Examine the institutional architecture of global AI cooperation, with reference to GPAI-OECD integration and BRICS-led alternatives.

Global AI cooperation rests not on a treaty but on plural, voluntary platforms — an OECD-anchored expert track, a universal UN layer, and now bloc-based alternatives. Examining them shows that fragmentation, rather than absence of rules, is the core governance problem.

The GPAI–OECD track: the standard-setting core

• GPAI was launched in 2020 with 15 founding members, including India, to bridge AI theory and practice [1][2].
• Its Secretariat functions merged with the OECD in July 2024, announced at the New Delhi Summit under India's Council chairship; the integrated partnership spans 46 countries and the EU [1][2].
• It works through four expert themes — Responsible AI, Data Governance, Future of Work, Innovation and Commercialisation — all recommendatory [1].
• Limitation: club-sized membership, and no pillar addressing model or compute access, the Global South's actual grievance.

The UN layer: universality without enforcement

• The UNESCO Recommendation on the Ethics of AI (2021), adopted by 193 countries, remains the first global ethical standard [4].
• The Global Digital Compact (2024) created a 40-expert Independent International Scientific Panel on AI and a Global Dialogue on AI Governance co-facilitated by Costa Rica and Spain [3].
• Limitation: the Panel's mandate covers only the non-military domain, leaving "weaponisation" of AI institutionally unaddressed [3].

BRICS-led alternatives

• At the 18th BRICS Summit, New Delhi (12–13 September 2026) under India's chairship [5], China proposed a "BRICS AI open source community" to dilute dominance of any single country's models.
• Merit: open weights cannot be recalled, insuring users against arbitrary vendor withdrawal.
• Weakness: open weights do not solve compute. India's own subsidised pool crossed 34,000 GPUs, largely rented from empanelled private providers [6] — a chokepoint export controls still command. Bloc-branding also invites mirror restrictions.

The architecture is thus layered but uniformly voluntary, and India alone sits credibly in all three. The way forward is to multilateralise the India AI Governance Guidelines [7] through the AI Impact Summit and anchor them in the UN Dialogue, converting India's convening power into rules that are inclusive, human-centric and durable.