Centre for Protecting Women Online

A successful first Global Dialogue must establish a shared, actionable understanding of how AI systems are already shaping experiences of harm, including technology-facilitated violence against women and girls (TFVAWG). This requires recognising that AI does not operate in a vacuum but reflects and amplifies existing power structures, including gender inequality and intersecting forms of discrimination, and should result in a clear action pathway with agreed reporting and accountability mechanisms. The Dialogue should centre the lived experiences of women and girls, alongside those of minoritised groups, to ensure that governance responses are grounded in the realities of how harm is produced and experienced. This includes acknowledging compounding inequalities and the ways in which AI systems can intensify abuse at scale and embedding mechanisms to ensure these perspectives are structurally integrated into decision-making processes (e.g. through representative advisory structures, see Question 16). A key outcome should be agreement on clear structures of responsibility and accountability across States, companies, and developers. At present, there is a significant gap in who is responsible for preventing, mitigating, and remedying harms such as AI-generated abuse, including deepfake sexual content and the amplification of online harassment. The Dialogue must also address the structural drivers of harm. These include monetisation models that incentivise harmful content, the use of biased or non-consensual training data, and the reinforcement of harmful social and cultural norms, including misogyny. Finally, success would include recognition of the global inequalities underpinning AI systems, including the exploitation of labour in the Global South within AI supply chains. Addressing these issues at the outset will be critical to ensuring that AI governance is both effective and equitable.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

Understanding the social, economic, ethical, and cultural implications of AI must come first. These dimensions are essential to identifying how harms such as TFVAWG are produced, including through the amplification of misogynistic norms, biased or non-consensual training data, and unequal power relations. Without this foundation, governance responses risk being reactive and narrowly focused on symptoms. Transparency, accountability, and human oversight are critical to ensuring that harms can be identified, challenged, and remedied. At present, there is a lack of clear responsibility across States, companies, and developers, allowing harmful systems to persist without effective redress for victims and survivors. For oversight to be meaningful, those responsible must have the necessary expertise, including mandatory training on ethical and social risks, and must work alongside independent specialists. Safe, secure and trustworthy AI, alongside the protection and promotion of human rights, must be operationalised in practice, not treated as high-level principles. Safety-by-design must include enforceable requirements such as risk assessments, safeguards against misuse, and accountability mechanisms across the AI lifecycle. Current approaches often prioritise increasing model capability over safety, which is particularly concerning in the context of TFVAWG, where increasingly sophisticated systems enable more scalable forms of abuse. Emerging evidence also highlights structural inequalities in AI's impact. Our recent research (Fernandez, Miriam et al., 2026, Assessing the Impact of Artificial Intelligence on Gender Disparities in the Labour Market, Centre for Protecting Women Online, The Open University) finds that female-dominated occupations show greater exposure to large language models, while women are less likely to adopt AI tools (Aldasoro et al., 2024; Stephany & Duszynski, 2026). This underscores the need for targeted capacity-building and upskilling programmes to support adaptation rather than displacement. Addressing these areas is essential to achieving meaningful human rights outcomes in AI governance.

A key cross-cutting issue is the political economy of AI, particularly the role of business models in driving harm. Monetisation strategies that prioritise engagement and scale incentivise the amplification of harmful content, including abuse targeting women and girls. Addressing TFVAWG therefore requires engagement with these underlying economic drivers, not only downstream content moderation. The sourcing and use of training data is another critical issue. AI systems are frequently trained on datasets that include biased, harmful, or non-consensually obtained material, including content depicting or enabling abuse. This directly contributes to discriminatory and harmful outputs. Labour exploitation within AI supply chains also remains insufficiently addressed. Many systems rely on low-paid workers, often in the Global South, to undertake data annotation and content moderation. These workers are frequently exposed to traumatic material, including violent and abusive content, with limited protections or support. AI systems also reinforce existing social and cultural norms, including misogyny and gender inequality. Governance approaches must therefore address structural power dynamics, rather than treating harms as isolated or purely technical issues. A further emerging issue is the governance of open-source AI, open data, and open model ecosystems. The democratisation of artificial intelligence through open data is changing the technological landscape and making AI more widely accessible (European Data Portal, 2025, Democratisation of AI through open data: Empowering innovation). While openness can support transparency, collaboration, and capacity-building, it can also lower the barriers to harmful reuse, including the repurposing of models and data for harassment, deepfakes, and other non-consensual or discriminatory applications. We have already seen how accessible tools can accelerate the spread of deepfake-based abuse when safeguards are insufficient. As such, open-source AI requires careful governance, including documentation, safety testing, and accountability measures to reduce the risk of harmful reuse. Finally, there is a need to prioritise the longer-term societal impacts of AI systems, which remain underexplored compared to short-term safety concerns.

Governance gaps are contributing to the rapid expansion of AI-enabled harms, including TFVAWG. Tools that enable the creation of synthetic non-consensual content, harassment at scale, and the targeting of individuals are becoming increasingly accessible, while accountability mechanisms remain weak or absent. A lack of transparency from developers and platforms makes it difficult to understand how systems are trained and deployed, limiting the ability to challenge harmful outcomes. At the same time, unclear or fragmented regulatory frameworks impact companies responsibility for addressing harm, leaving victims and survivors without effective avenues for redress, especially where harms have not been captured in criminal law. However, there are also opportunities. AI systems have the potential to support access to information, services, and innovation, including in addressing gender-based violence in some appropriate circumstances. Realising these benefits requires governance frameworks that prioritise safety, accountability, and inclusion from the outset.

The AI Dialogue can play a critical role in addressing harms that operate across borders such as TFVAWG, which cannot be effectively governed at the national level alone. AI-enabled abuse often operates across jurisdictions, exploiting regulatory gaps and inconsistencies. The Dialogue can support the development of shared norms and coordinated approaches to these harms, helping to ensure that protections are not limited by geography. This includes facilitating agreement on baseline standards for safety, transparency, and accountability. It can also enable the exchange of knowledge and expertise, particularly by amplifying the insights of civil society organisations working directly with affected communities. This is essential to ensuring that governance responses are grounded in lived experience and so that they directly address harms being experienced and those we can predict from developing technologies. Importantly, the Dialogue can help align existing initiatives and reduce fragmentation, while promoting interoperability across governance frameworks. It also has a role to play in encouraging measurable commitments from States and companies, particularly in relation to preventing and responding to TFVAWG.

The AI Dialogue should build on existing initiatives addressing TFVAWG, particularly those that already operate across borders and sectors. For example, UN Women has developed global frameworks on online violence and digital safety, including work on gender-responsive approaches to technology governance. The UNESCO Recommendation on the Ethics of AI provides an internationally agreed framework that explicitly recognises gender equality and the need to address harm. Multi-stakeholder initiatives such as the Global Partnership on AI have also contributed to research and policy development on responsible AI, including work relevant to harmful content and governance gaps. At a regional level, the Digital Services Act demonstrates emerging approaches to platform accountability, including obligations to address systemic risks such as gender-based harm. Civil society-led initiatives, including those focused on ending violence against women and girls, provide critical expertise, data, and survivor-centred approaches that are often not reflected in technical or policy processes. The added value of the AI Dialogue lies in its ability to connect these efforts into a coherent global framework, elevate TFVAWG as a core governance issue, and ensure that lessons learned are shared across regions. It can also ensure that civil society expertise is meaningfully integrated into global decision-making processes.

The Dialogue must ensure meaningful participation from a diverse range of stakeholders, particularly civil society organisations working to address violence against women and girls. This requires moving beyond consultative approaches towards models of engagement where stakeholders can actively shape outcomes. Participation must be adequately resourced, including through financial support, to enable engagement from underrepresented groups and organisations. Accessible formats, including multilingual and hybrid participation options, are essential. The Dialogue should also incorporate deliberative and interactive formats that allow for in-depth engagement, rather than relying solely on formal statements. Crucially, mechanisms must be established to ensure that stakeholder input, particularly from those working directly with affected communities, is reflected in outcomes and decision-making processes.

Global discussions on AI governance continue to underrepresent voices from the Global South, young people, and marginalised communities, including those most affected by technology-facilitated violence against women and girls. Civil society organisations working on ending violence against women and girls, as well as survivor-led and grassroots groups, are particularly excluded from technical and policy spaces, despite their critical expertise. Recent work by the Centre for Protecting Women Online, through the Towards a Safer Web for Women (TSWW'25) workshop at the ACM Web Conference, highlights the importance of addressing these gaps. The workshop convened interdisciplinary experts across computer science, law, social sciences, and civil society to examine the systemic drivers of online violence against women and girls and identified key governance recommendations that emerged through convergence across disciplines. First, there is a need to embed anticipatory and interdisciplinary governance practices. This requires earlier and deeper collaboration between technologists, legal scholars, gender experts, and policymakers to anticipate emerging harms before they become widespread, rather than relying on reactive regulatory responses. Second, governance must move beyond representation alone to centre the diversity of womens lived experiences. Increasing participation is necessary but insufficient without ensuring that decision-making reflects the realities of those most affected. This requires participatory governance and co-design processes that meaningfully involve women from marginalised and underrepresented groups across technology design, regulation, and enforcement. In practice, this could include institutionalising permanent, representative advisory structures (such as advisory boards convened by regulators and supported through mandatory industry levies) that are embedded throughout the technology lifecycle. These structures can help identify blind spots, ensure continuous input, and structurally integrate diverse perspectives into both policy and design, rather than treating them as a reactive consideration. Addressing underrepresentation therefore requires not only inclusion, but the redistribution of power within AI governance processes.

Innovative engagement formats should prioritise depth, inclusivity, and real-world relevance. This could include scenario-based exercises that explore how AI systems enable harms such as TFVAWG, as well as participatory workshops that bring together policymakers, technologists, researchers and civil society. Incorporating survivor testimonies and lived-experience perspectives, in a safe, appropriate and trauma informed way with specialists, is particularly important in grounding discussions in the realities of harm and ensuring that policy responses are effective.

Effective AI governance approaches are those that embed safety, accountability, and human rights from the design stage, rather than as reactive measures. This includes conducting impact assessments that explicitly consider risks related to gender-based violence, as well as implementing transparency measures around data use, system design, and deployment. Multi-stakeholder collaboration is essential, particularly involving civil society and organisations working on violence against women and girls in both policy development and oversight. Policies must also address the root causes of harm, including business models that incentivise harmful content and the use of biased or harmful datasets. Finally, governance approaches should include clear accountability mechanisms and access to remedy for those affected by AI enabled harms, ensuring that victims and survivors are supported and that responsible actors are held to account.