This paper discusses the next steps in establishing the International Network of AI Safety Institutes (AISI) following the Seoul Statement and offers recommendations for its members in anticipation of the inaugural meeting in San Francisco this November and the AI Action Summit in Paris in February 2025. The recommendations are structured around three fundamental questions: 1. **Goals of Collaboration**: The AISI network must define clear objectives for international cooperation among safety institutes to maximize returns on investment, considering both the benefits and inherent costs of collaboration. This clarity should be reinforced by specific priorities, deliverables, and timelines that guide the network's activities. 2. **Mechanisms of Collaboration**: Success hinges on the network members’ ability to act collectively toward common goals. Members need to explore various collaboration mechanisms, including leadership structures, research exchanges, shared platforms, and annual conferences. 3. **International Strategy**: The AISI network must clarify its distinct role within the broader AI governance landscape, which includes initiatives from G7, the UN, OECD, and others.

Members should articulate how the AISI differentiates itself, how it plans to engage with these existing initiatives, and the rationale behind its interactions. The paper provides foundational background on AI safety and the emergence of AI Safety Institutes (AISIs) globally, detailing their roles in mitigating risks associated with advanced AI systems through technical research, guidance, and collaboration with companies. It emphasizes the importance of public trust in AI safety as a catalyst for adoption and innovation. The text outlines nine salient questions geared towards guiding the development of the AISI network's goals, collaboration mechanisms, and international strategy. Recommendations include setting concrete projects to demonstrate initial momentum, identifying specific deliverables, scheduling key dates for deliverables related to international AI summits, establishing a regular meeting cadence among members, and deciding on leadership structures and voting processes. Furthermore, the paper discusses how to integrate the AISI network within the existing international AI governance framework while maintaining openness to new members who can contribute meaningfully. This would foster broader consensus and cooperation on AI safety, while also ensuring that the network retains its efficacy and focus on key safety issues. The establishment of the AISI International Network represents a significant step forward in global AI safety collaboration, and as members convene, they are encouraged to transform intent into actionable strategies to enhance AI governance and safety standards worldwide. This paper is authored by Gregory C. Allen, director of the Wadhwani AI Center at CSIS, and Georgia Adamson, a research associate there. The report is supported by Microsoft.