Sam Houston State University

Success requires outputs that are binding in direction, even where they are not immediately enforceable. Three concrete outcomes would mark the Geneva session as substantively consequential.First, a shared definitional baseline. The AI governance conversation is currently fragmented because states, technical communities, and civil society do not operate from agreed definitions of key concepts: trustworthy AI, meaningful human oversight, AI-generated evidence. The Dialogue should produce a working lexicon that member states commit to applying in domestic legislation and bilateral agreements.Second, a structured capacity gap assessment. Declarations about AI equity are common; documented, country-level assessments of AI governance capacity are not. The Dialogue should mandate a systematic mapping of where technical expertise, regulatory infrastructure, and computing access are absent — particularly across Africa, Small Island Developing States, and Least Developed Countries — and attach a concrete resource-mobilization mechanism to the findings.Third, an accountability framework for AI in criminal justice and security. This is the domain where AI is already making consequential decisions affecting individual liberty, yet it remains the least governed. A successful first Dialogue would produce agreed minimum standards for AI transparency in law enforcement contexts, including disclosure requirements when AI contributes to arrest, prosecution, or sentencing decisions.A successful Dialogue is not one that produces a comprehensive treaty. It is one that produces specific, actionable commitments in at least these three areas, with named follow-up mechanisms and a clear accountability structure for the May 2027 New York session.

• Safe, secure and trustworthy AI
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Open-source software, open data and open AI models
• Protection and promotion of human rights

These four priorities reflect the domains where AI governance failures are already producing measurable, real-world harm not speculative risk. Safe, secure and trustworthy AI is foundational because AI is now embedded in critical infrastructure: water treatment, energy grids, and industrial control systems. My doctoral research on explainable intrusion detection in SCADA/ICS environments demonstrates that AI deployed in these contexts operates in environments with near-zero tolerance for error and limited auditability. Governance must catch up to deployment. AI capacity-building is not a peripheral concern. During nine years as a Cybercrime Intelligence Officer at INTERPOL NCB Abuja, I observed African law enforcement agencies routinely receiving AI-derived threat intelligence they lacked the technical infrastructure to audit, challenge, or replicate. Capacity-building framed only as training misses the deeper problem: sovereign capability requires compute access, open models, and embedded technical partnerships. Protection and promotion of human rights is urgent because AI in law enforcement facial recognition, predictive policing, digital evidence triage is already affecting individual liberty in jurisdictions with weak judicial oversight of algorithmic outputs. The harm is not hypothetical; it is documented and disproportionately affects marginalized communities. Transparency, accountability, and human oversight are the operational requirements that make the other three meaningful. Without enforceable explainability standards and chain-of-custody requirements for AI-generated forensic evidence, governance frameworks remain aspirational. My work in digital forensics and cybersecurity compliance has made clear that oversight is not a feature that can be added after deployment; it must be architected in.

Two cross-cutting gaps deserve explicit attention in the Dialogue's thematic architecture. The first is AI governance in operational technology and critical infrastructure security. The listed themes address AI's societal and ethical dimensions comprehensively but largely assume a software-layer deployment context. AI embedded in industrial control systems, SCADA networks, and cyber-physical infrastructure introduces a distinct governance challenge: decisions are made at machine speed, human override may not be technically feasible in real time, and the consequences of failure are physical power outages, water contamination, infrastructure collapse. Existing cybersecurity frameworks were not designed to govern AI decision-making in these environments, and AI governance frameworks have not yet confronted the cyber-physical layer. This gap needs its own workstream. The second is the evidentiary and legal status of AI outputs in criminal proceedings. This issue is adjacent to transparency and human rights but is not fully captured by either. As AI tools become standard components of forensic investigation analyzing device data, reconstructing timelines, classifying malware, identifying network intrusions their outputs are entering courtrooms without agreed standards for admissibility, disclosure, or challenge. This is not simply a human oversight problem; it is a rule-of-law problem. The Dialogue should engage directly with international criminal justice bodies to develop minimum evidentiary standards for AI-assisted forensic evidence, applicable across jurisdictions and enforceable through mutual legal assistance treaty frameworks. Both issues are technically complex, jurisdictionally cross-cutting, and already generating real-world consequences. Waiting for the 2027 session to introduce them would be a missed opportunity.

My professional experience spans three intersecting contexts U.S. higher education cybersecurity, West African law enforcement, and doctoral research in digital forensic science and governance gaps manifest differently, but consequentially, in each. In the United States, AI is being integrated into institutional cybersecurity operations faster than compliance frameworks can accommodate. At the Texas Higher Education Coordinating Board, where I work as a Cybersecurity Analyst, the challenge is not access to AI tools; it is the absence of federal standards governing how AI-assisted threat detection outputs are documented, audited, and acted upon. When an AI system flags a network anomaly that triggers an incident response, there is no standardized accountability trail. This is a sector-wide problem across U.S. public institutions, and it is unresolved. In West Africa, the governance gap is structural. During nine years at INTERPOL NCB Abuja, the most persistent obstacle to effective cybercrime investigation was not investigator competence it was asymmetric access to AI-enabled forensic tools. Wealthier jurisdictions produced AI-derived intelligence that shaped multinational operations, while African member bureaus lacked the infrastructure to independently verify, replicate, or contest those outputs. The governance gap here is not regulatory lag; it is the absence of sovereign AI investigative capacity, compounded by inadequate data-sharing frameworks between jurisdictions operating under different legal standards. In the digital forensics sector broadly, the most significant emerging challenge is the evidentiary vacuum created by AI-assisted investigation tools entering legal proceedings without agreed admissibility standards. Courts in multiple jurisdictions are already receiving AI-generated forensic outputs network traffic analyses, malware classifications, timeline reconstructions with no framework for evaluating their reliability or disclosing their methodological limitations to defense counsel. The opportunity, across all three contexts, is the same: this Dialogue can establish the minimum standards that domestic institutions and national legislatures will reference for the next decade. The window to shape that foundation is now.

The AI Dialogue occupies a structural position that no existing mechanism does: it is the only forum where every UN member state has standing to participate in AI governance deliberation on equal terms. That universality is its primary asset, and the Dialogue should use it to do what bilateral agreements, regional blocs, and industry-led frameworks cannot establish a floor of shared obligations that applies regardless of a country's AI development stage or geopolitical alignment. Three specific cooperative functions are within reach. First, the Dialogue can serve as a norm convergence mechanism. The EU AI Act, the U.S. Executive Order on AI safety, the African Union's AI Continental Strategy, and ASEAN's AI governance frameworks are not incompatible, but they are not interoperable. The Dialogue is positioned to identify the common principles across these instruments and translate them into a reference framework that states can adopt, adapt, and invoke in bilateral and multilateral negotiations. Second, it can operationalize mutual legal assistance for AI-related evidence. Cybercrime investigations routinely cross jurisdictions, and AI-generated forensic evidence now regularly features in those cases. No existing MLAT framework addresses AI evidence standards. The Dialogue can mandate a working group with direct links to UNODC and INTERPOL to close this gap. Third, it can establish a mandatory transparency registry for AI systems deployed in law enforcement and critical infrastructure a mechanism through which states report what AI systems they are using in high-stakes public functions, under what oversight conditions, and with what incident disclosure obligations. Voluntary registries have not worked. A UN-anchored mechanism with reporting expectations attached to existing treaty obligations would. The Dialogue's value is not in replacing existing cooperation structures. It is in giving them a universal reference point they currently lack.

Several existing mechanisms provide substantive foundations the Dialogue should engage directly rather than replicate. INTERPOL's Cybercrime Directorate and its network of National Central Bureaus represent the most operationally mature global infrastructure for cross-border AI-assisted criminal investigation. The Dialogue should formally connect with INTERPOL's AI and cybercrime working groups to ensure that governance standards developed in Geneva are grounded in investigative realities, particularly the capacity constraints facing NCBs in developing regions. The ITU's AI for Good platform and the Global Initiative on AI and the Rule of Law, hosted by IDLO and UNESCO, have developed practitioner-facing guidance on responsible AI in justice systems. The Dialogue should treat these as technical inputs, not parallel tracks, and commission synthesis documents that translate their outputs into actionable governance language for member states. The NIST AI Risk Management Framework and the EU AI Act's conformity assessment architecture are the most operationally detailed governance instruments currently in force. Rather than producing a competing framework, the Dialogue should commission an interoperability analysis identifying where these instruments align, where they conflict, and what a harmonized baseline would require. The African Union's AI Continental Strategy is frequently cited but rarely integrated into global governance conversations. The Dialogue should establish a formal liaison mechanism with the AU Commission on digital affairs, ensuring that African priorities sovereign compute access, multilingual AI systems, and forensic capacity are not treated as afterthoughts to frameworks designed elsewhere. The added value the Dialogue uniquely provides is legitimacy and universality. Industry consortia can produce technical standards. Regional bodies can produce binding law. Only the UN can produce a framework that every state, regardless of development level or political alignment, has a recognized stake in shaping and implementing.

The Dialogue's legitimacy depends on whether its structure actually distributes influence or merely distributes seating. Formal multi-stakeholder participation has a consistent failure mode: governments deliberate, everyone else observes. Avoiding that outcome requires deliberate structural design, not good intentions. For member states, contribution should be conditioned on submission of a national AI governance inventory prior to each session a standardized disclosure of what AI systems are deployed in public functions, under what legal authority, and with what oversight mechanisms. This converts the Dialogue from a forum for aspirational statements into a mechanism for comparative accountability. For the technical community, contribution should be structured around problem-specific working groups rather than plenary panels. Practitioners in cybersecurity, forensic science, and critical infrastructure operations have precise, actionable knowledge that is lost in high-level debate formats. Dedicated technical tracks, with outputs that feed directly into negotiated text, would capture that expertise. For civil society and affected communities, the submission portal model is insufficient. Written inputs from organizations with limited English-language capacity or no permanent UN representation are systematically disadvantaged. The Dialogue should fund regional civil society coordinators in each UN regional group to aggregate and formally present constituency inputs, with guaranteed floor time in plenary. For the private sector, participation should require disclosure of commercial interests in any governance outcome being discussed. Unattributed industry influence on AI governance standards is a documented problem. Named, disclosed participation is a minimum condition for credibility. On structure: the two-session format across Geneva and New York is appropriate, but each session needs a published negotiating text going in, not just a thematic agenda. Dialogue without a draft to react to produces communiqués, not commitments.

Four communities are consistently present in AI governance conversations as subjects of concern but absent as architects of solutions. Practitioners in developing-nation law enforcement and forensic investigation are the most consequential gap. These are the professionals applying AI tools or being excluded from them in the highest-stakes operational contexts: cybercrime investigation, cross-border financial fraud, critical infrastructure protection. 10 years working at Law enforcement Abuja made it clear that the investigative realities of African, Southeast Asian, and Latin American law enforcement agencies bear almost no resemblance to the scenarios governance frameworks are designed around. Structured practitioner delegations from these agencies, with travel and interpretation support, should be a budget line item, not an afterthought. Indigenous and linguistic minority communities are underrepresented not only in governance rooms but in the AI systems being governed. AI tools trained on majority-language datasets produce discriminatory outcomes for speakers of minority languages and for communities whose cultural frameworks do not map onto Western categorical assumptions. Governance that does not include these communities will not govern these harms. The Dialogue should establish a dedicated indigenous and linguistic diversity advisory mechanism, modeled on UNPFII's engagement with the General Assembly. Early-career researchers and doctoral candidates in AI, cybersecurity, and digital forensics from the Global South represent a generation that will implement whatever governance architecture this Dialogue produces. Their inclusion is not symbolic; it is a quality-of-output issue. A structured fellowship track, with funded participation and formal submission rights, would capture perspectives that senior institutional representatives frequently do not hold. Finally, public defenders and legal aid practitioners who encounter AI-generated evidence in criminal proceedings are almost entirely absent from AI governance processes despite being among the most directly affected professionals. Their inclusion would materially improve the quality of deliberation on accountability and human rights.

Three format innovations would materially improve the quality and durability of the Dialogue's outputs. The first is structured adversarial review. For each major governance proposal that emerges from working groups, the Dialogue should formally commission a red-team panel composed of practitioners, civil society representatives, and technical experts from the Global South tasked specifically with identifying implementation failures, unintended consequences, and enforcement gaps. This is standard practice in cybersecurity policy design and almost entirely absent from intergovernmental AI governance processes. It produces better text and surfaces political and technical blind spots before commitments are made. The second is live scenario simulation. Abstract governance principles become concrete when tested against realistic operational scenarios: an AI system misidentifies a water treatment anomaly as a cyberattack and triggers an automated shutdown; an AI-generated forensic report is challenged in a criminal trial in a jurisdiction with no admissibility standard; a cross-border cybercrime investigation stalls because two countries' AI evidence disclosure requirements are incompatible. Running structured simulations of these scenarios in plenary, with practitioners narrating technical realities in real time, would anchor deliberation in consequences rather than principles. The third is asynchronous regional deliberation between sessions. The gap between the Geneva session in July 2026 and the New York session in May 2027 is eleven months. That interval should not be dead time. Regional deliberation hubs, operating in local languages and time zones, should convene quarterly to react to draft outputs, surface regional priorities, and feed formal inputs into the inter-session negotiating process. Technology exists to support this at low cost. The barrier is political will to treat between-session participation as substantively binding rather than consultative. These three formats share a common logic: governance quality is a function of the realism and diversity of the inputs that shape it.

The most instructive examples of effective AI governance share a common characteristic: they are grounded in operational specificity rather than general principle, and they attach accountability mechanisms to the frameworks they establish. The NIST AI Risk Management Framework represents the most technically rigorous voluntary governance instrument currently in wide use. Its tiered approach to risk categorization distinguishing between AI systems with limited consequence and those with high-stakes outputs in security, health, and justice contexts provides a replicable architecture that the Dialogue should formalize into a universal baseline. Its limitation is that it remains voluntary; the Dialogue's added value would be attaching reporting obligations to its core structure. The EU AI Act's prohibited practices list and conformity assessment requirements for high-risk AI systems demonstrate that binding governance is legislatively achievable at scale. Specifically, its requirement that high-risk AI systems maintain technical documentation sufficient for post-deployment audit is directly applicable to AI use in law enforcement and critical infrastructure - two domains where auditability is operationally essential and currently inconsistent. INTERPOL's Cybercrime Directorate's operational coordination model offers a governance-adjacent example worth formalizing. Its notice and diffusion system for cybercrime which enables cross-jurisdictional action without requiring full legal harmonization could serve as a template for AI incident disclosure: a mechanism through which member states report consequential AI system failures affecting public safety without requiring prior agreement on liability standards. At the research level, explainable AI frameworks being developed for intrusion detection in industrial control systems including work being conducted at institutions such as Sam Houston State University, demonstrate that transparency and real-time human oversight are technically achievable even in low-latency, high-consequence operational environments. Governance frameworks should reference and incentivize this class of applied research rather than treating explainability as a future aspiration.