Centre for Trustworthy Technology

A Global Alliance for AI Safety- Building a Foundation of Trust

The launch of the International Network of AI Safety Institutes (AISIs) during the Seoul Summit marked a pivotal milestone in the global effort to advance safe and trustworthy AI. In November 2024, the inaugural convening of the Network, hosted by the United States (U.S.), brought together representatives from Australia, Canada, the EU, Japan, Kenya, South Korea, Singapore, and the United Kingdom (U.K.), underscoring a shared commitment to aligning governance approaches on AI safety. The event highlighted the urgency of addressing AI safety, beyond existing national AISIs, in a way that does not harm innovation.

The rapid evolution of AI technologies has amplified the need for a unified global approach to safety. In recent years, governments, civil society, and industry leaders have stepped up efforts to tackle these challenges by creating specialized agencies, recruiting technical and policy talent, and launching new initiatives. However, despite preliminary progress, the current AI governance landscape faces several challenges. The AI governance space is increasingly crowded, often reactive, and struggling to keep pace with rapid technological developments. A lack of coordination among stakeholders, coupled with uneven resource distribution, has led to inconsistent global standards.

In this context, the International Network can play a key role, serving as a platform for collaboration, knowledge-sharing, and the development of unified and robust governance frameworks. Beyond mitigating adverse risks, the International Network emphasizes safety as a driver of adoption and innovation through strengthening public trust.

A Shared Mission for AI Safety

The creation of the International Network reflects a growing recognition that AI risks transcend national borders, requiring international solutions. From mitigating existential risks to addressing systemic biases, AI safety has evolved into a multidisciplinary field that spans technical, policy, and societal dimensions.

AI safety research can be broadly categorized into two areas:

  • Technical Safety focuses on improving the internal mechanisms of AI systems to prevent malfunctions or unintended outcomes.
  • Process-Based Safety is concerned with the policies, practices, and procedures that surround AI,

These areas align with the broader mission of national AISIs: advancing the science of AI safety through interoperability, and a shared understanding of risk. By uniting national safety efforts, the network seeks to accelerate progress and ensure equitable outcomes across diverse regional contexts.

Core Values of the International Network of AISIs

The Network builds on growing global initiatives for trustworthy technology, like the United Nations Global Digital Compact, the Global Partnership on AI (GPAI), and G7 efforts. The Network’s inaugural convening built on the themes of recent initiatives, with its mission statement focusing on four key priorities:

  1. Advancing Scientific Research: The Network outlined a research agenda to mitigate synthetic content risks and improve digital content transparency, with over $11 million in funding pledged for this effort, alongside new guidelines for safeguarding against harmful content.
  2. Joint Testing Exercises: The Network’s first joint exercise, led by experts from U.S. AISI, U.K. AISI, and Singapore’s Digital Trust Centre, tested Meta’s Llama 3.1 405B on multilingual AI safety, identifying key challenges and strategies to improve international AI evaluations.
  3. Shared Guidance: A joint statement was issued on AI risk assessments, aligning on a shared framework based on six principles: actionable, transparent, comprehensive, multistakeholder, iterative, and reproducible, advancing global AI safety efforts.
  4. Inclusive Information-Sharing: The AISI has an intentional focus on technical expertise and scientific collaboration through knowledge sharing. This goal is aligned with the national AISI’s which individual focuses on recruiting technical talent and supporting federal governments with ‘in-house’ technology expertise.

Common Understanding for Global Dialogue and Policy Harmonization

One of the main challenges in the AI governance landscape is the lack of uniformity in risk classifications and safety definitions.  The current ecosystem contains a patchwork of terminologies, classifications, and approaches to AI safety, leading to confusion and inefficiencies. Academic institutes have prioritized risk classification and multi-lateral organizations called for a common understanding to be a first step towards effective collaboration and redress of AI safety concerns. (Read our blog) In many ways, the initial priority of creating a common lexicon is the foundation for swift collaboration, effective policy-making, and global regulatory interoperability. As such, the international network of AISIs presents a unique opportunity for early alignment toward a common understanding of AI safety and risk classification.

From the National to the Global

AI safety efforts must be tailored to local priorities while aligning with global goals. AISIs are tasked with translating international safety standards into actionable national policies that cater to local needs and concerns. Given the diverse regulatory environments and the varying capacities of different nations, AISIs must work to harmonize their approaches without compromising on the distinct needs and regulatory landscapes of individual countries. For example, while the EU’s AI office has enforcement powers under the AI Act, others, like the U.K.’s AI Safety Institute, have a more advisory role, underscoring the need for flexible frameworks that allow for diverse implementation strategies. The ability of the AISI network to bridge these differences while maintaining a unified approach will be crucial to ensuring effective global governance.

As the network prepares for the AI Action Summit in Paris in February 2025, it has outlined an ambitious roadmap to establish benchmarks for international collaboration. These include aligning technical research agendas, scaling testing exercises, and defining a cohesive framework for regulatory interoperability.

Looking Ahead: The Promise of the AISI Network for a Trustworthy AI Future

The inaugural convening of the international network of AISIs concludes a year of momentous progress in global collaboration to design a trustworthy technology future. At the core of the network’s messaging lies the crucial relationship between safety and trust. As highlighted during the inaugural convening, “safety breeds trust, which speeds up adoption and leads to more innovation.” This central premise underscores the essential role of trust in fostering a virtuous cycle of innovation. Societal trust is earned, fostered, and upheld through resilient and inclusive safety measures. AI development which prioritizes safety not only protects against risk but fosters innovation.

By fostering a robust and collaborative global AI safety framework, the AISI network has the potential to shape the future of AI—one that is safe, ethical, and trustworthy. Through a concerted effort, aligned governance, and shared research, the network is laying the groundwork for a future in which AI technologies can thrive in a responsible, accountable, and beneficial manner.

The launch of the International Network of AI Safety Institutes (AISIs) during the Seoul Summit marked a pivotal milestone in the global effort to advance safe and trustworthy AI. In November 2024, the inaugural convening of the Network, hosted by the United States (U.S.), brought together representatives from Australia, Canada, the EU, Japan, Kenya, South Korea, Singapore, and the United Kingdom (U.K.), underscoring a shared commitment to aligning governance approaches on AI safety. The event highlighted the urgency of addressing AI safety, beyond existing national AISIs, in a way that does not harm innovation.

The rapid evolution of AI technologies has amplified the need for a unified global approach to safety. In recent years, governments, civil society, and industry leaders have stepped up efforts to tackle these challenges by creating specialized agencies, recruiting technical and policy talent, and launching new initiatives. However, despite preliminary progress, the current AI governance landscape faces several challenges. The AI governance space is increasingly crowded, often reactive, and struggling to keep pace with rapid technological developments. A lack of coordination among stakeholders, coupled with uneven resource distribution, has led to inconsistent global standards.

In this context, the International Network can play a key role, serving as a platform for collaboration, knowledge-sharing, and the development of unified and robust governance frameworks. Beyond mitigating adverse risks, the International Network emphasizes safety as a driver of adoption and innovation through strengthening public trust.

A Shared Mission for AI Safety

The creation of the International Network reflects a growing recognition that AI risks transcend national borders, requiring international solutions. From mitigating existential risks to addressing systemic biases, AI safety has evolved into a multidisciplinary field that spans technical, policy, and societal dimensions.

AI safety research can be broadly categorized into two areas:

  • Technical Safety focuses on improving the internal mechanisms of AI systems to prevent malfunctions or unintended outcomes.
  • Process-Based Safety is concerned with the policies, practices, and procedures that surround AI,

These areas align with the broader mission of national AISIs: advancing the science of AI safety through interoperability, and a shared understanding of risk. By uniting national safety efforts, the network seeks to accelerate progress and ensure equitable outcomes across diverse regional contexts.

Core Values of the International Network of AISIs

The Network builds on growing global initiatives for trustworthy technology, like the United Nations Global Digital Compact, the Global Partnership on AI (GPAI), and G7 efforts. The Network’s inaugural convening built on the themes of recent initiatives, with its mission statement focusing on four key priorities:

  1. Advancing Scientific Research: The Network outlined a research agenda to mitigate synthetic content risks and improve digital content transparency, with over $11 million in funding pledged for this effort, alongside new guidelines for safeguarding against harmful content.
  2. Joint Testing Exercises: The Network’s first joint exercise, led by experts from U.S. AISI, U.K. AISI, and Singapore’s Digital Trust Centre, tested Meta’s Llama 3.1 405B on multilingual AI safety, identifying key challenges and strategies to improve international AI evaluations.
  3. Shared Guidance: A joint statement was issued on AI risk assessments, aligning on a shared framework based on six principles: actionable, transparent, comprehensive, multistakeholder, iterative, and reproducible, advancing global AI safety efforts.
  4. Inclusive Information-Sharing: The AISI has an intentional focus on technical expertise and scientific collaboration through knowledge sharing. This goal is aligned with the national AISI’s which individual focuses on recruiting technical talent and supporting federal governments with ‘in-house’ technology expertise.

Common Understanding for Global Dialogue and Policy Harmonization

One of the main challenges in the AI governance landscape is the lack of uniformity in risk classifications and safety definitions.  The current ecosystem contains a patchwork of terminologies, classifications, and approaches to AI safety, leading to confusion and inefficiencies. Academic institutes have prioritized risk classification and multi-lateral organizations called for a common understanding to be a first step towards effective collaboration and redress of AI safety concerns. (Read our blog) In many ways, the initial priority of creating a common lexicon is the foundation for swift collaboration, effective policy-making, and global regulatory interoperability. As such, the international network of AISIs presents a unique opportunity for early alignment toward a common understanding of AI safety and risk classification.

From the National to the Global

AI safety efforts must be tailored to local priorities while aligning with global goals. AISIs are tasked with translating international safety standards into actionable national policies that cater to local needs and concerns. Given the diverse regulatory environments and the varying capacities of different nations, AISIs must work to harmonize their approaches without compromising on the distinct needs and regulatory landscapes of individual countries. For example, while the EU’s AI office has enforcement powers under the AI Act, others, like the U.K.’s AI Safety Institute, have a more advisory role, underscoring the need for flexible frameworks that allow for diverse implementation strategies. The ability of the AISI network to bridge these differences while maintaining a unified approach will be crucial to ensuring effective global governance.

As the network prepares for the AI Action Summit in Paris in February 2025, it has outlined an ambitious roadmap to establish benchmarks for international collaboration. These include aligning technical research agendas, scaling testing exercises, and defining a cohesive framework for regulatory interoperability.

Looking Ahead: The Promise of the AISI Network for a Trustworthy AI Future

The inaugural convening of the international network of AISIs concludes a year of momentous progress in global collaboration to design a trustworthy technology future. At the core of the network’s messaging lies the crucial relationship between safety and trust. As highlighted during the inaugural convening, “safety breeds trust, which speeds up adoption and leads to more innovation.” This central premise underscores the essential role of trust in fostering a virtuous cycle of innovation. Societal trust is earned, fostered, and upheld through resilient and inclusive safety measures. AI development which prioritizes safety not only protects against risk but fosters innovation.

By fostering a robust and collaborative global AI safety framework, the AISI network has the potential to shape the future of AI—one that is safe, ethical, and trustworthy. Through a concerted effort, aligned governance, and shared research, the network is laying the groundwork for a future in which AI technologies can thrive in a responsible, accountable, and beneficial manner.

Related Blogs

AI Agents at the Forefront

For decades, popular culture has envisioned Artificial Intelligence (AI) as autonomous machines capable of reasoning and even feeling like humans. However, today’s AI reality remains

Read More
Scroll to Top