AI Seoul Summit

The AI Seoul Summit was an international meeting on artificial intelligence governance held on 21-22 May 2024 and co-hosted by the Republic of Korea and the United Kingdom. It was the second meeting in the summit series begun at Bletchley Park in November 2023. A virtual leaders' session on 21 May was followed by an in-person ministers' session in Seoul on 22 May.^[1][2]

The summit broadened the series' agenda from frontier AI safety to three stated priorities: safety, innovation, and inclusivity. Its principal outputs were the Seoul Declaration for Safe, Innovative and Inclusive AI, the Seoul Statement of Intent toward International Cooperation on AI Safety Science, the voluntary Frontier AI Safety Commitments, and the Seoul Ministerial Statement.^{[5][11][12][14]} The Statement of Intent expressed an ambition to develop an international network of AI Safety Institutes; the network was formally launched in San Francisco six months later, rather than at the Seoul meeting itself.^[11][23]

The summit produced a common framework for government cooperation on safety science, a pledge by 16 AI companies to publish severe-risk frameworks, and a ministerial plan to develop proposals for assessing severe risks. These were voluntary political and corporate commitments, not a treaty or a binding international regulatory regime.^[12][14][19]

Key facts

Scroll sideways for more →

Background

From Bletchley Park to Seoul

The United Kingdom hosted the first AI Safety Summit at Bletchley Park on 1-2 November 2023. The resulting Bletchley Declaration, endorsed at the time by 28 countries and the European Union, described the possibility of serious or catastrophic harm from highly capable general-purpose systems and called for international cooperation on evaluation, public-sector testing capacity, and scientific research.^[3]

At the close of the Bletchley meeting, South Korea agreed to co-host a smaller virtual follow-up within six months, while France was to host the next in-person summit. Participants also commissioned an international state-of-the-science report chaired by Yoshua Bengio.^[4] Seoul therefore served as an interim meeting in a planned sequence, with a narrower leaders' format and a larger ministerial discussion.

The policy setting was also changing between the two summits. The United States was implementing Executive Order 14110, and the UK AI Safety Institute and US AI Safety Institute signed an April 2024 memorandum on shared testing methods, joint research, and information exchange.^[6][7] On 21 May, the first day of the Seoul meeting, the Council of the European Union gave final approval to the EU AI Act.^[8] The Seoul Declaration did not try to replace these national and regional efforts. It instead emphasized interoperability and referred to work at the United Nations, G7, G20, OECD, Council of Europe, and Global Partnership on AI.^[5]

Format and programme

The summit had two official sessions. On 21 May, Yoon and Sunak co-chaired a virtual leaders' session titled "Building on the AI Safety Summit: towards an Innovative and Inclusive Future." Selected industry participants were invited to report on their Bletchley commitments. On 22 May, Korean science minister Lee Jong-Ho and UK technology secretary Michelle Donelan co-chaired an in-person ministers' session at the Korea Institute of Science and Technology.^[2][10]

The second day had two substantive blocks. One covered emerging risks from current and near-future frontier models, national safety institutes, and international cooperation. The other addressed sustainability and resilience, including energy use, labor-market effects, misinformation, and bias.^[2] The Korean government also held an AI Global Forum alongside the ministerial programme. Contemporary analysis sometimes discussed the two events together, but the forum was an adjacent Korean-hosted event rather than a third official summit session.^[10][20]

Participants

The parties to the leaders' documents were Australia, Canada, France, Germany, Italy, Japan, the Republic of Korea, Singapore, the United Kingdom, the United States, and the European Union. These 10 states and the EU adopted both the Seoul Declaration and the attached Statement of Intent.^[5][11]

The final public participant list, updated on 13 June 2024, recorded 20 governments, the European Commission, the OECD, and the United Nations. It also named nine academic or civil-society organizations: the Ada Lovelace Institute, Carnegie Endowment, Chinese Academy of Science, Concordia AI, Future of Life Institute, Humane Intelligence, KAIST, Mila, and Seoul National University.^[9]

Industry and related organizations on that list included Amazon, Anthropic, Cohere, the Frontier Model Forum, G42, Google DeepMind, IBM, LG AI Research, Meta, Microsoft, Mistral AI, Naver, OpenAI, Samsung Electronics, SK Telecom, Technology Innovation Institute, Tencent, and xAI.^[9] The official list identifies organizations, not a complete roster of individual attendees. Claims about particular executives therefore require separate evidence and should not be inferred from an organization's inclusion.

The Seoul Ministerial Statement had a broader set of signatories than the leaders' documents: Australia, Canada, Chile, France, Germany, India, Indonesia, Israel, Italy, Japan, Kenya, Mexico, the Netherlands, Nigeria, New Zealand, the Philippines, the Republic of Korea, Rwanda, Saudi Arabia, Singapore, Spain, Switzerland, Türkiye, Ukraine, the United Arab Emirates, the United Kingdom, the United States, and the European Union.^[14] China took part in the in-person summit discussions but was not a signatory to the Ministerial Statement.^[15]

Agenda and proceedings

Leaders' session

The leaders' session organized its discussion around safety, innovation, and inclusivity. The co-hosts described these goals as interrelated rather than as separate negotiating tracks. The session produced the Seoul Declaration and its Statement of Intent, while 16 developers announced the Frontier AI Safety Commitments.^[1][5][12]

The Declaration supported risk-based, interoperable governance, cooperation among public safety bodies, and broader participation in international discussions. It also linked the summit process to the Hiroshima Process code of conduct, updated OECD principles, United Nations work, and other international forums.^[5]

The company commitments addressed severe risks from highly capable general-purpose AI models. They required signatories to describe risk assessment, thresholds, mitigations, internal accountability, and public transparency in an organization-level safety framework before the Paris summit. The commitments also called for internal and external red teaming, model-weight security, vulnerability reporting, and information about model capabilities and limitations.^[12]

Ministers' session

The ministers' first session considered the interim International AI Safety Report, cooperation among national institutes, and the methods used for model evaluation. The interim report had been published on 17 May, four days before the summit. It was chaired by Bengio, coordinated by the UK institute, and overseen by an expert advisory panel nominated by 30 countries, the EU, and the United Nations.^[2][16]

The second session covered sustainability and social resilience. The programme identified energy and environmental effects, employment, misinformation, and bias as topics. The Ministerial Statement later addressed resource-efficient systems and low-power chips, workforce training, access for small and medium-sized enterprises, public-sector use, digital divides, and participation by developing countries.^[2][14]

Documents and commitments

Seoul Declaration

The Seoul Declaration was the main leaders' document. It stated that safety, innovation, and inclusivity were interrelated, supported risk-based governance, and called for cooperation across public institutions. It also recognized the responsibility of organizations developing and deploying frontier systems and noted the separate corporate safety commitments.^[5]

The Declaration used broad political language rather than setting technical thresholds or creating an enforcement body. Its institutional detail came mainly from the attached Statement of Intent, while its wider contribution was to add innovation, inclusion, sustainability, and digital-divide concerns to the summit series' existing safety agenda.

Statement of Intent on safety science

The Seoul Statement of Intent committed the same leaders' group to closer cooperation on AI safety science. It described possible work on interoperable methods, information about capabilities and risks, monitoring harms and incidents, shared or jointly created evaluations and datasets, technical resources, research security, and international standards.^[11]

Its final paragraph expressed a shared ambition to develop an international network among key partners. That wording matters for the chronology: Seoul supplied the political basis for the later network, but did not itself establish the network's final membership, governance, or operating programme.^[11]

Frontier AI Safety Commitments

The initial corporate signatories were Amazon, Anthropic, Cohere, Google, G42, IBM, Inflection AI, Meta, Microsoft, Mistral AI, Naver, OpenAI, Samsung Electronics, Technology Innovation Institute, xAI, and Zhipu AI.^[12][13]

The commitments were organized around three outcomes:

1. Risk assessment and management. Developers would assess risks throughout the lifecycle, define thresholds at which unmitigated severe risks would be intolerable, specify mitigations, and describe what would happen if a threshold were reached.
2. Accountability. Developers would assign responsibilities and sufficient resources for applying and reviewing their frameworks.
3. Transparency. Developers would report publicly on implementation, subject to limits where disclosure would create risk or reveal disproportionately sensitive commercial information.

The most restrictive provision said that, in an extreme case, an organization would not develop or deploy a system if mitigations could not keep risk below its thresholds.^[12] The document described the commitments as voluntary and assigned no approval or audit role to a regulator.^[12]

The safety-framework model overlapped with company-specific approaches such as Anthropic's Responsible Scaling Policy, Google DeepMind's Frontier Safety Framework, and OpenAI's Preparedness Framework. Seoul established a common set of expected elements, but the summit record does not show that it originated each named company policy.^[12][27]

Seoul Ministerial Statement

The 27-state and EU Ministerial Statement covered a wider range of policy than the corporate commitments. On severe risks, it identified capabilities that could materially assist non-state actors with chemical or biological weapons. It also identified possible evasion of human oversight through safeguard circumvention, manipulation, deception, autonomous replication, or adaptation without explicit human approval.^[14]

Ministers planned to work with industry, academia, and civil society on proposals for severe-risk thresholds before the Paris summit. They also supported cooperation among national institutes on testing guidelines, evaluation datasets, and best practices. The statement did not itself define a numerical global threshold.^[14][15]

Other announcements

The United Kingdom announced up to £8.5 million for research on systemic AI safety during the second day. The programme was designed to study risks arising from the social and technical systems in which AI is deployed, including deepfakes and cyberattacks, rather than only the properties of a model.^[17]

The UK had announced on 20 May, immediately before the summit, that its safety institute would open a San Francisco office. This was a related national announcement, not a jointly adopted summit outcome.^[18] South Korea used the meeting to state its intention to create a domestic institute. The Korean Ministry of Science and ICT formally launched that institute on 27 November 2024.^[25]

Reception and limits

Independent reporting described Seoul as a "mini-summit" and emphasized that the company pledges were voluntary. The Associated Press treated the pause commitment and government research cooperation as the main concrete results, while noting that the meeting took place amid several overlapping efforts to govern AI.^[19]

The Center for Strategic and International Studies argued that the summit had less attendance and media attention than Bletchley but still produced substantive progress on the growth and coordination of national safety institutes. It also raised the possibility that adding innovation and inclusivity could dilute the summit series' distinctive focus on advanced-system safety.^[20] This was an analytical judgment, not a position adopted by summit participants.

Civil-society assessments were mixed. Computer Weekly reported support for international scientific cooperation, alongside calls for mandatory rather than solely voluntary commitments, more socio-technical evaluation, and wider participation by workers, affected communities, and public-interest groups. An Ada Lovelace Institute representative told the publication that participation concerns raised after Bletchley had not necessarily been remedied at Seoul.^[21] The criticism should not be confused with a 2023 open letter about the Bletchley summit; the official Seoul list did include nine academic and civil-society organizations.^[9]

Before the meeting, Chatham House researchers argued that an inclusion agenda should address local capability and stewardship in developing countries, not only access to foreign infrastructure and services.^[22] The Ministerial Statement did refer to capacity-building, digital divides, and cooperation with developing countries, but it did not create a funding mechanism for those aims.^[14]

Implementation after Seoul

The International Network of AI Safety Institutes was formally launched at an inaugural meeting in San Francisco on 20-21 November 2024. Its 10 initial members were Australia, Canada, the European Union, France, Japan, Kenya, the Republic of Korea, Singapore, the United Kingdom, and the United States. Germany and Italy had signed the Seoul Statement of Intent but were not listed as initial members of the launched network; Kenya was an initial network member despite not being a party to the leaders' statement.^[23][24]

The network adopted four priority areas: research, testing, guidance on interpreting test results, and international inclusion and information sharing. It also conducted an initial multilingual testing exercise and began work on shared approaches to risk assessment.^[23] By 2026, the UK institute described it as the "International Network for Advanced AI Measurement, Evaluation and Science," formerly the International Network of AI Safety Institutes, with the same 10 members and a narrower emphasis on measurement and evaluation science.^[26]

The official company-commitments page was updated on 7 February 2025 to add Magic, MiniMax, 01.AI, and NVIDIA, bringing the listed total to 20 organizations.^[12] A 2025 independent preprint identified 12 providers that had published frontier safety frameworks following Seoul and found substantial gaps against the authors' risk-management rubric. That study shows that publication could be assessed, but its scoring method was an independent research framework rather than a summit compliance process.^[27]

Place in the summit series

France and India co-chaired the next AI Action Summit in Paris on 10-11 February 2025. The French organizers explicitly described it as building on the UK and Korean meetings and adopted a broader "action" framing involving governments, industry, researchers, and civil society.^[28]

India then hosted the AI Impact Summit in New Delhi in February 2026, the first meeting in the series held in the Global South. The broader programme ran from 16-21 February, with the principal summit on 18-19 February. A later Indian government summary listed 92 countries and international organizations as endorsing the New Delhi Declaration on AI Impact.^[29] These later meetings broadened the series further, but they did not retroactively turn the Seoul documents into binding law.

See also

• AI governance

References

1. ^UK Department for Science, Innovation and Technology, "About the AI Seoul Summit 2024," accessed July 28, 2026. gov.uk/...about
2. ^UK Department for Science, Innovation and Technology, "AI Seoul Summit: day 1 and 2 programme," updated June 13, 2024. gov.uk/...ai-seoul-summit-day-1-and-2-programme
3. ^UK Government, "The Bletchley Declaration by Countries Attending the AI Safety Summit, 1-2 November 2023," updated February 13, 2025. gov.uk/...g-the-ai-safety-summit-1-2-november-2023
4. ^UK Government, "World leaders, top AI companies set out plan for safety testing of frontier as first global AI Safety Summit concludes," November 2, 2023. gov.uk/...-first-global-ai-safety-summit-concludes
5. ^UK Government, "Seoul Declaration for safe, innovative and inclusive AI by participants attending the Leaders' Session," May 21, 2024. gov.uk/...ders-session-ai-seoul-summit-21-may-2024
6. ^UK Government, "Collaboration on the safety of AI: UK-US memorandum of understanding," April 2, 2024. gov.uk/...-of-ai-uk-us-memorandum-of-understanding
7. ^President of the United States, "Executive Order 14110: Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence," Federal Register, November 1, 2023. federalregister.gov/...-of-artificial-intelligence
8. ^Council of the European Union, "Artificial intelligence act: Council gives final green light to the first worldwide rules on AI," May 21, 2024. consilium.europa.eu/...first-worldwide-rules-on-ai
9. ^UK Department for Science, Innovation and Technology, "AI Seoul Summit: participants list (governments and organisations)," updated June 13, 2024. gov.uk/...pants-list-governments-and-organisations
10. ^Republic of Korea Ministry of Science and ICT, "No Barriers Between Ministries in AI and Digital Diplomacy," May 10, 2024. msit.go.kr/...view.do
11. ^UK Government, "Seoul Statement of Intent toward International Cooperation on AI Safety Science," May 21, 2024. gov.uk/...afety-science-ai-seoul-summit-2024-annex
12. ^UK Government, "Frontier AI Safety Commitments, AI Seoul Summit 2024," updated February 7, 2025. gov.uk/...-safety-commitments-ai-seoul-summit-2024
13. ^UK Department for Science, Innovation and Technology, "Historic first as companies spanning North America, Asia, Europe and Middle East agree safety commitments on development of AI," May 21, 2024. gov.uk/...-safety-commitments-on-development-of-ai
14. ^UK Government, "Seoul Ministerial Statement for advancing AI safety, innovation and inclusivity," May 22, 2024. gov.uk/...ion-and-inclusivity-ai-seoul-summit-2024
15. ^UK Department for Science, Innovation and Technology, "New commitment to deepen work on severe AI risks concludes AI Seoul Summit," May 22, 2024. gov.uk/...evere-ai-risks-concludes-ai-seoul-summit
16. ^UK Department for Science, Innovation and Technology and AI Safety Institute, "International scientific report on the safety of advanced AI: interim report," May 17, 2024. gov.uk/...ific-report-on-the-safety-of-advanced-ai
17. ^UK Department for Science, Innovation and Technology, "Tech Secretary unveils £8.5 million research funding set to break new grounds in AI safety testing," May 22, 2024. gov.uk/...o-break-new-grounds-in-ai-safety-testing
18. ^UK Department for Science, Innovation and Technology, "Government's trailblazing Institute for AI Safety to open doors in San Francisco," May 20, 2024. gov.uk/...ai-safety-to-open-doors-in-san-francisco
19. ^Matt O'Brien, "AI companies make fresh safety promise at Seoul summit, nations agree to align work on risks," Associated Press, May 21, 2024. apnews.com/...2cc2b297872d860edc60545d5a5cf598
20. ^Gregory C. Allen and Georgia Adamson, "The AI Seoul Summit," Center for Strategic and International Studies, May 23, 2024. csis.org/...ai-seoul-summit
21. ^Sebastian Klovig Skelton, "AI Seoul Summit review," Computer Weekly, June 3, 2024. computerweekly.com/...AI-Seoul-Summit-review
22. ^Alex Krasodomski, Olivia O'Sullivan, and Rowan Wilkinson, "The Seoul AI summit should put local capacity at the heart of AI for international development," Chatham House, May 22, 2024. chathamhouse.org/...t-ai-international-development
23. ^US National Institute of Standards and Technology, "U.S. Department of Commerce and U.S. Department of State Launch the International Network of AI Safety Institutes at Inaugural Convening in San Francisco," November 20, 2024. nist.gov/...-department-state-launch-international
24. ^European Commission, "First meeting of the International Network of AI Safety Institutes," November 20, 2024. digital-strategy.ec.europa.eu/...safety-institutes
25. ^Republic of Korea Ministry of Science and ICT, "'AI Safety Institute' Launched Following the AI Seoul Summit in May," November 27, 2024. msit.go.kr/...view.do
26. ^UK AI Security Institute, "International consensus and open questions in AI evaluations," accessed July 28, 2026. aisi.gov.uk/...etwork-consensus-and-open-questions
27. ^Lily Stelling, Malcolm Murray, Simeon Campos, and Henry Papadatos, "Evaluating AI Companies' Frontier Safety Frameworks: Methodology and Results," arXiv:2512.01166, December 2025. arxiv.org/...2512.01166
28. ^Presidency of the French Republic, "Presentation: Artificial Intelligence Action Summit," updated January 17, 2025. elysee.fr/...presentation
29. ^Government of India Press Information Bureau, "India AI Impact Summit 2026 Witnessed Extensive Participation of Public," March 11, 2026, and "AI Impact Summit 2026 Concludes with Adoption of New Delhi Declaration," February 21, 2026. pib.gov.in/PressReleasePage ; pib.gov.in/PressReleasePage