Advancing Responsible AI Innovation A Playbook 2025

CASE STUDY 7 The Hiroshima AI Process International Code of Conduct and Reporting Framework In 2023, under Japan’s presidency, the G7 launched the Hiroshima AI Process, resulting in the Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems.72 This voluntary code promotes ethical, transparent and secure practices. To reinforce accountability, the G7 and OECD introduced a voluntary reporting framework in 2025 for organizations in member and partner countries. While initial reports were submitted, variations in detail and transparency highlighted limitations in consistency and comparability. The framework’s voluntary nature also raised challenges in participation and adherence. The G7, under its Canadian presidency, is exploring additional incentives and clearer guidance. There is also a forum led by Japan for broader collaboration through the Hiroshima AI Process Friends Group, which now comprises 56 countries and regions.73 Increasing participation by organizations across diverse jurisdictions will also require reporting requirements to consider language and timing. Key insight Commitments to voluntary frameworks alone are insufficient for ensuring transparent and accountable responsible AI practices by organizations. They likely require the layering of instruments to assess claims (see Table 2), such as standardized reporting. Instruments for reporting on responsible AI practices, by content TABLE 2 Content type Instruments Considerations (non-exhaustive) Commitments How an organization says it will implement responsible AI –Individual: Informal (blogs, speeches) or formal commitments (principles, policies, frameworks) e.g. Perplexity Acceptable Use Policy66 –Joint: Commitments from multiple organizations (see Case study 7)Advantages: –Agile method for signalling norms –Flexible to organization context Limitations: –Low adoption with varied adherence –Limited public evidence correlating responsible AI commitments with implementation67 Claims How an organization self-reports its responsible AI practice –Reports: Detailing practices e.g. Microsoft 2025 Responsible AI Transparency Report68 –Cards: Insights into the development, governance and safety of an AI model, system of models, or service e.g. Cohere Command R and Command R+ Model Card69Advantages: –Provides a benchmark for other companies –Promotes feedback and accountability Limitations: –Variability can hinder standardized comparisons across multiple companies –Self-reporting bias may occur Evidence How an organization substantiates its responsible AI practice –Certifications: A review typically aligned with a set criteria e.g. Anthropic certified by Schellman Compliance, LLC against ISO/IEC 42001:202370 –Sandboxes: Third-party controlled or monitored environments for AI testing e.g. the United Arab Emirates regulatory sandboxes71Advantages: –Provides credibility if certified by a reputable party –Incentivized adoption in pursuit of market differentiation Limitations: –Variability in certification methods risk practice fragmentation –Costly to implement and address renewal needs Advancing Responsible AI Innovation: A Playbook 26