This input is submitted by Forkit Dev and is informed by discussions and feedback from Prof. Dr. Alexander Jesser, Syed Amir Hamza, and Egor Yablokov, with affiliations to Hochschule Heilbronn / ICPS and E-Quadrat Science & Education. Institutional affiliations are provided for context and do not imply formal institutional endorsement unless separately stated.

The first Global Dialogue on AI Governance would be successful if it helps move AI governance from high-level principles toward practical, interoperable mechanisms that can work across borders, sectors, and institutional contexts. A strong outcome would be agreement on shared minimum governance baselines for AI systems, especially around provenance, human oversight, accountability, permissions, safety, incident learning, interoperability, capacity building, and compute sustainability. These baselines should not require a single centralized database, vendor platform, or uniform legal regime. Instead, they should allow global principles, regional and supranational frameworks, national policies, institutional controls, and human accountability to remain distinct but interoperable. This is especially urgent as AI systems move from static models and chatbots toward autonomous agents that can access tools, execute workflows, interact with APIs, and consume compute continuously. Institutions need reliable ways to document who authorized a system, what it was allowed to access, which model or agent was running, what changed over time, and who remained responsible for monitoring or escalation. A successful Dialogue should therefore encourage common terminology, practical accountability records, open reference implementations, and capacity-building pathways so that trustworthy AI governance is accessible not only to large technology companies or wealthy jurisdictions, but also to smaller institutions and lower-resource countries.

Safe, secure and trustworthy AI;AI capacity-building;Interoperability of governance approaches;Transparency, accountability, and human oversight

These priorities reflect the urgent shift from AI experimentation to operational deployment. AI systems are increasingly embedded in real workflows, and many are becoming agentic: they can access tools, call APIs, interact with data, execute tasks, and affect people or organizations directly. This makes safe, secure, and trustworthy AI essential. Interoperability of governance approaches is critical because countries, regions, and institutions will not adopt identical AI laws or technical systems. However, they still need a common accountability structure that supports cross-border understanding, auditability, and cooperation. Transparency, accountability, and human oversight are central because many AI failures are not only model failures. They may result from unclear permissions, weak review, poor deployment design, missing ownership records, or insufficient human responsibility. Governance must be able to show who approved an AI system, what it was allowed to do, and who remained responsible for monitoring, escalation, or revocation. AI capacity-building is equally important. Practical governance tools must be usable by smaller institutions, public agencies, universities, and lower-resource countries, not only by large technology firms. Open reference implementations, shared terminology, and minimal interoperable metadata baselines can help make trustworthy AI governance more accessible and operational.

One emerging issue is the governance of autonomous AI agents and AI-enabled workflows, not only individual AI models. Many current governance approaches still focus on models as static artifacts. However, real-world AI systems are increasingly composed of models, agents, prompts, APIs, tools, data sources, runtime environments, and human approvals. These systems can act across organizational boundaries and may change over time. This creates a need for portable accountability records that capture provenance, permissions, human oversight, deployment context, runtime evidence, and compute-use indicators. Such records could help institutions distinguish between model failure, unsafe permissions, weak human oversight, poor deployment design, or organizational misuse. A second cross-cutting issue is compute sustainability. As AI systems become more agentic and resource-intensive, governance should consider whether model choice and compute use are proportionate to the task. Token use, compute intensity, runtime duration, and efficiency rationale should become more visible and governable. A third issue is federated governance. Global principles, regional and supranational frameworks, national policies, institutional controls, and human accountability should remain distinct, but interoperable through shared governance records. This would allow different actors to govern differently while preserving a common structure for accountability and cross-border review.

Governance gaps are directly affecting the European AI and startup sector as organizations move from experimentation to production deployment. In Germany and the wider European region, companies are under growing pressure to adopt AI while also demonstrating safety, accountability, human oversight, and regulatory preparedness. The challenge is that many organizations still lack operational records showing which model or agent was used, who approved it, what permissions it had, what changed between versions, and what evidence exists after deployment. This creates several risks. First, accountability becomes fragmented across developers, vendors, business teams, compliance teams, and external AI providers. Second, autonomous AI agents can access tools, APIs, files, and workflows without sufficiently clear permission records. Third, smaller companies and public institutions may struggle to implement governance practices because existing solutions can be costly, complex, or designed mainly for large enterprises. Fourth, compute and token consumption are often poorly documented, making cost, efficiency, and sustainability difficult to evaluate. At the same time, this creates a major opportunity for Europe. The region can lead in practical, trustworthy AI governance by promoting interoperable accountability layers that work across sectors and jurisdictions. Minimal AI passport-style records for models, agents, and workflows could help organizations document provenance, human oversight, permissions, runtime evidence, and compute-use indicators without forcing them into one centralized platform. For the AI infrastructure sector, the opportunity is to build governance directly into development and deployment workflows. This can support safer innovation, auditability, cross-border cooperation, and stronger public trust while helping startups and smaller organizations participate in responsible AI adoption.

The AI Dialogue can play a central convening role by helping governments and stakeholders move from fragmented AI governance discussions toward shared, practical mechanisms for cooperation. Its value is not to replace national laws, regional frameworks, or existing technical standards, but to create a neutral forum where these approaches can become more interoperable. A key role for the AI Dialogue should be to identify minimal common governance baselines that different countries, regions, and institutions can adapt. These could include shared terminology and metadata expectations for provenance, human oversight, permissions, accountability, incident learning, and compute sustainability. Such baselines would help support cross-border trust without requiring a single global regulator or centralized database. The AI Dialogue can also help connect policymakers with technical communities, startups, researchers, civil society, and public institutions. This is important because many governance challenges only become visible at implementation level, when models and agents are deployed into real workflows. Finally, the AI Dialogue can support capacity building by ensuring that practical AI governance tools are accessible to lower-resource countries, smaller institutions, universities, public agencies, and startups. International cooperation should not only focus on principles, but also on the operational infrastructure that allows organizations to document, audit, and govern AI systems responsibly.

The AI Dialogue should build upon and connect with existing initiatives that already provide important foundations for AI governance. These include the OECD AI Principles and AI incident reporting work, the UN High-Level Advisory Body on AI, UNESCO's work on AI ethics, the EU AI Act, the Council of Europe Framework Convention on AI, regional initiatives such as those in the African Union and ASEAN, and technical standardization work from bodies such as ISO/IEC, IEEE, and NIST. The Dialogue should also connect with open-source and technical communities developing practical governance tools, as well as universities, startups, public agencies, and civil society organizations that are testing governance mechanisms in real-world environments. The added value of the AI Dialogue would be to connect these efforts into a more coherent international cooperation layer. Many existing initiatives focus on principles, legal requirements, risk management, standards, or incident reporting. What is still missing is stronger interoperability between them at the implementation level. The AI Dialogue could help define shared minimum accountability records for AI systems, including models, agents, and AI-enabled workflows. These records could support provenance, permissions, human oversight, deployment context, incident analysis, and compute-use visibility. This would help different governance systems communicate with each other without requiring identical laws or platforms. In this way, the AI Dialogue can add value by turning existing principles and frameworks into more operational, interoperable, and globally inclusive governance pathways.

Different stakeholders can contribute most effectively if the AI Dialogue combines high-level policy discussion with implementation-focused working formats. Governments can share national priorities, regulatory experiences, public-sector use cases, and capacity-building needs. Regional and supranational bodies can explain where interoperability is needed between legal and policy frameworks. Technical communities can contribute practical knowledge about model evaluation, provenance, security, agent permissions, compute monitoring, and deployment workflows. Academia can provide evidence, impact assessment methods, and independent research. Civil society can ensure that human rights, inclusion, accountability, and affected communities remain central. The private sector and startups can contribute practical deployment experience, including where governance breaks down in real-world systems. The format should include several layers: plenary sessions for shared priorities, thematic working groups for specific topics, written inputs for structured recommendations, and implementation labs where stakeholders can compare practical governance mechanisms. Working groups could focus on areas such as human oversight, incident reporting, AI agents, compute sustainability, public-sector deployment, and governance interoperability. The AI Dialogue should also include regional consultations before and after the main sessions, so that participation is not limited to those able to attend in person. A public synthesis document should map areas of convergence, disagreement, and future work. A useful structure would be: first, identify shared challenges; second, map existing frameworks and gaps; third, define practical cooperation mechanisms; fourth, create follow-up pathways through working groups, pilot projects, and capacity-building initiatives.

Several voices remain underrepresented in global AI governance discussions. First, lower-resource countries and smaller public institutions are often underrepresented, even though they will be deeply affected by AI adoption and may have less capacity to build governance infrastructure. Their inclusion is essential to avoid governance models that only work for wealthy jurisdictions or large technology companies. Second, startups and smaller AI builders are underrepresented. Many governance discussions focus on large technology firms, but smaller companies often build and deploy AI systems quickly and need practical, lightweight governance mechanisms. Third, technical implementers are not always sufficiently included. Engineers, open-source maintainers, MLOps teams, cybersecurity professionals, and AI infrastructure builders often see governance failures before they become policy problems. Their practical experience is important. Fourth, affected communities and frontline users need stronger representation. This includes workers, patients, students, public-service users, minority-language communities, people with disabilities, and communities affected by automated decision-making. Fifth, sustainability and infrastructure perspectives are still underrepresented. AI governance discussions should include experts on compute infrastructure, energy demand, data centres, and environmental impact. These groups could be included through regional consultations, travel support, remote participation, multilingual formats, open written submissions, public comment periods, and dedicated working groups. The AI Dialogue should also invite practical case studies from smaller organizations, not only formal statements from large institutions. Inclusion should mean participation in agenda-setting, not only attendance.

The AI Dialogue could foster meaningful engagement through formats that combine policy discussion with practical implementation. One useful format would be governance implementation labs, where stakeholders examine concrete AI deployment scenarios and identify what governance records, oversight mechanisms, and accountability structures are needed. For example, labs could explore AI agents in healthcare, education, public administration, robotics, or enterprise workflows. A second format could be interoperability roundtables, where regulators, standards bodies, startups, technical experts, and civil society compare how different governance frameworks handle provenance, human oversight, permissions, incident reporting, and compute sustainability. A third format could be regional challenge sessions, where participants from different regions identify specific barriers they face, such as lack of technical capacity, limited access to governance tools, infrastructure constraints, language gaps, or procurement challenges. A fourth format could be technical-policy demonstrations. These would allow stakeholders to show practical governance mechanisms, such as AI passports, model cards, incident reporting tools, audit logs, watermarking systems, or compute monitoring approaches. The purpose should not be product promotion, but shared learning about what works in practice. A fifth format could be living consultation documents, where stakeholders can comment on draft recommendations before and after the Dialogue. This would make the process more transparent and iterative. Finally, the Dialogue could use small mixed-stakeholder working groups with balanced representation from government, civil society, academia, technical communities, and private sector actors. These groups could produce short, actionable outputs rather than only general statements.

Effective AI governance requires both policy frameworks and operational tools. Several existing approaches provide useful foundations. At the policy level, the OECD AI Principles, UNESCO's Recommendation on the Ethics of AI, the EU AI Act, the Council of Europe Framework Convention on AI, and NIST's AI Risk Management Framework all provide important guidance on trustworthy AI, human oversight, accountability, transparency, risk management, and rights protection. These frameworks help establish shared expectations, but they still need practical implementation mechanisms. At the practice level, organizations are increasingly using model cards, system cards, data sheets, risk assessments, red-teaming, incident reporting, audit logs, human oversight procedures, and impact assessments. These practices are valuable, but they are often fragmented, static, or difficult to connect across teams, vendors, and jurisdictions. A promising concrete approach is the use of portable AI accountability records, such as model and agent passports. These can document provenance, version history, ownership, permissions, human oversight, deployment context, runtime evidence, and compute-use indicators. Such records can help organizations move from abstract governance principles to operational accountability. Open reference implementations and interoperable standards are especially important because they can lower adoption barriers for smaller institutions, public agencies, startups, and lower-resource countries. For example, open passport-style infrastructure can allow different actors to maintain their own governance systems while sharing a common accountability structure for cross-border review, incident analysis, and auditability. The most effective governance approaches will combine policy alignment, technical interoperability, practical evidence records, and capacity building. They should help institutions govern AI systems throughout the lifecycle, from development and deployment to monitoring, incident response, and continuous improvement.