Actions for government leaders –Drive development of context-specific frameworks: Where possible and practical with local norms, interpret widely accepted frameworks. This can prevent a fragmented landscape, where each industry and sub- industry develops its own AI governance terminology, definitions and practices. Recommended actions in developing context- specific frameworks include: –Engage stakeholders: Gather diverse expertise across sectors, company sizes, and impacted end-users and communities, e.g. the US AI Action Plan requires NIST to convene a broad range of stakeholders to accelerate the development and adoption of domain-specific national standards for AI systems.53 –Map context: Identify risks, opportunities, regulatory requirements and context-specific guidance, such as those provided by an industry group or civil society organization. –Draft framework: Detail policies with accountability structures, use cases and best practices. Provide tiered governance models with essential controls and progression paths as organizations grow. –Prototype: Gather public feedback on drafts and pilot-test with organizations (see Play 3). Ensure ongoing evaluation and refinement. –Incentivize adoption: Provide framework training and a Q&A channel for communication with policy-makers to enhance uptake. Test rewards for adoption, such as compliance recognition. –Promote open sharing of responsible AI resources: Sharing best practices, case studies and tools can enable access to context- specific insights and prevent unnecessary trial-and-error lessons already learned by other organizations. For SMEs, access to responsible AI tools is critical to implementation. Single- entry-point resource repositories should be developed or enhanced when they already exist. Examples include the Organisation for Economic Co-operation and Development’s (OECD) Catalogue of Tools and Metrics for Trustworthy AI,54 Canada’s AI and Data Governance Standardization Hub55 and the World Bank Group’s AI-in-Government Case Study Repository.56 Factors behind a successful repository include: –Scope: Clear objectives such as target audience, resources and outcomes –Accessibility: Multi-language support and navigable search and retrieval with relevant labels –Quality: Reliable, curated and up to date –Comprehensiveness: Diverse contributions –Inclusivity: Caters to diverse expertise and businesses with varying levels of maturity and resources –Transparency: Governance and curation –Sustainability: Long-term funding and maintenance –Incentivized contribution: Reputation promotion, privileged access to additional content, IP and confidentiality assurances –Cooperate internationally to reduce the digital divide: Insufficient participation from global majority countries in international AI governance discourse can lead to significant knowledge gaps about AI risks and opportunities. Additionally, cooperation is needed to ensure that AI addresses, rather than exacerbates, current structural limitations and power imbalances for the global majority related to infrastructure, data and talent.57 Advancing Responsible AI Innovation: A Playbook 23

Advancing Responsible AI Innovation A Playbook 2025