International Trends in AI Safety and Governance - Tokyo College

International Trends in AI Safety and Governance

2024.03.28 @ 10:00 – 12:00

Event Report

International Trends in AI Safety and Governance

 On March 28, 2024, the University of Tokyo's Institute for Future Initiatives and the University of Tokyo's Tokyo College held a public event titled “International Trends in AI Safety and Governance.” The event was held at the SMBC Academia Hall in the International Academic Building at the Hongo Campus of the University of Tokyo and was attended by 178 people online and about 30 people at the venue.

 While discussions on the safety of AI have been developing domestically and internationally with the spread of generative AI, there are various types of discussions on "safety" and their countermeasures. In addition, with the AI Safety Institute being established in the U.K., U.S., and Japan, it is important to understand the types of "safety" and countermeasures specific to Japan as a basis for future international collaboration. The event was attended by experts on AI governance from overseas, and discussions were held on international AI safety and governance trends.

Speakers and Panelists

Merve Hickok: President and Research Director at Center for AI & Digital Policy (CAIDP)

Cyrus Hodes: Lead, SAFE project at the Global Partnership on AI (GPAI)

Inma Martinez: Chair of the Multi-stakeholder Experts Group, Global Partnership on AI (GPAI)

Michael Sellitto: Head of Global Affairs at Anthropic

Yoichi Iida: Special Negotiator for Information and Communications International Strategy, International Strategy Bureau, Ministry of Internal Affairs and Communications

Hideaki Shiroyama: Professor, Institute for Future Initiatives, The University of Tokyo

Arisa Ema: Associate Professor, Tokyo College, University of Tokyo (Moderator)

(1) Opening remarks

 Professor Hideaki Shiroyama of the University of Tokyo's Institute for Future Initiatives first gave opening remarks. Focusing on the governance of emerging technologies, Professor Shiroyama explained how the Institute's Technology Governance Research Unit has contributed to international discussions on AI safety through its research on Risk Chain Models and participation in the GPAI. In light of recent rapid changes such as the explosive spread of AI and the G7 Hiroshima AI Process, he expressed his hope that this event will serve as a catalyst for discussion in the context of Japan, given the current need to organize issues related to the safety of AI and to develop a system to address them in a manner that is relevant to each site.

(2) Topics from presenters

 First, as an introduction to the discussion points from the panelists, Ms. Inma Martinez of the GPAI mentioned the GPAI's emphasis on equity and inclusion of vulnerable peoples as "AI for all," and the leadership role Japan has played in the GPAI discussion, including these issues. In addition, she introduced that the most recent GPAI activities have been conducted with an emphasis on consensus building toward the realization of common values, a feature of the G7 Hiroshima AI process.

 Ms. Martinez also explained that AI is not limited to automation, but will impact and transform all industrial sectors, and that while the "safety" of such AI can be interpreted in culturally diverse ways around the world, the "trustworthiness" of such AI is "technically functional," and that there is a consensus. It was then emphasized that while the GPAI seeks to build consensus, definitions on AI are no longer valid and should be in line with each country's culture and values and should not be monocultured.

 Next, Mr. Cyrus Hodes, also from GPAI, stated that GPAI is working with multi-stakeholders to guarantee the safety of generated AI.

 Mr. Hodes then noted that one of the risks of generative AI is that as AI systems become more sophisticated, which is bringing risks of misalignment, control and robustness of these systems and where tools addressing these raising issues will become increasingly important (such as audits, evaluations, cybersecurity red-teaming) and where an infrastructure for such alignment needs to be established, he expressed hope for collaboration with the AI Safety Institute. In addition, he mentioned that he expects Japan to cooperate in mapping the various set of tools developed by the global community and contribute to international coordination on AI safety.

 Ms. Merve Hickok of CAIDP then spoke, first explaining that the Center is tasked with providing recommendations on AI policy to governments and international organizations, and training of future AI policy leaders. She then introduced the current state of AI policy in the U.S., which is consistent across Administrations, and the development of binding presidential executive orders for government agencies and voluntary guidelines that can also be used in the private sector. She also explained that the bipartisan agreement on the need for AI regulation in the U.S. is a reflection of the failure to regulate harmful impact of social media. She noted the AI Safety Institute was established in the U.S., and that which ministry is in charge of this type of organization indicates what the nation is focusing on, she stated that in the U.S., unlike in UK, the definition of "safety" is broad and includes the economy and current risks of AI, and therefore, the Department of Commerce is in charge of this type of organization. In addition, she introduced recent initiatives such as the AI Safety Summit by the UK, upcoming AI Summit in France, and the Mini-Virtual Summit in South Korea.

 Ms. Hickok emphasized the importance of "interoperability" to avoid governance fragmentation. However, she also warned about reducing the protections to a minimum number of common elements in the discussion of AI and human rights. She underlined the importance of international collaboration with multi-stakeholder participation, and advancing the elements of the Hiroshima AI process.

 Finally, Mr. Michael Sellitto of Anthropic introduced the company's Responsible Scaling Policy, introduced that under the Responsible Scaling Policy, an AI Safety Levels (ASL), analogous to the biosafety level, is set and safety and security measures are taken according to the degree of risk. He also said that calls for a moratorium on AI development should not be based on abstract risks, but should be considered only when there is concrete evidence that safety or security measures may be insufficient.

 Mr. Sellitto also praised the international code of conduct developed during the Hiroshima AI process as a highly effective framework, and expressed hope that the public and private sectors will work together to monitor commitments and thus increase confidence in the code.

(3) Panel Discussion

 Following the introduction of the above issues, Mr. Iida, Special Negotiator for International Information and Communications Strategy, International Strategy Bureau, Ministry of Internal Affairs and Communications, and Prof. Shiroyama joined a panel discussion moderated by Associate Prof. Ema on the topic of “What is expected of Japan in AI governance.”

 First, Mr. Iida expressed his appreciation for the substantial presentations, including the introduction of overseas case studies, as well as his compliments for the ambitious efforts by each stakeholder to address AI safety. Mr. Iida also stressed the importance of ensuring commonality and interoperability in the diversity of AI policies, while pointing out that even among developed countries, there is still diversity, especially in approaches, as indicated by the comments of each speaker. He also noted Anthropic's voluntary efforts and willingness for international collaboration, which he appreciated and felt reinforced by such efforts.

 Professor Shiroyama posed the question as a discussant, what is safety and why is it important? He then asked for further views on two points: what are the new risks posed by advanced and generative AI that differ from conventional AI, and what does the existence of bipartisan consensus and differences in competent ministries mean when comparing AI policies in different countries?

 In response to the issues raised by Professor Shiroyama, Mr. Sellitto first responded that while there are a wide range of concerns and risks surrounding AI, "safety" in the context of Anthropic's focus is to ensure that AI can be used in a reliable and safe manner.

 Ms. Martinez then noted that the 21st century is the first century in which safety has been brought to all industries but pointed out that "safety" is about preventing harm, not causing harm.

 In response, Ms. Merve pointed out that while the objective function of AI is the starting point for trust and performance evaluation, it is not possible to envision all use cases for general-purpose AI. This makes it harder to manage risk and performance.

 Mr. Hodes also noted that in the era of AGI, any task can be subject to improvement by AI, but values must be maintained by adjusting AI systems in such a society.

 In response to these issues, Mr. Iida explained how the Hiroshima Process was launched to discuss the risks of generative AI but was later added to cover infrastructure systems and advanced AI as well. He also recognized that in international discussions, "safety" and "trust" have been discussed simultaneously, and that discussions on the definition of safety have been avoided, and that a detailed definition is needed in the course of taking concrete measures in the future.

 Associate Professor Ema, the moderator of the session, also raised the point that discussions on safety should be framed not only in terms of the safety of AI itself, but also in terms of the safety realized by AI, such as its use in law enforcement agencies, and the trade-off relationship with other values.

 In response, Mr. Iida noted that both Professor Shiroyama's and Associate Professor Ema's points of view are extremely important, but also expressed the view that the gap between political and administrative actors in terms of minimizing risk while advancing technology-based innovation is not so large. Mr. Iida also reiterated the importance of a multi-stakeholder approach in the AI policy-making process.

 Mr. Hodes agreed with Mr. Iida, pointing to the composition of the U.S. and China as the two giants, and praised Japan's efforts, such as the establishment of the AI Safety Institute, and expressed hope that Japan would play a coordinating role.

 Ms. Merve, while noting the differences in authority among ministries, emphasized the importance of a multi-stakeholder approach because of the need for diverse competencies, and praised Japan’s work to drive commonalities across actors.

 Ms. Martinez noted that even in Europe, the development of regulations pertaining to the Internet has been slow but said that regulations on AI have been developed under a global consensus based on principles, values, and commonalities, taking into account Japan's recommendations.

 Mr. Sellitto noted that in the early stages of technology development, there can be concern that regulations will hinder innovation, but people will gradually learn what to regulate, and that Anthropic's ASL was also a practice of first developing and implementing commitments and then publishing the lessons learned from them, and he hopes that this will lead to the development of best practices that can inform regulations in the future.

(4) Questions from an audience

 In response to a question from an online participant about what is needed to ensure the safety and reliability of AI, given that Japan has been the target of cyber-attacks in recent years, Mr. Sellitto explained that while there are currently no clear guidelines for AI cyber security, he explained that cybersecurity standards are being formed. Ms. Martinez also expressed the view that there have been many cyberattacks targeting AI, we can learn from them to increase resilience.

(5) Summary and closing remarks

 In concluding the event, Professor Shiroyama summarized the discussions and pointed out the need to organize a common vocabulary and know-how for "safety," although it seems better not to dare to establish a detailed definition. He also suggested that the dichotomy of hard law/soft law for regulation of AI is too simplistic, and that the learning process needs to start with abstract principles and shared experiences.

 In addition to thanking the participants, Associate Professor Ema mentioned the need to adhere to an agile process for AI security and safety, and ultimately AI governance, in the face of rapid technological innovation.

 Finally, Prof. Takeo Hoshi, Deputy Director of Tokyo College at the University of Tokyo, gave closing remarks. Prof. Hoshi pointed out the importance of today's discussion, and expressed the pleasure for Tokyo College to host this event together with the Institute for Future Initiatives. Drawing on debates on regulatory attempts to prevent financial crises, which is one of his areas of expertise, he stated that financial crises have been happening despite the various efforts to build sound and safe financial systems. here seem to be no regulatory mechanisms that make the financial systems completely safe.  The lesson is that, in addition to trying to prevent crises, we need to be ready to respond. Prof. Hoshi concluded the event by noting the need to prepare for AI crises while promoting human-centered AI development, and expressed his hope that today's discussion would serve as a starting point for future discussions.

Zoom Webinar
Date(s) Thursday, 28 March 2024, 10:00-12:00 JST

Zoom Webinar (In-person registration here, Online registration here)

Registration Advance registration is required. *Registration will close when we reach maximum capacity. *The Zoom URL will be emailed to those who register for the event on March 27.
Language English (Japanese interpretation)

With the proliferation of generative AI, discussions on the safety of AI are unfolding domestically and internationally. However, there are various types of discussions on “safety,” and corresponding measures. Moreover, what is considered “safe” or what threats and risks are emphasized can differ depending on the situation and context in each country. While AI Safety Institutes have been established in the UK, US, Japan and to be established in other countries, grasping the types of “safety” specific to Japan and its corresponding measures will become increasingly important as a foundation for future international cooperation.
In this event, we will host discussions with experts on overseas AI governance to delve into the trends in international AI safety and governance.



Opening Remarks

Hideaki Shiroyama (Institute for Future Initiatives, The University of Tokyo)



Introduction from panelists

Introduction from each panelist on current important issues in AI governance



Panel Discussion : What are the expectations for Japan?

Merve Hickok (President and Research Director at Center for AI & Digital Policy (CAIDP))

Cyrus Hodes (Lead, SAFE project at the Global Partnership on AI)

Inma Martinez (Chair of the Multi-stakeholder Experts Group, Global Partnership on AI)

Michael Sellitto (Head of Global Affairs, Anthropic)






EMA Arisa (Associate Professor, Tokyo College and Institute for Future Initiatives, The University of Tokyo)


Speaker Profile

Merve Hickok, President and Research Director at Center for AI & Digital Policy (CAIDP)
Merve Hickok is the President and Research Director at Center for AI and Digital Policy (CAIDP), advising governments and international organizations on AI policy and regulation. She is a globally renowned expert on AI policy, ethics and governance. Her contributions and perspective have featured in The New York Times, Guardian, CNN, Forbes, Bloomberg, Wired, Scientific American, The Atlantic, and Politico. Her work focuses on impact of AI systems on individuals, society, public and private organizations – with a particular focus on fundamental rights, democratic values, and social justice. Merve is also the founder of She is the Data Ethics lecturer at University of Michigan School of Information, and the Responsible Data and AI Advisor at Michigan Institute for Data Science.


Cyrus Hodes, Lead, SAFE project at the Global Partnership on AI
Cyrus Hodes is a co-founder of Stability AI a leading generative AI platform, which he exited to launch (AIGC Chain), the first foundation model of AI generated content on a blockchain. He is a General Partner at 1infinity Ventures, a global fund investing in responsible AI ventures. Cyrus leads the Safety and Assurance of Generative AI (SAFE) project at the Global Partnership on AI, responding to the G7 Hiroshima AI Process. He previously co-founded and chaired the AI Initiative at The Future Society—a think tank incubated at Harvard Kennedy School. Cyrus is a member of the OECD Expert Group on AI and a Board member of Intelmatix (Saudi’s largest AI company). Educated at Sciences Po Paris, M.A. Paris II University, M.P.A. Harvard.


Inma Martinez, Chair of the Multi-stakeholder Experts Group, Global Partnership on AI
Inma Martinez is technology pioneer and AI scientist who advises leaders in business and government on technology as competitive advantage and contribution to societal progress. She was a pioneer of digital technologies and AI in the 2000s and has combined her career in innovation with advisory appointments at government agencies in the United Kingdom (UKTI and the Innovation Fund of the Department of Sport, Media and Culture), Spain (State Secretariat for Artificial Intelligence at the Ministry of Economy and Digital Transformation) as well as provided expert testimonies across various technology boards at the European Commission since 2002. She has collaborated with the United Nations Industrial Development Organisation (UNIDO) highlighting the implications of the 4IR for developing countries in the post-pandemic world and is a UNESCO Ambassador for Intercultural Values in AI. She is a guest lecturer at Imperial College Business School in London and a published author of scientific books and research papers on emerging technologies.


Michael Sellitto, Head of Global Affairs, Anthropic

Michael Sellitto is the Head of Global Affairs at Anthropic, an AI safety and research company. He is also an Adjunct Senior Fellow in the Technology and National Security Program at the Center for a New American Security, and a Member of the Council on Foreign Relations.

Prior to joining Anthropic, Michael was the founding Deputy Director of the Stanford Institute for Human-Centered Artificial Intelligence (HAI), which is dedicated to advancing AI research, education, policy, and practice to improve the human condition. As HAI’s first staff member, he was instrumental in designing and executing the Institute’s strategic plans and establishing HAI’s global reputation among policymakers.

Michael served in the White House as Director for Cybersecurity Policy on the National Security Council staff from 2015-2018. He led international engagement on cybersecurity policy and strategy, promoted international adoption of a framework for strategic stability in cyberspace, and advanced issues related to the digital economy and Internet governance. Before that, Michael served as Special Assistant to Deputy Secretaries of State William J. Burns and Antony Blinken, advising the Deputies on political, energy, security, and trade issues related to South and Central Asia and on worldwide cyber policy and counterterrorism strategy.

Organized by Institute for Future Initiatives, The University of Tokyo Tokyo College, The University of Tokyo
Contact E-mail: ifi_tg[at] (replace [at] with @)

Upcoming Events

Panel discussion “The Economy of Japan Viewed from the Outside” (Speakers: Prof. Takatoshi ITO, Prof. Nobuhiro KIYOTAKI)

イベント予定パネルディスカッション/Panel discussion

Friday, 23 August 2024, 16:00-17:30 JST

This panel discussion will feature two distinguished Japanese economists from overseas. They will discuss the current situation of the Japanese economy and the challenges it faces. This discussion will provide new insights into the Japanese economy from an international perspective.

Previous Events

Peace, security and Artificial Intelligence


Friday, 12 July 2024, 14:00-15:00

This lecture will delve into the inherent risks that AI systems pose across the broader security domain, which are mentioned above, and will conclude with some insights on proposed governance models to prevent and mitigate the risks associated with these technologies. The afore include the need to elaborate binding norms, standards, and guidelines, as well as oversight, monitoring, validation and verification functions through a centralised authority with the appropriate mechanisms to enforce these regulations and ensure compliance through accountability, remedies for harm and emergency responses.

Book Launch “The Faraway Sky of Kyiv. Ukrainians in the War” (Lecture by Dr. Olga KHOMENKO)


Friday, 28 June 2024, 15:30-16:30

On July 25, 2023, Chuo Koron Shinsha published Dr. Komenko's book, 'The Faraway Sky of Kyiv. Ukrainians in the War', offering a unique perspective on the war in Ukraine.
This book originated from her experience of the war in Ukraine and stories from family members, friends, and former students. Her motivation to write this book came from being interviewed by Japanese media in early 2022. The questions she was asked lacked general knowledge of Ukrainian history and culture; therefore, she decided not to give any further interviews and to focus on writing in Japanese to provide a voice for Ukrainians instead.

A Cultural History of Hacking (Lecture by Prof. Federico MAZZINI)


Monday, 24 June 2024, 15:00-15:45

The traditional historical narrative locates the birth of hacker culture in US universities in the 1960s. This talk will look at hackers as part of a longer chronology, beginning with science fiction novels at the end of the 19th century, continuing with radio hams in the 1910s and "phone phreaks" in the 1970s, and ending with computer hackers in the late 20th century. It will examine both what hackers and proto-hackers wrote about themselves and how they were perceived by the print media. It will show not only that hacker culture existed before computers, but also that it is an integral part of modern Western technoculture, influencing its ideas about innovation and positive human-machine relationships, as well as media coverage of technology and online communication strategies.

The Future of Globalization: A History (Lecture by Bill EMMOTT)


Tuesday, 4 June 2024, 16:00-17:30 JST

We are in an era in which globalization -- the connection of countries through trade, finance and ideas -- appears to be in retreat, as geopolitical tensions force governments to prioritize economic security and to try to "de-risk". Yet this is not the first time when globalization has been said to be reversing. By looking into history, we can understand what factors will truly determine the future course of globalization.

Family-run Medical Institutions in Japan (Lecture by Prof. Roger GOODMAN)


Thursday, 30 May 2024, 14:00-15:30 JST

Around 80% of all hospitals and around 90% of clinics in Japan are private. Of these private institutions in total, up to 75% are family-run. This lecture sets out to fill a puzzling gap in the literature by describing the development and significance of dōzoku keiei iryō hōjin in the context of how the health system as a whole operates in Japan.

Central Banks in the 21st Century (Lecture by Prof. Luiz Awazu PEREIRA DA SILVA)


Wednesday, May 29th, 2024, 15:00-16:30 JST

Central banks, and central bankers, stand at a crossroads. They face five major forks in the 21st century requiring careful reflection: (1) the re-emergence of inflation and uncertainties; (2) climate change; (3) inequality; (4) digital financial innovation; and (5) artificial intelligence. Modern central banks have always strengthened their analytical thinking when facing challenges in the past, balancing risks properly and choosing the best path. Now, these new issues imply that central banks will have to carefully identify and analyze their challenging implications.