Anna August Advisory

Success would not be measured by the number of declarations signed or frameworks produced. The AI governance space already has frameworks. What it lacks is convergence between the people who write them and the people who implement them. For me, a successful first Dialogue would achieve three things. First, it would produce a shared language. Regulators, technologists, ethicists and practitioners currently speak in parallel. A SHAP output means nothing to a policymaker. Article 14 means nothing to an engineer. A successful Dialogue creates genuine translation, not summary documents, but a working vocabulary that allows all parties to interrogate the same system with the same rigour. Second, it would surface the implementation gap honestly. Most AI governance conversations happen at the level of principle. The harder conversation is about evidence, what does documented accountability actually look like in a healthcare AI system, a credit scoring tool, a recruitment platform? A Dialogue that stays at principle level will produce another framework nobody operationalises. Third, it would include voices from jurisdictions building governance infrastructure now, not only those with mature regulatory regimes. The EU AI Act sets a global standard. But companies in the Gulf, in emerging markets, in post-Brexit UK are navigating compliance without the institutional support that EU-based organisations have. Their experience matters. I work at the intersection of regulation, technical systems and organisational practice across EU, UK and Gulf markets. What I would bring to this Dialogue is not a theoretical position but a grounded one, built from auditing real systems, translating requirements into evidence, and standing in rooms where all three worlds need to speak to each other.

• Transparency, accountability, and human oversight
• Protection and promotion of human rights
• Interoperability of governance approaches
• Social, economic, ethical, cultural, linguistic and technical implications of AI

My four priorities reflect where the gap between governance intent and operational reality is widest, and where my work is most directly engaged. Transparency, accountability, and human oversight is where most governance frameworks exist at the level of principle but fail at the level of evidence. My Seven Core Ethical Pillars framework was built specifically to operationalise these requirements across jurisdictions, turning regulatory obligations into auditable, documented proof. This is not a theoretical priority for me, it is the daily work. Protection and promotion of human rights follows directly. High-risk AI systems in employment, healthcare and financial services make decisions that determine access to opportunities. A Fundamental Rights Impact Assessment is not a compliance exercise -it is a recognition that behind every dataset is a person whose rights either are or are not protected. My audits are structured around this recognition. Interoperability of governances approaches is where I see the most urgent practical need. Organisations operating across EU, UK, Gulf and US markets are navigating four regulatory frameworks simultaneously with no common methodology. My framework is the only independent audit methodology that maps all four jurisdictions to a single structured baseline. The Dialogue needs to address this fragmentation directly. Social, economic and ethical implications of AI closes the loop. Governance that ignores the structural effects of AI on labour markets, on marginalised communities, on democratic participation is governance that protects systems rather than people. This has to remain central.

Two issues stand out as systematically underrepresented in the current thematic architecture. The first is what I would call the evidence gap. Current governance frameworks focus heavily on what organisations should do. They are largely silent on what proof of compliance actually looks like in practice. There is no shared standard for what constitutes sufficient evidence that a high-risk AI system is safe, fair and accountable. This creates a situation where organisations can be technically compliant on paper while remaining operationally exposed. The Dialogue needs to address not just principles and obligations, but the evidentiary standards that make those obligations verifiable by regulators, auditors and affected individuals. The second is jurisdictional fragmentation at the implementation level. The listed themes address interoperability of governance approaches at the policy level. But the operational reality for organisations deploying AI across multiple markets is more granular and more urgent. A company deploying a recruitment AI system in the EU, UK, Saudi Arabia and the United States simultaneously faces four distinct regulatory frameworks with different timelines, different documentation requirements and different enforcement mechanisms. No existing multilateral initiative addresses this at the level of practical implementation. The Dialogue has an opportunity to develop shared minimum standards for cross-jurisdictional AI compliance documentation that reduce duplication without compromising the integrity of any individual framework. Both issues share a common root: the distance between what governance frameworks demand in theory and what compliance looks like in practice. Closing that distance requires voices with direct operational experience of implementing governance across systems, sectors and jurisdictions. That is the perspective I would bring.

Anna August Advisory operates through registered entities in both the United Kingdom and Poland, bringing EU membership into the practice directly and serving clients across Europe. This dual positioning creates a particular vantage point on the governance gaps that neither purely UK-based nor purely EU-based practitioners experience in the same way. The most significant challenge in both markets is the implementation gap. In Poland and across Central and Eastern Europe more broadly, awareness of EU AI Act obligations among small and mid-size organisations is critically low. The regulation is perceived as a large-company problem. It is not. Any organisation deploying AI in recruitment, credit assessment, healthcare or education is subject to the same high-risk obligations as a multinational. The capacity to comply, however, is vastly unequal. In the UK, the challenge is different but equally urgent. Post-Brexit divergence means UK organisations entering EU markets carry UK compliance frameworks that do not map directly onto EU AI Act requirements. UK GDPR, ICO guidance and the Equality Act provide a strong data and anti-discrimination baseline. They do not require a Quality Management System, a Fundamental Rights Impact Assessment, Annex IV technical documentation or a conformity assessment pathway. UK companies assume their compliance transfers. It does not. The opportunity in both markets is the same: organisations that build governance infrastructure now, before August 2026, will have a structural advantage in procurement, investment and cross-border expansion. Governance is becoming a market differentiator, not just a regulatory obligation. My work sits precisely at this intersection, helping organisations in both markets understand not just what compliance requires, but what evidence of compliance actually looks like in practice.

The AI Dialogue has an opportunity that most multilateral initiatives miss: it can function as a bridge between the production of governance frameworks and their implementation in practice. Most international AI governance efforts operate at the level of principle. They produce declarations, taxonomies and recommendations that are technically sound but operationally incomplete. The gap between a UN resolution and a compliance decision made by a Chief Compliance Officer in Warsaw, Riyadh or London is enormous. The Dialogue can close that gap if it is designed to do so from the start. Three roles would make it uniquely valuable. First, the Dialogue can serve as a convergence mechanism for jurisdictional fragmentation. Organisations operating across the EU, UK, Gulf and US markets currently navigate four distinct regulatory regimes with no common methodology. The Dialogue is positioned to develop shared minimum standards for cross-border AI compliance documentation that reduce duplication without displacing national frameworks. Second, it can establish evidentiary benchmarks. What does documented accountability actually look like for a high-risk AI system? What constitutes sufficient evidence of human oversight, fairness testing or transparency? These questions remain unanswered in most current frameworks. The Dialogue can convene regulators, auditors and practitioners to define what proof of compliance means in practice, not just in principle. Third, it can amplify implementation capacity in underserved markets. Awareness of AI governance obligations among small and mid-size organisations across Europe, US, and the Gulf remains critically low. The Dialogue can create practical guidance that reaches beyond the organisations already at the table. The measure of success is not the quality of the document produced. It is whether the organisations most affected by AI governance gaps are better equipped to act.

Several existing mechanisms provide foundations the AI Dialogue should build upon rather than duplicate. The OECD AI Principles and Policy Observatory offer the most comprehensive cross-jurisdictional mapping of AI governance frameworks currently available. The Dialogue should connect directly with this infrastructure rather than create a parallel taxonomy. The Global Partnership on AI provides a multilateral research and policy function that complements rather than competes with what a UN-level Dialogue can achieve. At the regional level, the EU AI Act represents the most operationally detailed governance framework in existence. The Dialogue should treat it not as one regional approach but as the most advanced working model of binding AI governance in practice, while remaining open to the legitimate governance choices of other jurisdictions. The IEEE standards work on ethically aligned design and ISO/IEC 42001 provide technical baselines the Dialogue should reference when discussing evidentiary standards. The added value the AI Dialogue can bring is not more principles. It is four things none of the existing mechanisms have delivered. First, a structured process for translating principles into minimum evidentiary standards that regulators and auditors can apply consistently across jurisdictions. Second, a mechanism for organisations outside mature regulatory environments to access practical implementation guidance, not just high-level frameworks. Third, a forum where the distance between governance producers and governance implementers is deliberately closed. The people who write frameworks and the people who audit compliance against them rarely occupy the same room. The Dialogue can change that. Fourth, a shared methodology for organisations navigating multiple regulatory regimes simultaneously. AI has no borders, but governance does. Practitioners working across jurisdictions are developing cross-jurisdictional audit approaches out of necessity, not design. The Dialogue has the authority to elevate this into a systemic standard rather than leave it as an exception.

The AI Dialogue will only deliver its potential if its structure is designed around the diversity of its participants, not despite it. Different stakeholders bring genuinely different knowledge, and the format should create conditions for that knowledge to meet. Regulators and policymakers bring jurisdictional authority and legislative experience. Technical practitioners and auditors bring operational knowledge of what governance looks like when it leaves the document and enters a real system. Civil society and affected communities bring the perspective of those whose rights are at stake, often the last to be consulted and the first to be affected. Academia and research institutions provide the empirical evidence base that separates governance grounded in data from governance grounded in assumption. Standards bodies and technical organisations provide the methodological infrastructure needed to translate principles into verifiable requirements. Private sector organisations, particularly small and mid-size enterprises deploying AI across multiple jurisdictions, bring the implementation reality that large platforms rarely reflect. None of these contributions is sufficient on its own. A Dialogue composed only of regulators and academics will produce frameworks that are intellectually coherent and operationally unrealistic. A Dialogue dominated by large technology companies will produce frameworks that protect incumbents. A Dialogue without affected communities will produce frameworks that protect systems rather than people. The exchange between these groups is where the real value lies. Understanding the practical constraints of those responsible for implementation and enforcement is not a secondary concern. It is the primary test of whether governance is fit for purpose. The most important conversations in any effective Dialogue will be the ones where a policymaker and an auditor disagree about what compliance actually looks like in practice.

The most underrepresented voices in global AI governance fall into two distinct categories, each absent for different reasons. The first is those who implement compliance, not design it. Auditors, compliance officers and operational practitioners working inside organisations subject to AI regulation rarely appear in governance forums. Their absence is significant because they are the ones who know where frameworks fail in practice, which requirements are operationally unworkable, and what evidence of compliance actually looks like when it leaves the document and enters a real system. Their silence is not disengagement. It is a structural exclusion. A governance framework designed without their input will produce obligations that are theoretically sound and practically unenforceable. The second is small and mid-size organisations deploying high-risk AI. Global AI governance conversations are shaped disproportionately by large technology companies and large regulators. The organisation with eighty employees deploying a recruitment AI tool across three markets, with no dedicated compliance function and no legal team with AI expertise, is absent. Yet this organisation carries the same regulatory obligations as a multinational. Its experience of those obligations, its constraints, its confusion and its practical knowledge of what compliance actually costs, is entirely different and entirely absent from the room. Including these voices requires more than open calls for participation. It requires active recruitment, funded participation, and formats designed around the knowledge these groups hold. The most important conversations will not happen in plenary sessions. They will happen when a policymaker and a compliance practitioner sit in the same working group and agree about what accountability actually looks like in practice.

The most significant barrier to meaningful engagement at governance dialogues is not the absence of good ideas. It is the format that privileges those already fluent in the language of multilateral institutions and marginalises those who are not. Plenary sessions and panel discussions reward prepared statements over genuine exchange. They produce the appearance of dialogue while the real positions remain unchanged. If the AI Dialogue wants to be different, it needs formats that are designed around knowledge exchange rather than position performance. The most effective format would pair regulators and compliance practitioners in structured problem-solving sessions around specific implementation scenarios. Not "what should accountability look like" but "here is a real system, here is what the operator knows, here is what the regulator needs, where is the gap." That conversation has never happened at a multilateral level. It is the most important conversation the Dialogue could host. Reverse briefings would serve a similar purpose. Instead of regulators briefing practitioners on what frameworks require, practitioners brief regulators on what implementation actually costs, where requirements are unworkable, and what evidence looks like in practice. This reversal of the usual information flow would produce insights that no amount of expert testimony can replicate. Structured disagreement sessions, where participants are asked to argue against their own position, would surface assumptions that remain invisible in consensus-driven formats. The Dialogue should also create written channels for contributions from those who cannot attend in person, with genuine mechanisms for those contributions to influence outcomes rather than being archived and ignored.

The most effective examples of AI governance come not from policy documents but from the gap between what frameworks require and what organisations can actually implement. At the policy level, the EU AI Act's risk classification system and the GDPR's data protection impact assessment process are the most instructive models. Both demonstrate that rights-based frameworks with enforcement consequences change organisational behaviour in ways voluntary guidance cannot. At the practice level, the most concrete solution I encounter in my own cross-jurisdictional audit work is the absence of a shared evidentiary standard. Every client asks the same question: what does proof of compliance actually look like? The answer differs by jurisdiction, by auditor, and by regulator. The most effective practice I have developed is mapping all applicable regulatory requirements to a single structured audit baseline before any client engagement begins. This removes the duplication of building a compliance methodology from scratch for each market. At the platform level, AI governance tools such as model risk monitoring platforms and automated fairness testing suites exist but remain inaccessible to small and mid-size organisations without dedicated technical teams or significant budget. The governance infrastructure available to a multinational is simply not available to the organisations that need it most. The most significant challenge is not the absence of good examples. It is that effective governance approaches remain isolated within individual practitioners, firms or jurisdictions rather than becoming shared standards. The Dialogue has a concrete opportunity here: to examine what is already working at the practitioner level and elevate it into a shared multilateral methodology rather than leaving convergence to those few operating across jurisdictions.