Understanding the Hiroshima AI Process

Countries, companies and international organizations participating in the Hiroshima AI Process agreed on a 2026 action plan at the second in-person meeting of the HAIP Friends Group in Tokyo this March, marking a shift from principle-setting to practical implementation of safe, secure and trustworthy artificial intelligence and governance. 

With participation from 66 countries and 38 organizations, the HAIP plays a significant role in promoting the adoption of AI governance at the corporate level.

Understanding the HAIP requires looking at its origins, structure and voluntary reporting mechanism and exploring how this relates to the AI governance efforts privacy experts are engaged in. By combining international principles, participating governments and organizations, and a voluntary reporting framework the HAIP creates a layered model of governance that bridges state-level commitments and corporate disclosure. 

The HAIP and the HAIP Reporting Framework

The HAIP was proposed at the G7 Summit in May 2023, during which Japan served as the chair country. It was established as the first international policy framework consisting of guiding principles and a code of conduct, aimed at promoting the widespread adoption of safe, secure and trustworthy advanced AI systems. The Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems recommends a risk-based approach and identifies 11 key action areas, ranging from risk and incident management to strengthening internal governance, promoting transparency and enhancing security.

At the G7 Summit held in Italy in June 2024, the HAIP Reporting Framework was established as a tool for organizations to voluntarily report on their specific initiatives regarding the code of conduct. The HAIP Reporting Framework was announced at the AI Action Summit held in Paris in February 2025.

The reporting framework enables organizations to report on implementation of the Code of Conduct in a clear and standardized manner. Companies that complete the HAIP Reporting Framework questionnaire will be listed on the Organisation for Economic Co-operation and Development's website, and all responses will be made public. The provision of a standard form for reporting on efforts related to the HAIP Reporting Framework enables comparison and sharing among different organizations, opening the possibility that, in the long term, organizations will be able to use it to benchmark their own governance activities.

HAIP Friends Group and Partners Community

The HAIP began as a framework proposed by the G7, but it is open to countries that share its spirit at large. Countries and regions that join the Friends Group are expected to support the spirit of the HAIP's international guiding principles and code of conduct. Thus far, 66 countries and regions, including the G7, have joined the HAIP Friends Group. Given that there are 193 United Nations Member States, this means approximately one-third of them are participating, indicating the HAIP is growing into a framework with significant global influence.

In addition to the Friends Group, the HAIP also includes a Partners Community, which was launched at the first in-person meeting of the HAIP Friends Group. It was established as a voluntary framework to support the activities of the Friends Group, with participation from private-sector AI-related companies and international organizations that endorse the spirit of the HAIP. 

Through this community, the aim is to support the governments of Friends Group member countries in gaining a deeper understanding of and implementing the HAIP, while also promoting the participation of AI developers from various countries in the Reporting Framework, with the goal of realizing safe, secure and trustworthy AI on a global scale. A total of 37 private-sector AI-related companies and international organizations are participating in the Partners Community.

AI governance and the HAIP Reporting Framework

The HAIP aims to promote international cooperation on AI governance. Emphasizing the importance of all AI stakeholders bearing responsibility for promoting safe, secure and trustworthy AI, the HAIP encourages all stakeholders to read and understand the Hiroshima Process International Guiding Principles for Organizations Developing Advanced AI Systems and to fulfill their responsibilities appropriately, taking full account of their capabilities and roles throughout the AI life cycle.

The questionnaire provided for the HAIP Reporting Framework consists of seven sections corresponding to the 11 actions included in the Code of Conduct. The questions are structured to align with other governance initiatives, such as the OECD AI Principles, and organizations are encouraged to submit reports at least once a year. Regular reporting enables continuous monitoring of practices and the accumulation of insights, making it easier to measure the effectiveness and impact of the Code of Conduct.

The questions in the HAIP Reporting Framework and brief supplementary explanations cover: risk identification and evaluation — definition and/or classification of different types of risks related to AI; risk management and security — measures to address risks and vulnerabilities across the AI life cycle; transparency reporting — measures to share information with stakeholders regarding the results of evaluations of risks and impacts related to advanced AI systems; organizational governance, incident management and transparency — communication of risk management policies and practices to users and/or the public; content authentication and provenance mechanisms — use of content provenance detection, labeling or watermarking mechanisms to enable users to identify content generated by advanced AI systems; research and investment to advance AI safety and mitigate societal risks — collaboration on and investment in research to improve the current state of content authenticity and provenance; and advancing human and global interests — support for digital literacy, education or training initiatives to improve user awareness and/or help people understand the nature, capabilities, limitations and impacts of advanced AI systems.

Twenty-five companies have currently submitted and published reports.

Potential and challenges of the HAIP Reporting Framework

A key feature of the HAIP is its emphasis on policy formulation at the corporate level rather than the government level. Given that the OECD AI Principles and the U.N. Educational, Scientific and Cultural Organization's Recommendation on the Ethics of AI are aimed at governments, the HAIP can be considered a more accessible international cooperation framework for companies. 

Since participating companies answer the same set of questions and their responses are aggregated on the OECD website, the HAIP Reporting Framework also offers the benefit of allowing companies to easily investigate the status of other companies' initiatives when benchmarking their own activities. As responses to the published HAIP Reporting Framework accumulate, it may eventually become possible to deduce reasonable governance standards for companies regarding each question in the report. In other words, the HAIP Reporting Framework has the potential to form a baseline for AI governance. 

At the second in-person meeting of the HAIP Friends Group, the "HAIP Friends Group Action Plan 2026" was formulated. It outlines activities to promote AI governance — such as outreach efforts for the OECD Principles and the HAIP Reporting Framework, the hosting of seminars and workshops for knowledge sharing on AI, and the implementation of group studies to examine interoperability across various governance approaches — as well as activities to promote the societal implementation of AI, including the sharing of best practices and insights on AI implementation and initiatives to build AI capabilities. Through this Action Plan, the HAIP has become a practical program aimed at driving action among stakeholders and further solidifying the AI governance framework.

Of course, there are drawbacks. Like many outcomes of international consultations, the HAIP merely provides high-level guiding principles and a code of conduct. Aside from the point that “a risk-based approach should be adopted,” the current situation lacks specificity. Since the decision of whether to report — and, furthermore, how to report — is left to the discretion of the reporting entity, the level of detail in publicly available reports is inconsistent. This could pose a constraint for companies when benchmarking their governance activities. 

Furthermore, corporate disclosure of information may invite unnecessary scrutiny from external stakeholders. It is also possible that organizations participating in the HAIP Reporting Framework with good intentions could end up catching the eye of activists and becoming targets of attack. What, then, is the incentive for companies to publicize their initiatives even at the risk of such consequences? Since the success of the HAIP Reporting Framework depends on the diversity and number of participating companies, if this point remains ambiguous, the number of participants may not grow, and the framework could become a mere formality. It is essential to clearly articulate the benefits of participation.

Summary

The HAIP is a groundbreaking initiative that provides a venue for various stakeholders to engage in dialogue regarding AI governance. Without a place to gather and discuss, this dialogue might not happen. Since stakeholders with differing cultures, values, economic circumstances and interests are involved, progress is naturally slow. 

Nevertheless, those involved continue to strive for the better implementation of AI in society, exercising the utmost care and consideration. In these difficult times, there is hope in that such a forum exists.

While discussions within the HAIP are still ongoing, global consensus-building is steadily progressing. As regulatory frameworks for AI continue to diverge across jurisdictions and the technology rapidly advances, the HAIP offers companies an opportunity to actively participate, contribute and help realize a better implementation of AI in society.