Dassault Systemes

A strong outcome would be alignment on risk prioritization. Even if countries disagree on many issues, identifying a small set of urgent risks (e.g., misuse, systemic bias, loss of control in advanced systems) and agreeing to collaborate on them would be a major step forward.

Safe, secure and trustworthy AI;Interoperability of governance approaches

AI systems operate across borders, while governance frameworks remain fragmented. Without greater interoperability, differing rules and standards risk creating gaps, inconsistencies, and inefficiencies that can weaken both oversight and innovation. Strengthening interoperability is therefore essential to enable coordination, mutual understanding, and more coherent governance across jurisdictions. At the same time, ensuring AI is safe, secure, and trustworthy is fundamental to its long-term adoption and societal acceptance. Trust must be grounded in concrete measures, such as risk mitigation, accountability, and transparency ,rather than broad principles alone. These two topics reinforce one another: interoperable governance frameworks can support more consistent safety practices globally, while a shared commitment to trustworthy AI provides a basis for aligning different approaches. This is key to building an AI ecosystem that is both innovative and responsibly governed.

Although related to AI Capacity building, another emerging challenge is compute and resource concentration. Advanced AI development depends on significant computational power, which is controlled by a small number of actors and countries. This raises concerns about unequal influence, barriers to entry, and global imbalances in who shapes AI systems and standards.

Governance gaps are increasing short-term risk and complexity, but they are also redefining competition in the software sector, rewarding companies that can integrate safety, interoperability, and trust directly into their products and development processes. - New markets for governance and safety tools - Competitive differentiation through trust - Interoperability as a design advantage

It can serve as a neutral platform for convergence. In a fragmented landscape, the Dialogue can help align different governance approaches around shared priorities. It can also facilitate practical cooperation, not just high-level discussion. By convening regulators, industry, and technical experts, the Dialogue can support the development of common tools Lastly, it should act as an early warning and coordination mechanism. Regular exchanges on emerging risks, such as misuse, systemic failures, or security vulnerabilities, must be implemented.

We recommend the OECD AI Principles, which provide widely endorsed baseline norms and the Global Partnership on AI (GPAI), which advances research and policy collaboration.

Coming from the Industry , I think we can provide practical insights on implementation, how safety, security, and compliance measures work in real systems. We can also contribute data (e.g., risk assessments, incident learnings) and help pilot common standards. Technical and academic experts can support the Dialogue with independent evidence, evaluation methodologies, and foresight on emerging risks, helping ground discussions in technical reality. For the format, we should combine high-level plenaries with smaller working groups. Plenaries can set direction and political momentum, while working groups focus on specific issues like evaluation standards, interoperability, or risk governance. Then we need to ensure continuity beyond a single event. The Dialogue should operate as an ongoing process, with clear timelines, deliverables, and follow-up mechanisms (e.g., periodic meetings, progress reports). Finally, prioritize output-oriented discussions. Each track should aim to produce concrete outcomes, such as shared definitions, voluntary guidelines, or pilot initiatives.

Workers and labor organizations: The impact of AI on jobs, working conditions, and economic distribution is central, yet worker representation is often indirect or absent.

The key is to move beyond traditional panel discussions toward structured collaboration and real-time problem solving. 1. Problem-solving "labs" (challenge-based sessions) 2. Multi-stakeholder simulation exercises Participants role-play real-world scenarios (e.g., a cross-border AI failure, deepfake crisis, or model misuse incident) 3. Rapid feedback cycles (mini-plenaries) Short iterative sessions where working groups present interim findings, receive feedback, and refine outputs within the same forum

The Global Partnership on AI (GPAI) is a practical model for collaboration between governments, industry, and researchers. It focuses on applied projects such as responsible AI development, data governance, and safety evaluation.