Building a Global Framework for Safe AI Innovation: The Inaugural Meeting of the International Network of AI Safety Institutes
The rapid advancement of artificial intelligence (AI) presents both immense opportunities and significant risks. In response to these challenges, a new initiative, the International Network of AI Safety Institutes (AISI), has been launched to foster international collaboration and research on AI safety. Co-hosted by the U.S. Department of Commerce and the U.S. Department of State, this inaugural meeting in San Francisco marks a critical step in global efforts to ensure that AI technologies are safe, secure, and trustworthy.
A Global Effort to Address AI Safety Challenges
The AISI aims to promote international cooperation by aligning technical expertise, facilitating research, and developing best practices to manage the growing risks associated with AI. Comprising 10 initial members—including the U.S., Australia, Canada, France, Japan, the European Union, and others—the Network is set to act as a global forum to share knowledge and expertise on AI safety. The U.S. will serve as the inaugural chair, with the goal of leveraging the combined technical capacity of the member nations to harmonize approaches to AI safety and mitigate risks effectively.
Key Developments Announced at the Convening
At the heart of this convening, several pivotal announcements were made that will shape the future of AI governance. First, the Network unveiled its joint mission statement, emphasizing the importance of cultural and linguistic diversity in AI safety efforts and underscoring the need for globally interoperable principles and best practices. The statement reflects a shared commitment to advancing AI safety across all stages of technological development, ensuring that the benefits of AI are shared equitably among countries.
Additionally, the AISI announced a joint research agenda on synthetic content. With the rise of generative AI, creating synthetic content—ranging from digital images to deepfake videos—has become alarmingly easy. This capability poses substantial risks, from fraud to the distribution of harmful content such as child sexual abuse material and non-consensual imagery. In response, the Network has secured over $11 million in funding to research ways to mitigate these risks and develop safeguards against harmful AI-generated content.
Multilateral Testing and Risk Assessments
A critical area of focus for the AISI is the development of risk assessment frameworks for advanced AI systems. At the convening, members agreed on six key principles for conducting AI risk assessments: they must be actionable, transparent, comprehensive, multistakeholder, iterative, and reproducible. These principles will form the basis for future international alignment on AI risk evaluations, ensuring that AI systems are tested thoroughly across various jurisdictions.
The Network’s first multilateral testing exercise focused on AI models, particularly Meta’s Llama 3.1 405B, testing topics like academic knowledge, hallucinations, and multilingual capabilities. This exercise is a precursor to more comprehensive international testing efforts, intended to foster reproducible and robust safety evaluations of AI technologies.
Strengthening Global Research and Collaboration
To support this framework, the U.S. AI Safety Institute (US AISI) and other AISI members have committed substantial resources to foster research on AI safety. The U.S. Agency for International Development (USAID) is contributing $3.8 million toward AI safety research, with a focus on mitigating synthetic content risks. Other governments, including Australia and South Korea, have pledged funding to enhance research on detecting and preventing AI-related harm, including the development of model safeguards and transparency methods.
This multi-pronged approach—spanning technical, social, and humanistic research—will form the foundation of global collaboration in AI safety, enabling nations to share knowledge, avoid duplicative efforts, and establish clear, actionable frameworks for mitigating risks.
National Security and AI
Beyond safety and ethical concerns, the U.S. government is also focused on the national security implications of AI. The Testing Risks of AI for National Security (TRAINS) Taskforce was announced to coordinate research and testing of AI models that could impact critical national security domains, such as cybersecurity, chemical and biological security, and military capabilities. This taskforce aims to ensure that AI development does not inadvertently undermine U.S. national security and that AI technologies are used responsibly in sensitive contexts.
Looking Ahead: The AI Action Summit
The discussions at this inaugural convening will set the stage for the AI Action Summit hosted by France in February 2025. The global community of AI developers, policymakers, and safety experts will come together to further align efforts on the critical issues of AI safety and governance. With a shared commitment to collaboration, transparency, and responsible AI innovation, the International Network of AI Safety Institutes is poised to play a central role in shaping the future of AI safety on the global stage.
Conclusion: A Call for Global Cooperation
The launch of the AISI marks a significant step forward in the global effort to address the risks and challenges posed by AI. As AI technologies continue to evolve, fostering international cooperation will be essential in ensuring their safe and ethical deployment. By aligning research agendas, establishing clear guidelines, and developing robust risk assessment frameworks, the Network aims to create a safer, more secure future for AI innovation. Legal professionals, policymakers, and stakeholders across sectors will play an essential role in supporting these efforts, ensuring that AI technologies benefit society while minimizing potential harms.