Coordination between states, multilateral institutions, safety institutes, and frontier laboratories on the governance of artificial intelligence, spanning safety summit declarations, shared model evaluation programmes, harmonised risk thresholds for frontier models, compute and export control alignment, and scientific exchange, aimed at managing risks that no single jurisdiction can address alone.
Semantic Classification
Content
Definition
International AI cooperation is the AI-specific specialisation of international cooperation: the web of summits, declarations, institutions, and working-level programmes through which governments attempt to govern a technology whose development, deployment, and risks cross borders by default. Frontier models are trained on globally sourced data, served worldwide within days of release, and — if they pose catastrophic risks — pose them everywhere at once. No national regulator can evaluate, contain, or set thresholds for such systems alone, which is why frontier-model governance structurally bridges into international coordination.
The modern architecture of this cooperation formed quickly. The UK’s Bletchley Park AI Safety Summit (November 2023) produced the first joint declaration on frontier AI risk signed by 28 countries including both the United States and China; follow-on summits in Seoul (2024) and Paris (2025) extracted voluntary frontier safety commitments from leading laboratories and broadened the agenda towards adoption and resilience, and the India AI Impact Summit in New Delhi (February 2026) completed the shift from a risk-centred to an impact-centred agenda. In parallel, national AI safety/security institutes — led by the UK AI Safety Institute and its US counterpart — formed an International Network of AI Safety Institutes to share evaluation methods, run joint model testing, and align on risk thresholds, giving the summit diplomacy a standing technical machinery.
Beyond the summit track, cooperation runs through the OECD AI Principles, the G7 Hiroshima Process code of conduct, the Council of Europe’s AI framework convention (the first binding international AI treaty), UN initiatives including the Global Digital Compact and an independent scientific panel on AI, and the International AI Safety Report chaired by Yoshua Bengio — a deliberate echo of the IPCC model for building shared scientific baselines.
Current Landscape
-
India AI Impact Summit (New Delhi, 16–21 February 2026): the fourth summit in the Bletchley lineage concluded with the New Delhi Declaration, endorsed by roughly 90 countries and international organisations (official tallies range from 88 to 92), including the United States, China, the UK, and the EU — far broader than Bletchley’s 29, Seoul’s 11, or Paris’s 60 signatures.
-
From risk to impact: New Delhi deliberately reframed the summit agenda from frontier “risk” towards development “impact”, structuring the Declaration around seven pillars including democratising AI resources, secure and trusted AI, AI for science, and resilient AI systems.
-
Frontier laboratory commitments (February 2026): thirteen frontier model developers signed the New Delhi Frontier AI Impact Commitments, pledging published statistical insights on real-world AI usage and stronger multilingual and low-resource-language evaluations by the next summit.
-
New standing machinery: the summit launched the Global AI Impact Commons, announced a Trusted AI Commons with 22 partner countries, created a Network of AI for Science Institutions with 19 founding countries, and released Guidance Notes on AI Governance endorsed by 22 countries.
-
Money follows diplomacy: infrastructure-related investment pledges at the summit exceeded USD 250 billion, with roughly USD 20 billion more in venture and deep-tech commitments — a reminder that cooperation and compute competition are now inseparable.
The central tension is between cooperation and strategic competition. The same states convening safety summits are simultaneously racing for compute, talent, and military AI advantage, and export controls on advanced chips sit awkwardly beside calls for inclusive global governance. Practical near-term cooperation has therefore concentrated where interests align most cleanly: shared evaluation science, incident reporting, and thresholds for the most severe misuse risks (biosecurity, cyber offence), rather than harmonised regulation of everyday AI.
Open problems include bringing China durably into technical safety exchange, giving the Global South genuine standing rather than observer status, making laboratory commitments verifiable rather than voluntary, and deciding whether the summit-and-institute network should harden into treaty-based institutions. The trajectory of compute governance — the most enforceable lever currently available — will substantially determine whether international AI cooperation matures into a regime or remains a communiqué culture.
Sources:
-
https://www.pib.gov.in/PressReleasePage.aspx?PRID=2231208®=48&lang=2
-
https://www.techuk.org/resource/key-outcomes-from-the-2026-ai-impact-summit-with-techuk.html