The Sidney Group PLLC

A successful Dialogue should move beyond principles and produce clarity on how AI is actually governed in practice. First, it should identify where real authority sits. In many systems, control is not at the point of use. It is set earlier through procurement, contract terms, data access, and system design. If governance frameworks do not address those points, they will miss where decisions are truly shaped. Second, it should establish shared expectations for accountability. Not just that systems are "responsible," but that there is a clear answer to who can delay, override, or stop a system when risk changes. Third, it should focus on auditability. If a system cannot be explained after the fact in a way that holds up under regulatory or legal review, governance has not been achieved. Finally, it should produce practical outputs. That could include baseline expectations for decision authority, audit trails, and execution controls that can be applied across jurisdictions. Success is not another set of high-level commitments. It is clearer alignment on how authority, accountability, and control are actually exercised in real systems.

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

These priorities are closely linked. Transparency, accountability, and human oversight are foundational because governance depends on whether decisions can be explained, traced, and challenged. Without that, other safeguards are difficult to enforce. Safe, secure, and trustworthy AI depends on more than model performance. It requires systems that can be controlled in real conditions, including the ability to intervene when circumstances change. Interoperability matters because AI systems and the companies that deploy them operate across jurisdictions. If governance approaches are not aligned, authority gaps will emerge and enforcement will weaken. Human rights remain central because many high-impact uses of AI affect access to credit, employment, healthcare, and public services. Governance frameworks must ensure that these systems do not produce outcomes that are difficult to contest or remedy. Across all four areas, the key issue is not only whether standards exist, but whether they can be applied, enforced, and audited in practice.

A key issue that is not fully captured is where authority actually sits in AI systems. In many cases, institutions appear to be making decisions, but the real terms of action are set earlier. Procurement choices, vendor dependencies, data access arrangements, and cross-border data flows can determine what a system can and cannot do before it is ever used. This creates a gap between formal responsibility and actual control. Another issue is the timing of authority. Many systems are validated at one point in time and then execute later without reassessing whether conditions have changed. Governance frameworks should address whether systems still have a valid basis to act at the moment of execution. Finally, auditability in practice remains underdeveloped. It is not enough for systems to be technically explainable. Organizations must be able to reconstruct decisions in a way that holds up under legal and regulatory scrutiny. These issues cut across all themes and are central to whether AI governance works in real environments.

In the United States, governance gaps are showing up most clearly in high-impact sectors like finance, healthcare, employment, and public services. The core challenge is a gap between formal responsibility and actual control. Organizations are held accountable for outcomes, but key decisions are often shaped earlier through vendor systems, procurement terms, and data dependencies. By the time a system is deployed, important constraints are already set. Another challenge is that many systems are evaluated at deployment but not continuously reassessed. Conditions change, but systems often continue to operate as if the original approval still applies. This creates risk in areas where decisions affect people's access to jobs, credit, or services. There is also growing pressure from regulators to demonstrate accountability. It is no longer enough to say a system works. Organizations are being asked to explain decisions, document processes, and show how risk is managed in practice. At the same time, there is a clear opportunity. Institutions are beginning to invest in auditability, clearer decision ownership, and stronger internal controls. There is also movement toward aligning governance expectations across jurisdictions, particularly in transatlantic contexts. The opportunity is to build systems where authority, accountability, and control are clear and testable. The risk is that without that clarity, governance will remain formal on paper but weak in practice.

The AI Dialogue can be useful if it helps move the conversation from broad principles to real points of control. A lot of international AI governance still lives at a high level. The harder question is where authority actually sits when systems are built, bought, deployed, and used across borders. In practice, control often shifts through procurement, contract terms, data access, technical standards, and vendor dependence. If the Dialogue helps countries talk honestly about that, it can add real value. It can also help by creating a common language for accountability. Not just whether systems are called safe or trustworthy, but whether there is a clear answer to who can explain a decision, challenge it, delay it, or stop it when conditions change. The Dialogue should also help connect legal, technical, and operational conversations that are too often kept apart. Countries do not need one identical model. But they do need more shared understanding of audit trails, human responsibility, and how governance holds up in practice. Its value will not come from another abstract statement of principles. It will come from helping countries close the gap between formal responsibility and actual control, and from making international cooperation more practical, more honest, and easier to use. The UN has framed the Dialogue as a platform for international cooperation, sharing best practices, and open discussion, which gives it room to play exactly that role.

The Dialogue should build on the OECD, UNESCO, the G7 Hiroshima AI Process, the Council of Europe, and recent UN work. The world does not lack principles. It lacks alignment around where authority actually sits. That is the gap the Dialogue can help close. Its value is not in repeating broad commitments. It is in connecting those frameworks to the places where control is really set: procurement, contracts, standards, data access, audit trails, system design, and cross-border dependence. It should also create room for countries and institutions that are usually asked to live under frameworks they did not shape. The goal should be simple: make the global AI governance landscape more coherent, more inclusive, and more tied to how power actually works in real systems.

Different stakeholders should do more than give statements. They should help test where governance works and where it breaks. Governments can speak to law, enforcement, and public accountability. Companies can speak to deployment, controls, and operational limits. Civil society can identify where rights and remedies fail in practice. Technical experts can clarify what is and is not actually possible. Academia can help separate evidence from aspiration. The format should reflect that. Keep formal statements short. Use more structured discussion around specific governance problems, such as auditability, cross-border control, procurement, and who has authority to stop a system when risk changes. The Dialogue will be most useful if it is built around real governance questions, not long general remarks.

Too much of this conversation is still being set by the people with the money, the standards, or the platforms. Missing voices include countries in the Global South that are expected to live under rules they did not write, frontline regulators, labor, public-interest advocates, and the people inside institutions who are actually told to make governance real. Inclusion should mean more than being invited into the room after the agenda is set. It should mean real power to shape the agenda, the discussion, and the outputs.

Use formats that force people to be concrete, interactive, action-focused, and clear about remedies. Short sessions. Real cases. One hard question at a time. For example: who had the authority to stop this system, why did that fail, and what would have prevented it? That will do more than a room full of prepared remarks. The Dialogue should push participants to show where governance holds, where it breaks, and who has the power to act when it matters.

It's a set of operating practices that show how governance actually works day to day, not just what it should look like in theory. One example is clear ownership. Someone inside the organization is responsible for the system, not just the vendor. If something goes wrong, there is no confusion about who answers for it. Another is audit trails. Decisions can be reconstructed after the fact in a way that holds up under review. Not just what happened, but why it happened. Third is the ability to stop or override a system. If no one can pause it when conditions change, governance is not real. Fourth is regular review. Systems should not be approved once and left alone. They should be checked as conditions change. Finally, procurement matters. Many of the real decisions are made when systems are bought and set up. If control is given away at that stage, it is hard to get back later. These are simple practices, but they make governance real.