorithm framework that translates legal requirements into practical guidelines for public organisations, including a fairness handbook, bias testing instruments, and risk profiling assessments . She also mentioned the fundamental rights and algorithm impact assessment that helps identify human rights risks in AI systems , and the algorithm register that includes all algorithms used by the Dutch government in the interest of transparency .

Aerdts argues that when making policy for children and young people, they must be genuinely included in the decision-making process. She highlights a gap where young people face issues that are too big for their educators or parents but too small for the police, leaving them without tangible support.

Aerdts noted that the Netherlands speaks to UNICEF youth panels and wants to include parents as well, while ensuring platforms take their responsibility . She recounted that when speaking to young people, they mentioned that some issues they face are too big for their educators or parents but feel too small for the police, leaving them looking for a tangible way to address them .

This argument has already been captured above under Jhalak M. Kakkar's entries.

This argument was made by Alvitta Ottley, not Sonia Livingstone. Please see Alvitta Ottley's entry for this argument.

Lametti describes Canada's proposed legislative approach to AI safety for children, which would require AI chatbots to be proven safe before they could be exposed to children. A digital commissioner would be established to ensure compliance with this requirement.

Lametti described Canada's new policy, AI for All, which aims to ensure privacy rights and safety in the use of AI, and a new proposed law on children that would restrict any social platform to children under 16 until proven safe . He noted that AI chatbots would also have to be proven safe before exposure to children, with a digital commissioner to ensure compliance .

Henriques argues on behalf of over 120 signatories of a joint statement that children's rights must be central to AI governance as a cross-cutting consideration. Children are among the earliest and most frequent users of new technology, yet they remain largely missing from national AI strategies and governance frameworks.

Henriques spoke on behalf of over 120 signatories of the Joint Statement on Children's Rights at the AI Dialogue, a global campaign coordinated by Five Rights Foundation . She noted that children are among the earliest and most frequent users of new technology, yet remain largely missing from national AI strategies and governance frameworks .

Henriques argues that children must not be treated as a homogeneous group in AI governance discussions. Indigenous children, girls, children with disabilities, and children of African descent face disproportionate impacts from gaps in AI governance, and their intersectionalities must be recognised.

Henriques stated that children are not a homogeneous group and that we must recognise their intersectionalities, noting that indigenous children, girls, children with disabilities, and children of African descent have their rights disproportionately impacted as a result of gaps in AI governance . She highlighted Brazil's ECA Digital as a landmark framework for the protection of children's rights in the digital environment .

Henriques argues that children should only interact with AI systems that have been proven safe and designed to respect children's rights, considering their perspectives in decisions that affect them. Responsibility must sit with those who build and profit from the technology, not with children or parents.

Henriques stated that children should interact with AI systems only if they are proven safe and designed to respect children's rights and considering their perspectives in decisions that affect them . She emphasised that responsibility must sit with those who build and profit from this technology .

Eghobamien-Mshelia argues that CEDAW provides an essential framework for addressing the gendered realities of AI governance. It requires states not only to prohibit discrimination but to achieve substantive equality by tackling structural and regulatory barriers, stereotypes, and unequal power relations.

Eghobamien-Mshelia stated that CEDAW requires states not only to prohibit discrimination but to achieve substantive equality by tackling structural and regulatory barriers, stereotypes, and unequal power relations . She noted that CEDAW also recognises that discrimination is often compounded by factors such as race, disability, age, migration status, and poverty, requiring tailored and proactive action to curb intersectional discrimination .

Smyth argues that AI-amplified disinformation campaigns and hate speech undermine the rights of women and girls to exercise freedom of expression by forcing them to self-censor, close their digital accounts, or withdraw from online spaces. This contributes to a broader chilling effect on their full and equal participation in society.

Smyth stated that AI-amplified disinformation campaigns and hate speech undermine the rights of women and girls to exercise freedom of expression by forcing them to self-censor, close their digital accounts, or otherwise withdraw from participating in online spaces . She noted that this contributes to a broader chilling effect on their full, equal, and meaningful participation in society .

Smyth argues that the deployment of AI for online content moderation without human oversight can amplify disinformation, undermining the integrity of the information landscape. This disproportionately affects women and girls.

Smyth stated that the deployment of AI online content moderation without human oversight can amplify disinformation, undermining the integrity of the information landscape, and that this disproportionately affects women and girls .

Brant highlights Brazil's legislative action against AI-facilitated gender-based violence, including a decree establishing proactive responsibilities for platforms and AI companies, including the prohibition of creating synthetic intimate images based on real third parties.

Brant described Brazil's digital statute for children and adolescents online, which requires platforms to adopt child protection measures from the design phase and prohibits targeted advertising based on children's personal data . He also described a decree released the previous month to address violence against girls and women in online spaces, including the prohibition of creating synthetic intimate images based on real third parties .

Brant argues for a broad perception of human rights when discussing AI, highlighting information integrity as a key issue. He frames information integrity as part of Article 19 in its collective or social dimension, encompassing the need for an information ecosystem that provides consistent, accurate, and reliable information.

Brant called for taking information integrity as part of Article 19 of the government in a broad perception of access to information in its collective or social dimension, and stated that we need an information ecosystem that can provide consistent, accurate, and reliable information . He also noted that the way AI is supporting and affecting the economic arrangement and sustainability of journalism should be a concern .

Park argues that while individual companies may define AI risk and evaluate their models accordingly, the problem arises when multi-agent systems from different companies are interconnected, as each company has its own definition of AI risk. Common and well-designed standards are needed to ensure consistent human rights due diligence across these systems.

Park stated that the problem comes when multi-agent systems from different companies are interconnected, as companies have their own definition of AI risk and some agents may consider human rights more than others . He referenced the AI risk taxonomy from the BTEC project at OHCHR as important work that needs to be developed in a more concrete and practical way to find a common baseline with many stakeholders .

Park argues that today's children are AI natives who use AI naturally in search, recommendation, and learning in their daily lives. It is therefore important for them to develop critical thinking skills to engage with AI in a balanced way, and their voices should flow into governance discussions.

Park stated that today's children are AI natives who use AI naturally in search, recommendation, and learning in their daily lives, and that it is important for them to accept it with critical thinking . He suggested that when school programmes for education and discussion are connected with NGOs and relevant committees, children can voice how AI affects them in a balanced way, and those voices can flow into governance discussions .

Ottley argues that there is a fundamental mismatch between the questions that engineers ask about AI (can we make it faster, more accurate, more capable?) and the questions that societies are asking (are human rights protected, are institutions accountable, is society better off?). These are simply different questions that lead to different evaluations of success.

Ottley illustrated the mismatch by noting that a parent does not ask if the model scored 94% on a recent benchmark but asks whether their child can use it and decide when not to trust it, and a physician does not ask if it is the latest model but whether they can rely on it when a patient's life is at stake . She stated that the conversations in the room are not asking whether AI is more capable than last year but whether human rights are protected, institutions remain accountable, and society is better off .

Ottley highlights a significant evidence gap: while there is abundant evidence about how well AI models perform, there is very little evidence about how AI is affecting people's decision-making, children's learning, clinical decisions, or public servants' work. Answering these questions requires longitudinal studies and interdisciplinary collaboration.

Ottley stated that we have a lot of evidence about how well AI models perform but very little evidence about how AI is affecting people's decision-making, how well AI has helped teachers teach, how AI affects clinical decisions, how AI helps children learn, or how AI affects public servants . She noted that answering these questions requires longitudinal studies, partnerships with organisations, interdisciplinary collaborations, and careful work with protected and underserved communities .

Ottley argues for a new way of thinking about the relationship between society and science in AI governance. Society must decide what success looks like, science must determine how to measure it, and together they must build the evidence to determine whether success is being achieved.

Ottley stated that society has to decide what success looks like and science has to determine how to measure it, and together they have to build the evidence that tells us whether we are actually achieving it . She concluded that good decisions under uncertainty do not come from eliminating uncertainty but from making values explicit, evidence rigorous, and uncertainty understood .

Ottley highlights a methodological challenge that is rarely discussed: children are a protected category in research, meaning there are limitations on what can be studied and how. This partly explains the lack of evidence on AI's impact on children and means that things are often developed without children being involved.

Ottley noted that children, along with pregnant women and people in prison, are considered a protected category of the population, meaning there are limitations on what can be studied and how, which is one of the reasons why there is a lack of evidence on the impact of AI on children . She stated that this also means that a lot of times things are developed without having children involved .

Paullier argues that young people are the most users of AI and many of the innovators within AI technology firms, yet they are almost never the decision-makers. This is a significant governance gap given that they will live longest with the consequences of decisions made today.

Paullier stated that young people are the most users of artificial intelligence and are actually most of the innovators and creators within some of these huge AI technology firms, but almost always they are not the decision-makers . He noted that they will live the longest of the consequences of the decisions that we make today .

Paullier argues that meaningful spaces for young people to engage in AI governance must be created especially at the national level, not only in global forums. Young people are those who first experience the impact of technology in their education, employment, and everyday digital life, and they best understand both the opportunities and the harms.

Paullier stated that young people are those who first experience the impact of technology in their education, employment cycles, and everyday digital life, and that they forge their identities in the digital space . He argued that young people are the ones who understand the most about which opportunities AI can bring but also the ones who will help identify biases, discriminations, and exclusions .

Paullier warns that if meaningful spaces for young people to engage in AI governance are not created, young people will find their own way and create their own spaces, as has happened in the climate space. This underscores the urgency of proactive inclusion.

Paullier stated that if meaningful spaces for young people to engage are not created, young people will find their way and create their own spaces, as has happened within the climate space .

Paullier reports that across regions, young people consistently raise the human dimension and the impact of AI on well-being, mental health, and meaningful human relationships as their primary concern. While young people value technology, what they value most is belonging, trust, and meaningful human relationships.

Paullier stated that what young people and citizens in general value most is belonging, trust, and meaningful human relationships . He called for emphasis on the impact of technologies on the well-being and mental health of communities, and for working not only for youth but with youth .

Osterling argues that journalism is a keystone species of the information ecosystem that stabilises trust, anchors verification, structures accountability, and enables other institutions to function. AI systems are trained on journalistic content scraped without consent, compensation, or credit, undermining journalism's business models and public trust.

Osterling cited her colleague's description of journalism as a keystone species of the information ecosystem, noting that removing it means the ecosystem does not adapt but collapses . She stated that AI systems are trained on journalistic content scraped without consent, compensation, or credit , and called for public digital infrastructure, industrial policies ensuring public interest AI, and AI governance requiring human rights to be embedded across the full lifecycle of AI systems .

Tan argues that AI-generated disinformation and online hate speech have become direct protection risks for refugees and displaced people. She presents concrete examples from Libya and the Rohingya community where AI-generated content has incited hostility, abuse, and violence.

Tan cited the example of Libya, where violent hate speech and dangerous misinformation including AI-generated content incited hostility and abuse of refugees and affected the safety of UNHCR and humanitarian workers . She also noted that the Rohingya continue to face dehumanising narratives online even in exile, and that this kind of hate can fuel abuse, exclusion, violence, and fear .

Bellion-Jourdan argues that geopolitical competing interests and power asymmetries are impediments to global AI governance, and that leveraging the tools and techniques of multilateral negotiation can help navigate these challenges. The goal is to shift from positional bargaining and win-lose dynamics to integrative negotiation and win-win outcomes.

Bellion-Jourdan argued that if we leverage the power of negotiation with all its tools and techniques to navigate power politics and power asymmetry, this would allow us to forge common ground . He called for investment in process design, process management, and innovative formats of negotiations, noting that it will not be enough to lock 193 states in a room and expect agreement .

Kownacki welcomes the Council of Europe Framework Convention on AI as the first binding treaty to place AI within the order of human rights, democracy, and the rule of law. He argues that Poland is turning these principles into practice at home through its national digitisation strategy and AI policy.

Kownacki stated that Poland welcomes the Council of Europe Framework Convention on AI, describing it as the first binding treaty to place AI within the order of human rights, democracy, and the rule of law . He noted that Poland's national AI policy commits to artificial intelligence that is human-centric and trustworthy .

Kownacki argues that the human factor — keeping a person with the final word to understand, contest, and switch off AI decisions — is non-negotiable and represents the common ground on which truly global AI governance can be built. This must be written into values, laws, standards, and institutions, not left as a declaration.

Kownacki stated that however capable AI systems become, a person must keep the final word, to understand the decision, to contest it, and where it matters most to switch the systems off, and that this is non-negotiable above all in high-risk systems . He argued that the human factor belongs to no single nation and is the common ground on which a truly global governance of AI can be built .

Shahani argues that compatibility between national approaches to AI governance does not require uniformity. Instead, a common floor of rights safeguards — including disclosure, contestability, and remedy — must apply to every person affected by AI, with genuine pluralism permitted above that floor.

Shahani stated that compatibility between approaches globally does not require uniformity, and that we can have a common floor of rights safeguards including disclosure, contestability, and remedy for every person affected, but with genuine pluralism above that floor .

Shahani argues that safety claims for AI systems must be accompanied by the conditions under which systems were tested, known limitations, and failure rates disaggregated by language, gender, disability, region, and skin tone. Without this, it is impossible to hold a system accountable for where it fails.

Shahani stated that safety claims must be accompanied by the conditions under which systems were tested, known limitations and failure rates disaggregated by language, gender, disability, region, skin tone, and more, and that disclosure of training data provenance must be standard . She noted that we cannot hold a system accountable if we cannot see where it fails .

Shahani argues that global majority countries largely import AI models trained elsewhere and deploy them in high-stakes settings without the means to evaluate them first. She calls for publicly financed, regionally hosted capacity to evaluate AI systems before deployment and open benchmarks built with and for under-resourced languages.

Shahani stated that global majority countries largely import models trained elsewhere and deploy them in high-stakes settings without the means to evaluate them first, and called for publicly financed, regionally hosted capacity to evaluate AI systems before they are deployed and open benchmarks built with and for under-resourced languages .

Shahani argues that indigenous and traditional knowledge is being absorbed into AI training data without consent, recognition, or return. She calls for community-led data stewardship to be recognised as part of human oversight.

Shahani stated that indigenous and traditional knowledge is being absorbed into training data without consent, recognition, or return, and that community-led data stewardship must therefore be recognised as part of human oversight .

Bonyo argues that the concentration of AI governance conversations in Geneva, combined with visa restrictions that rely on opaque algorithms, effectively locks out the rest of the world from these discussions. She calls for transparency in the algorithms that determine whether people can travel to participate in governance forums.

Bonyo cited a report stating that 51% of AI governance conversations this year happen in Geneva, meaning the rest of the world is effectively locked out because they cannot access visas . She called for algorithmic openness around visa processes and asked the global north to reconsider the kind of algorithms outsourced to determine whether people travel .

Bonyo presents the Africa AI Governance Index, launched by Lawyers Hub, as a tool for tracking AI strategies, laws, and institutions across all African states for the first time. This provides an assessment tool that supports inclusive governance by making African AI governance visible and comparable.

Bonyo stated that the Africa AI Governance Index tracks AI strategies, laws, and institutions across all African states, and that for the first time there is an assessment that covers everyone across the African continent . She invited participants to visit AIpolicy.africa to interact with the work of the AI Lab .

De Bellis Bonilla highlights Uruguay's commitment to driving forward a digital transformation that includes AI development and use in an inclusive and gender-based approach. She points to the Global Partnership for Human Rights as a broad multi-stakeholder platform that connects efforts on AI's cross-cutting impacts.

De Bellis Bonilla noted Uruguay's adherence to the Madrid Declaration on the Fifth Ministerial Conference on the Feminist Foreign Policy , and highlighted the Global Partnership for Human Rights from OHCHR as a broad multi-stakeholder platform and network connecting all efforts with a cross-cutting focus on the impact of AI .

Gigena calls on behalf of Access Now for a halt to the development and use of AI tools that are facilitating violations of international humanitarian law and international human rights law, especially in contexts of conflict or war. He also calls for binding legal standards and global governance mechanisms to regulate digital surveillance technologies and AI.

Gigena called for a halt to the development and use of AI tools facilitating violations of international humanitarian law and international human rights law, especially in contexts of conflict or war . He also echoed calls for the adoption of human rights-centred, binding legal standards and global governance mechanisms to regulate digital surveillance technologies and AI, addressing chilling effects .

Cha argues that access to AI should be treated not as a privilege to be earned but as a basic right to be enjoyed. He notes that global efforts to build AI readiness remain fragmented, and that the Republic of Korea has declared a vision for establishing a Global AI Hub to contribute to strengthening global AI capacity.

Cha stated that the Republic of Korea considers access to AI not as a privilege to be earned but as a basic right to be enjoyed . He noted that as countries deepen their readiness for AI, their approaches to governing it also deepen, but globally these efforts remain fragmented , and described Korea's vision for establishing the Global AI Hub together with nine participating AI and related organisations .

The APC representative argues that before communities can contribute data, build local AI solutions, influence policy, or benefit from innovation, they need meaningful, affordable, trusted, and locally relevant connectivity. Community-centred connectivity shifts power closer to communities and enables women, indigenous people, rural communities, and speakers of underrepresented languages to become creators and decision-makers.

The speaker stated that community-centred connectivity is more than providing internet access; it is about communities owning, managing, and governing the infrastructure that connects them, creating space for local knowledge, local languages, and community priorities to thrive . The speaker concluded that communities are not the last mile of connectivity or of AI but are actually the first mile .

Kiladze argues that the question of AI and children's rights is ultimately a moral one about whether there is the will from states, companies, and international organisations to design, develop, and govern AI that respects children's rights. History will judge us not by the intelligence of our AI but by whether we protected those with the smallest voice and the greatest stake in the future.

Kiladze stated that history will not judge us by how intelligent our AI or machines are, but by whether we used this extraordinary moment to protect those who had the smallest voice but the greatest stake in the future . She referenced the joint UN statement on AI and the Rights of the Child, co-led by ITU, the CRC, and UNICEF, which unified 13 UN agencies and over 60 organisations, with children from all five UN regions helping to shape it .

This argument was primarily made by Shumaila Hussaini Shahani. Clara Chappaz's role in the session was as moderator, and her closing remarks synthesised the panel's contributions rather than advancing this specific argument independently.

Hendricks argues that innovation must not be pursued at the expense of children's safety and rights. Children should only interact with AI systems that have been proven safe and designed with their rights in mind, and responsibility must rest with those who build and profit from the technology rather than with children or families.

Hendricks stated that if there is potential harm, we need to prioritise protection, and that children should interact with AI systems only if they are proven safe and designed to respect children's rights and considering their perspectives in decisions that affect them . She emphasised that responsibility must sit with those who build and profit from this technology, and called for building an AI future worthy of every child's trust and safety .

Baruch argues that responsible AI governance requires that creators and artists retain control over how their work is used to train AI systems. Agreements reached freely and voluntarily between creators, rights holders, and technology developers are the foundation for trust, legal certainty, and sustainable innovation.

Baruch stated that creators and artists must remain in control of how their work is used to develop AI systems, and that agreements reached freely and voluntarily between creators, rights holders, and technology developers create trust, legal certainty, and sustainable innovation . He referenced Universal Music Group's partnerships with companies including NVIDIA, Stability AI, and Spotify as practical examples of this approach .

Baruch argues that creators must know when and how their works have been used to develop AI models and how they will be used for outputs. Without transparency, there can be no accountability, no trust, and no meaningful collaboration between the creative and technology sectors.

Baruch stated that creators should know when and how their works have been used to develop AI models and how they will be used for outputs, and that without transparency there can be no accountability, no trust, and no meaningful collaboration . He argued that with transparency, artistic integrity can be protected and, where necessary, enforced .

Baruch argues that AI is an extraordinary creative tool but must remain subordinate to human creativity rather than replacing it. Music, as a profoundly human art form carrying memory, identity, emotion, and lived experience, exemplifies why human oversight of AI in the creative sector is essential.

Baruch stated that AI is an extraordinary creative tool but it should amplify human creativity, not replace it, and that music is profoundly human, carrying memory, identity, emotion, and lived experience . He cited Bjorn Ulvaeus's description of music as testimony, and argued that if applied responsibly, AI can ignite a new renaissance of human creativity preserving cultural heritage and inspiring future generations .

Baruch grounds the music industry's approach to AI governance in international human rights law, arguing that the rights of creators to protection of their moral and material interests are enshrined in foundational human rights instruments and remain just as relevant in the AI era.

Baruch stated that Article 27 of the Universal Declaration of Human Rights and Article 15 of the International Covenant on Economic, Social, and Cultural Rights recognise that creators are entitled to protection of the moral and material interests from their work, and that those principles are just as relevant in the age of AI .

Hicks presents the results of a survey conducted among session participants, revealing that many are uncertain or concerned about the adequacy of current AI governance measures. The survey identified accountability and rule of law, and the impact of AI on children, as the top human rights concerns among participants.

Hicks reported that a plurality of survey respondents were quite concerned that we do not have the measures in place to address the challenges and opportunities discussed in the session . She noted that the top two human rights issues identified in the survey were accountability and rule of law, and the impact of AI on children, followed by surveillance and use of AI by law enforcement, and environmental sustainability .

Hicks presents survey findings showing that participants identified a hierarchy of responsibility for agentic AI: first, companies that develop AI models; second, states that must ensure regulatory environments prevent harm; and third, organisations that deploy agentic systems. This is consistent with the UN Guiding Principles on Business and Human Rights.

Hicks reported that survey respondents identified the companies that develop AI models as the primary responsible party for agentic AI actions, followed by states and the public authorities that allowed the system, and then the organisations that deploy it . She noted that this is consistent with the UN Guiding Principles on Business and Human Rights, under which the first pillar is that states are responsible for ensuring the regulatory environment exists to prevent companies from causing harm .

Hicks references a survey of 1,000 children in 49 countries conducted with ITU, which found that children want meaningful participation in shaping AI rather than simply being protected from it. Children expressed that they want AI to respect their rights and want to be included in governance discussions.

Hicks described a survey called 'Me and AI' of 1,000 children in 49 countries, noting that children said they want AI to be shaped with them, not just for them, and they want it to respect their rights and do not just want to be shielded from it .

Pipan argues that Slovenia's view is that trust in AI must be understood not merely as an ethical aspiration but as a practical prerequisite for the broader adoption of AI systems. This trust must be built on a human-centric approach grounded in human rights, democratic values, ethics, safety, transparency, and accountability.

Pipan stated that in Slovenia's view, trust is not only an ethical objective but also a prerequisite for the broader adoption of AI, and that this approach is reflected in Slovenia's national AI strategy 2030, adopted in March of that year . She noted that the strategy clearly promotes a human-centric approach to AI grounded in human rights, democratic values, ethics, safety, transparency, and accountability, with the objective of ensuring that AI is developed and used responsibly in the public interest .

Pipan argues that declaring principles is insufficient; building genuine trust in AI requires translating those principles into effective and practical implementation. This entails ensuring responsible governance, legal certainty, appropriate support mechanisms, and sustained investment in AI literacy and technology skills.

Pipan stated that the strategy recognises that building trust requires effective and practical implementation, and that this entails ensuring responsible governance, legal certainty, appropriate support mechanisms, and sustained investment in AI literacy and technology skills .

Pipan highlights Slovenia's particular concern as a small linguistic community that inclusive AI must reflect and preserve all languages and cultures. She argues that AI governance must ensure that no community is left behind due to linguistic or cultural underrepresentation in AI systems.

Pipan stated that as a small linguistic community, Slovenia places particular importance on protecting linguistic and cultural diversity, and that inclusive AI should reflect and preserve all languages and cultures, ensuring that no community is left behind .

Pipan calls for strengthened international cooperation as a key pillar of advancing trustworthy AI globally. She argues that this cooperation should take the form of exchanging good practices, developing interoperable approaches, and building capacity across countries.

Pipan outlined four forward-looking priorities for Slovenia, including strengthening international cooperation through the exchange of good practices, interoperable approaches, and capacity building to advance trustworthy AI globally . She also called for complementing regulatory frameworks with practical implementation support, including capacity building, guidance, and AI literacy .

Ng emphasises the significance of the dialogue by characterising it as timely and important, and acknowledges the substantial preparatory work done by OHCHR and his counterpart. This implicitly argues that effective AI governance requires coordinated effort across UN system organisations rather than siloed approaches.

Ng thanked OHCHR and the co-chairs for doing a lot of the heavy lifting in preparing the session, and specifically acknowledged his OHCHR counterpart Li Zhou for his massive efforts .

The Co-moderator explains that the audience intervention segment is organised to alternate between one representative from member states and one representative from other stakeholders. This procedural arrangement reflects a substantive commitment to inclusive multi-stakeholder participation as a governance principle, ensuring that civil society, private sector, and other non-state voices are formally embedded in the dialogue structure alongside governments.

The Co-moderator stated that in the interest of promoting broad participation from stakeholders, the segment would proceed with one representative from member states and one representative from other stakeholders, alternating between the two .

The Co-moderator emphasises that a timer has been set and that microphones will be automatically switched off once the allocated speaking time expires. This procedural commitment reflects a substantive principle that governance forums must actively manage participation to ensure the broadest possible range of voices can be heard, rather than allowing a few speakers to dominate.

The Co-moderator noted that a timer had been set visible on screen and at the lectern, and that microphones would be automatically switched off once the allocated speaking time expired, encouraging all speakers to respect the allotted time so that the largest possible number of speakers could be accommodated .