Human rights must move from the margins to the centre of AI governance if societies are to harness AI without undermining democracy, equality and public trust, speakers argued during the fourth thematic discussion of the UN Global Dialogue on AI Governance.
Bringing together governments, UN agencies, civil society, academia and industry, the session examined how AI systems can better respect human rights through stronger transparency, accountability and human oversight. Participants agreed that AI governance should be grounded in international human rights law throughout the entire AI lifecycle, from design and development to deployment and oversight.
AI deserves the same safeguards as medicines
UN High Commissioner for Human Rights Volker Türk opened the discussion by comparing AI regulation to the approval process for new medicines. Drugs undergo years of testing before reaching patients, he noted, yet AI systems are being deployed at unprecedented speed despite already contributing to mass surveillance, online disinformation, discrimination and growing risks to children.
Türk rejected the notion that regulation inevitably slows innovation, arguing instead that robust safeguards enable societies to trust new technologies. International human rights law, he said, already provides a binding framework for addressing issues such as privacy, equality, non-discrimination and access to justice, and should guide AI governance rather than being treated as an afterthought.
He also stressed that human oversight must be meaningful rather than symbolic, with clearly identified individuals empowered to intervene or halt AI systems when necessary. Summarising his vision for responsible innovation, Türk contrasted the technology industry’s pursuit of ‘bigger, faster, better’ with what he described as a more appropriate goal: ‘smarter, kinder, wiser.’
Women and children bear disproportionate AI risks
The first panel focused on how AI is amplifying existing inequalities, particularly for women, children and other vulnerable groups.
UN Women Executive Director Sima Bahous presented evidence showing that 44% of assessed AI systems exhibit gender bias, while up to 99% of online sexual deepfakes target women. She also noted that women remain significantly underrepresented in AI development, with only a minority of national AI strategies explicitly addressing gender equality.
Bahous argued that governments remain the primary duty bearers under international human rights law and called for mandatory human rights impact assessments before and after AI deployment, alongside the meaningful participation of women, indigenous communities, disability advocates and civil society in AI governance.
Sonia Livingstone, a member of the Independent International Scientific Panel on AI, highlighted growing evidence that AI-generated child sexual abuse material is increasing rapidly and warned that many AI companion systems currently fail basic child safety standards. Rather than excluding young people from digital technologies, she argued, policymakers should ensure that children’s rights to participation, education and expression remain protected while embedding safeguards into AI systems from the outset.
Agentic AI raises new accountability challenges
Speakers also warned that increasingly autonomous AI systems are exposing significant legal and governance gaps.
Morocco’s Minister Delegate Amal El Fallah Seghrouchni described agentic AI as one of the most important governance challenges of the coming decade. As AI systems increasingly rely on networks of autonomous agents making decisions without direct human instruction, identifying responsibility when something goes wrong becomes considerably more difficult.
She proposed several practical measures, including documenting the actions of AI agents throughout decision-making processes, ensuring that a clearly identifiable human remains responsible for AI-enabled public services, and guaranteeing timely avenues for redress when individuals are harmed.
Samuel Arias Arzeno, Judge of the Supreme Court of the Dominican Republic, similarly argued that governance only becomes meaningful when someone believes an AI system has violated their rights and seeks justice. Courts, he said, must remain central institutions for ensuring that AI-assisted decisions remain subject to human accountability.
Rights protections should not depend on geography
A recurring concern throughout the discussion was that meaningful human rights protections are often applied unevenly across different regions.
Digital Rights Foundation founder Nighat Dad argued that robust human rights due diligence is largely conducted only where legislation requires it, particularly in Europe, while identical AI systems may be deployed elsewhere without comparable safeguards. She described this as a structural choice rather than a capacity gap, creating what she called a ‘two-tier’ human rights regime.
Dad called for mandatory gender and child rights impact assessments before deployment, consistent due diligence obligations across all markets where AI systems operate, and repeated assessments whenever AI capabilities change significantly.
Alvitta Ottley, also a member of the Independent Scientific Panel on AI, highlighted what she described as an ‘evaluation mismatch’. Current AI assessments often measure technical performance such as speed and accuracy, she explained, while policymakers and societies are instead asking whether AI protects human rights, strengthens accountability and improves people’s lives. Closing this evidence gap will require interdisciplinary research and much stronger evaluation of AI’s long-term societal impacts.
UN Assistant Secretary-General for Youth Affairs Felipe Paullier added that young people remain among AI’s most active users and innovators, yet rarely participate in decisions shaping the technology’s future. He urged governments to create meaningful opportunities for youth participation within national AI governance frameworks.
Global South voices call for more inclusive governance
Audience interventions reinforced the need for AI governance that is genuinely inclusive rather than shaped primarily by a handful of countries and companies.
Brazil highlighted its Digital Statute for Children and Adolescents, which requires child protection measures to be incorporated from the design stage and restricts platform features that encourage excessive use. Poland pointed to the Council of Europe Framework Convention on AI as an important legally binding instrument placing AI within the broader framework of human rights, democracy and the rule of law, while the Republic of Korea presented its AI Basic Act, which requires human rights assessments for high-impact AI systems.
Civil society organisations called for stronger global action. Access Now urged governments to establish binding human rights safeguards and prohibit AI applications that pose unacceptable risks, while the Association for Progressive Communications argued that communities should be viewed as ‘the first mile, not the last mile’ of AI governance, emphasising that meaningful connectivity and local participation remain prerequisites for equitable AI development.
In the closing discussion, co-chair Linda Bonyo highlighted another overlooked barrier to inclusive governance: many Global South experts remain unable to participate in international discussions because of restrictive visa processes, illustrating that exclusion from AI governance can begin long before negotiations start.
Closing the session, Spain’s Minister for Digital Transformation and Public Service Óscar López Águeda acknowledged that governments are already behind the pace of technological change but insisted the direction ahead is clear. AI governance, he argued, is ultimately about defending democracy, human dignity and human agency, ensuring that AI helps societies become better rather than simply more technologically advanced.
Track all key moments from the Global Dialogue on AI Governance inaugural meeting on our dedicated page.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!
