International AI Cooperation
International AI cooperation is the set of treaties, declarations, summits and standards bodies through which governments try to coordinate how advanced artificial intelligence is developed and governed across borders, in a field where models, chips, talent and risks cross national boundaries far faster than any single jurisdiction's law can. It is not a single institution: it is a patchwork of overlapping tracks — safety-focused summits, G7 and OECD policy principles, United Nations resolutions and a first binding treaty — that differ sharply in who participates, how binding they are and what they actually require.
The summit series began at Bletchley Park in November 2023, where 28 countries and the European Union, including both the United States and China, signed the Bletchley Declaration. The declaration recognized that frontier AI poses risks that warrant shared attention, singled out areas such as cybersecurity and biotechnology, and committed signatories to cooperate on scientific understanding of those risks. It was a political statement, not a treaty, creating no enforcement mechanism. The follow-up AI Seoul Summit in May 2024, co-hosted by South Korea and the United Kingdom, produced the Seoul Declaration, a commitment to build a network of safety institutes (see AI safety institute), and Frontier AI Safety Commitments in which sixteen companies agreed to publish safety frameworks. The AI Action Summit in Paris on 10–11 February 2025 widened the agenda toward innovation, public-interest AI and energy; its concluding statement on inclusive and sustainable AI was endorsed by roughly sixty countries including China and India, while the United States and United Kingdom did not sign it, which showed how thin the consensus had become on how much emphasis safety should receive.
The G7 Hiroshima AI Process and the OECD supply much of the shared vocabulary. The OECD adopted its AI Principles in 2019 — the first intergovernmental standard on AI, covering values such as transparency, robustness and accountability — and updated them in 2024 to address general-purpose and generative systems. The G7, at its May 2023 summit in Hiroshima, launched a process that led in October 2023 to International Guiding Principles and a voluntary Code of Conduct for organizations developing advanced AI systems, which later fed into a reporting framework developed with the OECD. These instruments are soft law: they shape corporate and national practice mainly through reputation, procurement and later legislation such as the European Union's AI Act, which is regional rather than international.
The United Nations track aims for universality at the cost of specificity. In March 2024 the General Assembly adopted by consensus a United States–led resolution promoting safe, secure and trustworthy AI for sustainable development, co-sponsored by well over a hundred member states, and later resolutions addressed capacity building for developing countries. The UN also convened an advisory body whose reports argued for a more inclusive global governance conversation. These texts are non-binding, but they matter because they bring in many countries absent from the summit and G7 circles.
The Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law is the first legally binding international treaty on AI. It was adopted in May 2024 and opened for signature in September 2024, with signatories including the European Union, the United Kingdom and the United States, among others; as with any treaty, its force depends on ratification and entry into force, and its obligations are framed around human rights, democratic processes and accountability for AI use rather than around frontier-model capabilities. A complementary scientific effort, the International AI Safety Report chaired by Yoshua Bengio, published an interim report in 2024 and a first full report in January 2025, aiming to play a role loosely analogous to the climate panel's assessments.
Two pitfalls recur when reading this landscape. The first is to mistake declarations for governance: most summit outputs are voluntary and non-enforceable, so their significance lies in agenda-setting, shared evaluation practice and norm-building rather than in compliance. The second is to treat the field as a unified bloc, when participants differ in priorities — safety and catastrophic risk, innovation and competitiveness, human rights, development and digital sovereignty — and when export controls and strategic competition between major powers run alongside, and sometimes against, cooperative diplomacy.
Read international AI cooperation as a layered and uneven attempt to build shared rules and shared measurement for a technology no one country controls: broad but soft where participation is wide, binding only where few commit, and always shaped by the competing interests of the states at the table.