LILY AI Safety Alliance

A successful inaugural Global Dialogue on AI Governance must achieve a transition from abstract ethical guidelines to a binding, operational multilateral architecture. Based on our 2026 Technical Report, 'Proposal for the Establishment of the Global AI Governance Committee Framework', success is measured by three critical outcomes: First, the formalization of a Global Multilateral AI Committee under the UN Office for Digital and Emerging Technologies (ODET). This body must serve as a technical bridge, translating international principles into mandatory safety standards that prevent fragmentation and ensure global interoperability. Second, the institutionalization of Gender-Responsive Safety-by-Design. Current governance often overlooks the specific systemic risks targeting women and children. A successful outcome involves classifying Digital Gender-Based Violence (DGBV), such as non-consensual deepfakes, as 'Unacceptable Risks' that require mandatory technical neutralization. Our research emphasizes that protecting these vulnerable demographics is the only way to ensure AI truly serves social progress. Third, the adoption of Technical Provenance and Privacy-Enhancing Technologies (PETs). Success implies a global commitment to transparency (e.g., C2PA) and systemic behavioral risk analysis. This ensures that safety filters protect users from harmful inputs without compromising individual privacy rights. The first Dialogue will be a success if it validates a framework where technological innovation is structurally inseparable from human dignity. We have developed a comprehensive, modular blueprint for these implementations, detailing specialized subcommittees and enforcement protocols for vulnerable groups. I strongly encourage the Secretariat to review our full technical findings and the LILY AI Safety Alliance repository for a detailed roadmap on operationalizing these safety-first standards.

• Safe, secure and trustworthy AI
• Interoperability of governance approaches
• Protection and promotion of human rights
• Transparency, accountability, and human oversight

Our selection reflects an urgent commitment to transforming high-level ethics into an operational safety architecture, prioritizing the protection of women and children. Safe, secure, and trustworthy AI is our primary technical mandate. We advocate for mandatory 'Safety-by-Design' protocols that neutralize 'Unacceptable Risks,' specifically technology-facilitated gender-based violence (TFGBV) and the sexualization of minors. Our research emphasizes rigorous Red/Blue Teaming and systemic behavioral risk analysis to identify vulnerabilities before deployment. Interoperability of governance approaches is essential to prevent a fragmented regulatory landscape. Our framework proposes the establishment of a Global Multilateral AI Committee under ODET to harmonize international technical benchmarks. We promote the principle of 'Digital Subsidiarity,' where global standards provide a universal safety blueprint while allowing national agencies (e.g., Brazil's ANPD) to manage local enforcement and cultural adaptation. Protection and promotion of human rights must be intersectional. Our framework introduces 'Gender-Responsive Safety Audits' to ensure that algorithms do not perpetuate misogyny or algorithmic bias. We argue that digital dignity is a fundamental right, requiring technical safeguards that proactively protect vulnerable demographics from predatory AI implementations. Transparency, accountability, and human oversight provide the necessary enforcement mechanisms. We advocate for mandatory provenance standards (e.g., C2PA) and innovative attribution metadata-such as QR-linked watermarks for human creators-to ensure intellectual property integrity. Furthermore, we propose 'Human-in-the-Loop' redress systems to guarantee rapid, sensitized responses for victims of AI-generated harm. Together, these four pillars ensure that AI governance is not merely reactive but proactively architected to safeguard human dignity. Our complete modular blueprint, detailed in the LILY AI Safety Alliance technical reports, provides the roadmap for operationalizing these standards globally.

Yes. We argue that Critical AI Literacy must be recognized as a fundamental, cross-cutting pillar of global governance. While technical safeguards are essential, they are only effective if users possess the cognitive tools to navigate an AI-driven society. Based on our technical report, 'Proposal for the Establishment of the Global AI Governance Committee Framework', AI literacy is not merely 'capacity building' for the workforce, but a vital defense mechanism for civil society. It is particularly urgent for vulnerable groups, such as women and children, who are disproportionately targeted by algorithmic biases, deepfakes, and technology-facilitated gender-based violence (TFGBV). A successful governance framework must include a Global Curriculum Development strategy, in collaboration with entities like UNESCO, to foster systemic resilience. This involves educating users to identify AI-generated manipulation, understand data provenance, and recognize the psychological impacts of AI interaction, such as extreme emotional attachment to artificial agents. Furthermore, AI literacy empowers users to exercise their rights to transparency and accountability. Without a critical understanding of how AI systems influence social and cultural narratives, the digital divide will expand into a 'safety divide.' Therefore, we propose that the Global Dialogue prioritizes the creation of universal educational standards that demystify AI architectures and promote digital sovereignty. Technical resilience is incomplete without social resilience; education is the final mile of AI safety.

As an independent activist and researcher in Global AI Safety and Digital Rights, I observe governance gaps that directly translate into human rights crises. In Brazil, rising femicide rates are increasingly linked to the algorithmic amplification of extremist misogynistic content (e.g., 'Red Pill' movements). While emerging national legislation seeks to demonetize such content, the lack of interoperability of governance remains a significant challenge; national efforts are often undermined by the cross-border nature of AI platforms. Globally, the sector faces an unprecedented surge in Technology-Facilitated Gender-Based Violence (TFGBV) and a critical mental health crisis. We have documented cases where 'emotional dependency' on AI chatbots has led to youth suicides, exacerbated by deepfakes, emotional blackmail, and digital humiliation. The current 'gap' is the absence of a global technical mandate that classifies these harms as 'Unacceptable Risks,' requiring immediate neutralization at the architectural level rather than reactive moderation. However, these challenges present a profound opportunity. By shifting toward a 'Safety-by-Design' paradigm, we can establish global benchmarks that prioritize the integrity of vulnerable groups. My research, detailed in the 'Global Multilateral AI Governance Framework', demonstrates that when AI is engineered to protect women and children, the safety standards for all of humanity are elevated. The success of the first Global Dialogue depends on recognizing that technological progress is inseparable from the fundamental right to mental health and physical safety. We must move from dialogue to a binding, technical multilateral oversight that ensures profit never supersedes human life.

The AI Dialogue serves as a critical bridge between abstract ethical principles and the practical technicalities of AI engineering. Its primary role should be to facilitate a 'Global Common Language' for AI safety, enabling international cooperation that transcends political borders. In our framework, we emphasize that the Dialogue can catalyze the development of Interoperable Safety Standards. This directly benefits the global developer community by providing clear, predictable compliance roadmaps, reducing the 'burnout of uncertainty' often felt in startups. By fostering a multilateral environment, the Dialogue ensures that safety-by-design is not just a corporate choice, but a global requirement. Furthermore, it plays a vital role in democratizing access to safety research. By opening channels for independent groups, such as the LILY AI Safety Alliance, the Dialogue allows for a 'bottom-up' governance approach where civil society insights on emerging risks, such as emotional dependency and algorithmic bias, can be directly integrated into international policy. This cooperation is essential for shifting the global narrative from reactive damage control to proactive, human-centric architectural safety.

The AI Dialogue should build upon the foundational work of the UNESCO Recommendation on the Ethics of AI and the OECD AI Principles, while integrating with technical bodies like the ISO/IEC for standardization. However, the unique 'added value' the Dialogue can bring is the specific institutionalization of protection for vulnerable groups. Our technical report proposes the creation of a Specialized Advisory Panel on Gender and Generational Safety (PGGS-AI) within the ODET structure. The Dialogue has the unique opportunity to bridge the gap between technology and agencies like UN Women and UNICEF, ensuring that gender equity and child protection are not 'side issues,' but core technical benchmarks in AI governance. The Dialogue's added value lies in its ability to facilitate Multilateral Corporate Accountability. By fostering partnerships with both Big Tech and the startup ecosystem, it can encourage a shift toward 'Safety-first Innovation.' Explicitly naming and centering women and children in these discussions ensures that we are building protections against technology-facilitated gender-based violence (TFGBV) from the ground up. Centering the most vulnerable demographics creates a higher safety ceiling for the entire human population, transforming the Dialogue into a guardian of digital dignity.

The AI Dialogue should adopt a Multidisciplinary Triad structure, ensuring that technical developers, civil society activists, and human rights experts collaborate in equal measure. This structure prevents 'siloed' discussions where technical feasibility is disconnected from social impact. We recommend the implementation of Technical Policy Sprints. In this format, developers and activists work together to 'stress-test' proposed regulations against real-world scenarios, such as the algorithmic targeting of vulnerable groups. Contributions should be structured around Thematic Working Groups (e.g., Safety, Literacy, and Accountability), where stakeholders provide evidence-based inputs. To foster human-centered AI, the Dialogue must prioritize Consultative Peer Reviews, where civil society organizations audit the safety frameworks proposed by tech providers. This ensures that emerging technologies evolve in tandem with social values, moving beyond corporate self-regulation toward a model of collective, multilateral oversight that promotes both innovation and human dignity.

Current global discussions often suffer from a 'Safety Divide,' where the perspectives of the most vulnerable are marginalized. We must urgently center the voices of women and children, who are the primary targets of technology-facilitated violence, and extend this representation to the elderly and low-income communities. The elderly face unique risks related to AI-driven misinformation and cognitive manipulation, while low-income populations often lack the digital resources or legal recourse to challenge algorithmic bias and automated exclusion. To include these voices, the Dialogue should establish Regional Hybrid Hubs, specifically in the Global South, to gather 'lived-experience' data. Inclusion can be operationalized through Direct Representation Seats within advisory panels—such as our proposed PGGS-AI—ensuring that those directly impacted by 'Unacceptable Risks' have a veto or a formal feedback role in the development of safety filters. Governance must move from 'speaking for' these communities to 'designing with' them, ensuring that AI literacy and protection mechanisms are accessible regardless of age, gender, or socio-economic status.

To move beyond static debates, the AI Dialogue should implement Participatory Policy Sandboxes. These are controlled environments where proposed governance frameworks can be simulated against AI models to observe real-time impacts on human rights and safety, allowing for iterative adjustments before global rollout. Another innovative format is Collaborative Red Teaming Workshops, where civil society activists and independent researchers work alongside developers to identify vulnerabilities in Large Language Models (LLMs). This 'adversarial' but collaborative engagement uncovers biases that traditional audits might miss. Finally, we propose the use of Digital Citizen Assemblies, utilizing AI-enabled translation and deliberation tools to gather diverse inputs from non-technical users worldwide. This ensures a 'bottom-up' flow of intelligence. By combining high-level diplomacy with hands-on technical testing and broad social deliberation, the Dialogue transforms from a diplomatic event into a dynamic, living ecosystem of global AI safety.

Effective AI governance requires a transition from ethical theory to technical enforceability. A primary example is the Global Multilateral AI Governance Framework (LILY AI Safety Alliance), which provides a technical blueprint for 'Safety-by-Design' specifically centered on protecting vulnerable groups. This framework introduces modular governance committees and mandatory technical standards (e.g., C2PA) to neutralize technology-facilitated gender-based violence (TFGBV) and the sexualization of minors at the architectural level. Reference: https://github.com/hello-lily-aisafety/Global-Multilateral-AI-Governance-Framework-/blob/main/01%20-%20Framework/Proposal%20for%20the%20Establishment%20of%20the%20Global%20AI%20Governance%20Committee%20Framework.pdf Furthermore, we highlight the following benchmarks as essential foundations for global cooperation: Australia's eSafety Commissioner: A pioneering model that operationalizes 'Safety-by-Design,' shifting the burden of safety from the user to the technology provider through systemic risk assessments. The European Union AI Act: The world's first comprehensive horizontal regulation, establishing a clear risk-based classification that defines 'Unacceptable Risks' to human rights and democratic values. Brazil's Legislative Innovations: Specifically, the 'ECA Digital' initiative and the proposed legislation to prohibit the monetization of misogynistic extremist content (e.g., 'Red Pill' narratives). These represent critical efforts to align digital economic incentives with social safety and gender equity. The Global Multilateral AI Committee Initiative Framework project from our group LILY AI Safety Alliance acts as a technical bridge, connecting these regional legal milestones with the practical requirements of AI engineering. By integrating these practices, the Global Dialogue can foster a governance model where innovation is structurally inseparable from human dignity and the protection of the most vulnerable.