Page 24 - AI for Good - Impact Report

AI for Good

Other approaches at the global level mostly focus on GenAI and its risks. In accordance, the G7
countries (Canada, France, Germany, Italy, Japan, the United Kingdom, and the United States),
and the EU as an observer have taken a proactive approach to establishing principles for AI
governance through the G7 Code of Conduct (CoC). Moderated by the Japanese G7 presidency,
the document is termed the “Hiroshima process.” The G7 CoC is a voluntary framework that sets
high expectations for the responsible development and deployment of AI technologies. Key
86
components of the G7 CoC include encouraging transparency and accountability throughout
the AI lifecycle, from development to deployment. The Code also promotes risk management
strategies that anticipate and mitigate potential harms, such as biases in AI decision-making or
vulnerabilities that could be exploited by cyberattacks. Furthermore, the Code touches upon
themes such as incident reporting, watermarking, data privacy and intellectual property as well
as common research and the development and use of international technical standards. While
this allows for flexibility and innovation, it also relies heavily on the willingness of companies
and governments to adopt and enforce these principles.
A series of “AI Safety Summits” were kicked off by the United Kingdom government in November
2023, hosting the Bletchley Park Summit. For the first time governments and companies on
a global scale came together to discuss mitigating risks of AI technologies, specifically so-
called frontier models, the most advanced and sophisticated AI models. Governments and
AI companies recognized their shared responsibility in ensuring the safety of these models,
particularly in areas critical to national security, societal well-being, and public safety. At the end
of the Summit, 29 governments and international institutions signed the Bletchley Declaration.
The declaration reinforces the global regulatory framework by addressing the safe use of
AI. Moreover, it focuses on a set of concrete measures aimed at enhancing the safety and
responsible development of frontier AI technologies. This approach emphasizes the shared
responsibility between governments and AI model developers to ensure that these technologies
are rigorously assessed for potential risks, particularly those that could impact national security,
public safety, and societal well-being. In addition to safety testing, the declaration also calls for the
development of shared international standards and best practices for AI governance. A further
measure outlined in the declaration is the establishment of a collaborative framework for testing
AI models both before and after deployment. The initiatives agreed upon at Bletchley Park,
including the establishment of the United Kingdom’s AI Safety Institute, lay the groundwork
87
for ongoing international collaboration in AI governance, ensuring that the development of AI
technologies can proceed safely and responsibly. A further summit took place in the Republic
of Korea in May 2024 . Since then, other governments including the United States , Canada ,
89
90
88
Japan , the Republic of Korea, and Singapore have announced the establishment of AI Safety
91
93
92
Institutes. In February 2025 France will be hosting the next session of the AI Safety summits –
coined the AI Action Summit. 94
In 2023, the OECD revised its definition of AI systems to reflect the latest advancements in
technology, providing a foundational framework that governments can use to legislate and
regulate AI. This updated definition not only enables harmonization among national policies
but also contributes to the development of cohesive global policy frameworks for AI. The EU
95
AI Act – elaborated in the following sections – aligned its definition with the one drafted by the
OECD.

These three approaches—the UN-led ethical frameworks, the G7’s voluntary code of conduct,
and the practical safety measures from the AI Safety Summits, accompanied by the OECD’s
definition of AI—demonstrate diverse yet complementary strategies for global AI governance.

14