Fab Campaigns Ltd

Success requires more than consensus on principles. The world already has principles. What it lacks is architecture — mechanisms that make ethical progress irreversible. I propose the Dialogue succeeds if it produces three durable outcomes: First, a ratchet commitment. Governance frameworks must be designed so that minimum standards can only rise, never retreat. Voluntary codes have failed because they allow backsliding under competitive pressure. A successful Dialogue would establish that once a protection is agreed — for children, for marginalised groups, for democratic integrity — it becomes a floor, not a ceiling. The ratchet only turns one way. Second, structural accountability rather than content accountability. Current governance debates focus on outputs: harmful content, biased decisions, discriminatory outcomes. But these are symptoms. The Dialogue should produce agreement that architectural choices — how engagement accumulates, how influence decays, how past behaviour determines future access — are themselves subject to governance. Harm is designed in. It must be designed out. Third, participation parity. The communities most harmed by AI systems — those targeted by scapegoating algorithms, those in low-resource linguistic environments, those in jurisdictions without regulatory capacity — are systematically underrepresented in technical governance. A successful Dialogue would establish participation structures that correct this asymmetry, not merely acknowledge it. Capacity-building is not charity; it is the precondition for legitimacy. Measured against these three outcomes: a Dialogue that produces a new voluntary framework of aspirational norms is a missed opportunity. A Dialogue that produces binding minimum standards, structural accountability obligations, and genuine participation infrastructure would mark a genuine turning point. The bar for success should be: could a bad actor look at what was agreed and see no constraint on their behaviour? If yes, the Dialogue has failed.

• Safe, secure and trustworthy AI
• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Interoperability of governance approaches

These four themes form a coherent system rather than a menu of options - which is precisely why I have selected them together. Safe, secure and trustworthy AI is the necessary foundation. But "trustworthy" must mean more than transparency disclosures and audit trails. It requires that systems are designed to resist weaponisation - not just at the content layer, but at the architectural layer. A recommendation algorithm can be fully disclosed and still be engineered to amplify division. Trust requires architectural accountability. AI capacity-building is the legitimacy prerequisite. Governance agreed without genuine participation from the Global South, from low-resource language communities, from civil society in jurisdictions without AI regulatory infrastructure, is governance for the powerful by the powerful. Capacity-building is not an add-on; it is what makes the entire process credible. Social, economic, ethical, cultural, linguistic and technical implications is the broadest theme and the most important corrective to technocratic framings. AI governance is too often treated as a technical problem. The harms are social, economic and cultural. The solutions must be too. This theme insists on breadth of impact assessment rather than narrow safety benchmarking. Interoperability of governance approaches matters because fragmentation is itself a governance failure. If strong frameworks in one jurisdiction simply push harmful deployments to another, nothing structural has changed. Interoperability - not harmonisation, but mutual recognition and coherence - is the mechanism by which local progress becomes global progress. Taken together, these four themes address: what we are building (trustworthy), who can participate (capacity), what effects we must account for (implications), and how local governance connects globally (interoperability). They are mutually reinforcing. Selecting fewer would leave the framework incomplete.

Two issues are substantially absent from the listed themes and deserve dedicated treatment. Architectural harm as a distinct governance category. Current frameworks address AI harms as content problems - harmful outputs to be detected, removed or labelled. But the most consequential harms are architectural: built into how influence accumulates, how engagement metrics distort incentive structures, how past behaviour forecloses future optionality. A recommendation system can be fully transparent, technically safe, and architecturally engineered to amplify scapegoating. The Dialogue needs a governance category for design choices that cause systemic harm independent of any specific output. Without this, governance will always be reactive - a moderation layer bolted onto an architecture that was never designed for the public good. Children's digital rights as a first-order governance priority, not a sub-topic. Children are the cohort most profoundly shaped by AI-mediated social environments, and the least represented in governance processes. The risks - algorithmic manipulation of developing identity, data extraction before meaningful consent is possible, engagement architectures optimised for dependency - are not edge cases. They are the design. Yet children's digital rights appear nowhere as a standalone theme. Existing frameworks (GDPR, COPPA, UK Children's Code) provide a starting point but operate nationally in a global system. The Dialogue should establish minimum global standards for AI systems that interact with minors, with particular attention to consent architecture, algorithmic transparency for parents and guardians, and the prohibition of engagement-maximisation systems directed at children. Both issues share a common thread: they require governance at the level of system design, not system outputs. The ethical ratchet must be embedded in the architecture - not applied to what the architecture produces.

I write from two positions simultaneously: as a practitioner based in the United Kingdom, and as a founder building environmental intelligence infrastructure in the Alpine territories of northern Italy. The governance gaps I observe are not abstract — they are shaping operational decisions now. In the UK, the post-Brexit divergence from the EU AI Act has created genuine regulatory ambiguity for cross-border ventures. The UK's principles-based, sector-led approach offers flexibility but provides insufficient certainty for small ventures and civil society actors seeking to build trustworthy AI systems. Without interoperability between the UK framework and EU regulation, ventures operating across jurisdictions face duplicated compliance costs that larger incumbents absorb and smaller innovators cannot. This is a structural barrier to ethical AI development, not a technical one. In the Alpine region, the governance gap manifests differently. Environmental intelligence — using satellite, sensor and geospatial data to provide operational certainty to mountain territories — is an emerging sector where the regulatory framework is essentially absent. AI systems informing land management, climate risk assessment and infrastructure decisions in fragile ecosystems have no dedicated governance standard. The EU AI Act's risk classification does not adequately address environmental decision-support systems. Meanwhile, the communities most affected — small municipalities, mountain cooperatives, territorial bodies — have no capacity to participate in governance processes that will ultimately determine what data about their territory can be extracted, by whom, and on what terms. Across both contexts, the most significant opportunity is also the most neglected: small actors with genuine domain expertise and community embeddedness are precisely the partners needed to make AI governance work in practice. The gap is not only in regulation — it is in participation architecture. Governance processes that reach only large institutions and nation-states will produce frameworks that serve them.

The Dialogue can play a role that no treaty process can: it can function as a ratchet — a mechanism that raises the floor of practitioner knowledge, embeds that knowledge into governance structures, and makes it difficult for institutions to retreat to comfortable ignorance. I offer a concrete reference point. I recently participated in a senior NED peer group — a structured conversation among non-executive directors about what boards actually understand about AI and data systems. The most striking finding was not that board members lacked knowledge. It was that they lacked the right questions. They were receiving reports they could not interrogate, and making decisions based on assertions they had no framework to test. Out of that experience I developed a ten-question framework for board-level AI and cyber governance — deliberately simple, deliberately interrogative, deliberately non-technical. The point was not expertise. The point was structured accountability: giving practitioners the tools to ask whether assurance is genuine or performative. The AI Dialogue can scale this logic internationally. Not by producing another set of principles that boards will file and forget, but by producing interrogation frameworks — structured questions that senior practitioners in every jurisdiction can bring into the rooms where AI systems are actually being deployed. Governance does not happen in intergovernmental meeting rooms. It happens in audit committees, in procurement decisions, in board papers that either do or do not ask the right questions. The Dialogue's specific opportunity is to connect international standard-setting with practitioner networks — the NEDs, the civil society leaders, the cooperative boards, the municipal councils — who sit between regulation and deployment. These practitioners are not waiting for perfect frameworks. They are making decisions now. The Dialogue should reach them. International cooperation advances when the same questions are being asked in every room.

The most valuable initiatives the Dialogue should connect with are not the ones with the largest budgets. They are the ones where practitioners are already doing the thinking — quietly, in public, without waiting for permission. Three existing mechanisms deserve attention. The Signal Foundation's open-protocol model demonstrates that architectural transparency is possible at scale and that ethical design choices can be embedded structurally rather than added as governance afterthoughts. The EU AI Act's risk-classification framework provides a starting legislative vocabulary, however imperfect. And the emerging NED and senior practitioner communities — peer networks of board-level leaders who are already asking governance questions inside real institutions — represent an underutilised civil society infrastructure for AI accountability. The added value the Dialogue could bring is connection: between these practitioner networks and the international standard-setting process. Not as consultation exercises, but as ongoing learning journeys — structured processes where practitioners and policymakers develop understanding together, in public, over time. I have been building this kind of thinking publicly through my Ideas series at fabrizio.deliberali.com. "The Empty Space We Need to Create" (fabrizio.deliberali.com/the-empty-space-we-need-to-create) explores why breaking platform dependency requires architectural intervention, not just regulation. A forthcoming piece, the Ethical Ratchet Framework, proposes a governance architecture derived from cryptographic protocol design — one that makes ethical progress structurally irreversible. These are not academic papers. They are practitioner thinking, written to be interrogated and built upon. The Dialogue's added value would be to create the conditions where this kind of work — distributed, cross-jurisdictional, practitioner-led — feeds into governance design rather than remaining on the margins of it. I would welcome the opportunity to contribute directly: through a practitioner session, a public working group, or a learning journey that grows alongside the Dialogue itself.

The question of who contributes matters less than the question of how. Most multi-stakeholder processes fail not because the wrong people are in the room, but because the structure rewards performance over presence — polished position papers over genuine exchange, statements over questions, representation over accountability. Three structural recommendations follow directly from what works in practice. Replace submissions with interrogation rounds. The most generative governance conversations I have participated in — including structured NED peer sessions — share a common design: questions are circulated before the meeting, responses are brief, and the majority of time is spent in structured challenge. The Dialogue should adopt this format. Each thematic session should open with ten practitioner-sourced questions and spend the majority of its time on responses, not presentations. This is Embedding Over Advising in governance form: you learn more by being inside the problem than by observing it from outside. Create a tiered contribution architecture. Not all stakeholders have the same capacity to contribute, and pretending otherwise produces either tokenism or paralysis. A Stage Gate approach works: an open consultation layer (written submissions, broad participation), a practitioner layer (structured peer sessions with accountability for follow-through), and a working group layer (small, mandated, time-bound, publicly reported). Each gate should filter for quality of engagement, not institutional status. Build in a learning journey, not a conference cycle. A single dialogue event produces a communiqué. A structured multi-session process — with published interim findings, public iteration, and named practitioners who evolve their positions over time — produces genuine governance development. The Dialogue should commission a small cohort of practitioners to develop their thinking publicly across the process, demonstrating that engagement changes understanding. Governance built on partnership logic — few but good, embedded not advisory, present not performing — produces better outcomes than governance built on representation logic alone.

The most underrepresented voices in AI governance are not marginal. They are the majority — and their absence is structural, not accidental. Territorial communities without regulatory capacity. I work across Alpine municipalities in northern Italy — small territorial bodies managing land, water, forests, and climate risk. AI systems are already informing decisions that affect these communities: land classification, environmental monitoring, infrastructure planning. Yet no one in these communities sits in any governance process. They are not passive victims of distant technology; they are active stewards of complex ecosystems who have developed sophisticated local knowledge over generations. That knowledge is invisible to AI governance. Including them requires deliberate outreach to sub-national bodies — regions, provinces, cooperatives, mountain communities — not just nation-states. Children and young people. The generation most shaped by AI-mediated environments has no formal voice in their governance. Youth panels that produce statements are not sufficient. The Dialogue should commission structured participation processes — designed with developmental appropriateness and proper facilitation — that allow young people to interrogate the systems being built around them, not merely react to adult framings of the problem. Non-English-speaking practitioner communities. Global AI governance discourse is conducted almost entirely in English, which systematically excludes the practitioners, civil society actors, and researchers whose work is developed in other languages. The Dialogue should fund simultaneous interpretation and active multilingual synthesis — not as courtesy, but as epistemic necessity. A governance framework that cannot be interrogated in Italian, Arabic, or Swahili is not a global framework. Caregivers and educators. The people closest to AI's daily human impact — parents, teachers, social workers — are almost never present in governance spaces. Their practical knowledge of how AI systems affect development, learning, and family life is irreplaceable. Inclusion requires structural redesign, not open invitation. The door being unlocked is not the same as the door being open.

The least effective format for a dialogue about AI governance is a conference. Keynotes produce performance. Panel discussions produce positioning. Open floor Q&A produces the loudest voices, not the most important ones. The Dialogue's ambition requires formats that are structurally designed to produce genuine exchange — and genuine exchange requires skin in the game. Three formats that work in practice. The Pre-Committed Question Round. Circulate ten practitioner-sourced questions one week before each session. Participants must submit brief written responses in advance — not position statements, but direct answers. The session itself is then spent interrogating those responses in small groups of six to eight, with a named rapporteur who synthesises findings publicly. This is the format that works in senior NED peer communities precisely because it makes preparation mandatory and performance difficult. You cannot bluff your way through a question you answered in writing the previous week. The Ratchet Session. Each thematic session ends not with conclusions but with commitments: specific, named, time-bound actions that participants agree to report back on at the next session. These commitments are published. The ratchet only turns forward — every session begins by accounting for what was promised in the last one. This creates continuity across what would otherwise be disconnected events, and makes absence costly. The Practitioner Learning Cohort. Select a small group — twelve to fifteen practitioners from different sectors, jurisdictions, and linguistic communities — who commit to the full Dialogue process. They publish their evolving thinking between sessions, respond to each other publicly, and present their development arc at the closing session. Not experts delivering verdicts, but learners demonstrating that engagement changes understanding. Presence Over Performance is not a cultural preference. In governance design, it is a structural choice. Build formats where showing up fully is the only way to contribute meaningfully.

Effective AI governance already exists. It is not waiting to be invented - it is waiting to be recognised, connected, and scaled. Three examples from different layers of practice illustrate what works and why. The Signal Protocol as architectural governance. The most durable ethical AI design decision I know of was not a policy. It was a technical choice: Moxie Marlinspike and Trevor Perrin designed the Double Ratchet Algorithm to assume compromise and recover forward. End-to-end encryption, open-source code, nonprofit structure - each decision made ethical behaviour structurally irreversible rather than voluntarily maintained. This is the model the Dialogue should propagate: governance embedded in architecture, not appended to it. The Signal Foundation demonstrates that values can be load-bearing. The UK Cyber Governance Code of Practice as interrogation framework. The Code works not because it mandates specific technical controls but because it gives board-level practitioners the right questions to ask. My own ten-question board framework, developed from the Code and tested in senior NED peer sessions, demonstrates that accessible interrogation tools change governance behaviour faster than compliance requirements alone. The Dialogue should commission equivalent frameworks for AI governance - designed for non-technical board members, municipal councils, and cooperative leadership - and distribute them freely. Smart Mountains as territorial AI governance in practice. The environmental intelligence infrastructure I am building across Alpine territories embeds a specific governance principle: the communities whose territory generates the data must retain sovereignty over it. We do not sell data. We sell operational certainty. This model - where AI creates value for the territory it observes rather than extracting value from it - is replicable across any data-rich environment where communities currently have no stake in the systems built on their knowledge. Governance that cannot point to working examples is not governance. It is aspiration.