AI cooperation

1. At a Glance

• AI cooperation refers to multilateral/bilateral frameworks governing safe, inclusive development and deployment of Artificial Intelligence, avoiding its "weaponisation" amid US-China rivalry.
• India hosted the 18th BRICS Summit in New Delhi (12–13 September 2026) and the India AI Impact Summit 2026, positioning itself as a bridge between US-led AI models and a BRICS "open source" alternative [1][3].
• India is a founding member (2020) of the Global Partnership on Artificial Intelligence (GPAI), which merged its Secretariat function with the OECD in July 2024 [4][5].
• Relevant for GS-II (International Relations/Groupings) and GS-III (Science & Tech, Cyber/AI security).

2. Why in the News

• At the 18th BRICS Summit (New Delhi, Sept 2026), PM Modi warned against "weaponisation" of technology and critical minerals, while China's President Xi Jinping proposed a "BRICS AI open source community" to counter dominance of any single country's AI models [6].
• The Hindu Businessline op-ed (16 Sept 2026) frames this as AI development standing at a "crossroads" — the US government has disregarded frontier AI developers' own call for a global slowdown over hacking/misalignment risks, while the BRICS bloc offers an alternative multilateral posture [6].
• The op-ed cites the Global South being caught in great-power rivalry, exemplified by a pull-out of Anthropic's models ("Fable" and "Mythos") from certain markets [6].
• India hosted a Bharat Innovates Exposition (11–12 September 2026, Bharat Mandapam) showcasing Deep-Tech/AI innovation ahead of the BRICS Summit [1].

3. Background & Evolution

• 2019 — OECD adopts Recommendation on Artificial Intelligence, the first intergovernmental AI policy standard.
• 2020 — GPAI launched with 15 founding members, including India, to bridge AI theory and practice responsibly [4][5].
• December 2023 — India hosted the GPAI Summit in New Delhi (12–14 Dec 2023); the "New Delhi Declaration" adopted [5].
• July 2024 — GPAI and OECD formally merge functions to create an integrated international AI partnership under the GPAI brand, with a renewed vision announced at a New Delhi summit (3–4 July 2024) under India's Council chairmanship [4][5].
• September 2026 — 18th BRICS Summit (New Delhi) and India AI Impact Summit 2026 mark India's dual engagement with Western-anchored (GPAI/OECD) and BRICS-led AI cooperation tracks [1][3].

4. Core Static Facts

5. Multi-Dimensional Analysis

Geopolitical/Strategic

• India navigates a bipolar AI landscape (US-led proprietary frontier models vs China-backed BRICS open-source push), avoiding formal alignment with either bloc [6].
• Framing of AI/mineral "weaponisation" links AI cooperation to broader strategic-resource geopolitics (export controls, critical minerals) [6].

Economic

• Massive capital expenditure by US AI firms in data centres is framed domestically as necessary for US-China technological competition [6].
• India's relatively "fledgling" AI ecosystem shapes its cautious, participative-rather-than-confrontational stance, seeking dividends from being present in the AI value chain [6].

Scientific/Technological

• Open-source AI models (BRICS proposal) vs closed frontier models (US firms like Anthropic) represent competing technological governance philosophies.
• Frontier AI developers themselves have called for a global slowdown in model development citing hacking and misalignment risks — a rare industry self-regulation signal not heeded by the US government [6].

Ethical/Governance

• GPAI/OECD framework emphasises "human-centric, safe, secure and trustworthy AI" as a normative standard [4].
• Global South vulnerability is highlighted by supply-side disruptions (e.g., withdrawal of Anthropic models from certain markets), raising equity concerns in access to frontier AI [6].

Administrative

• India simultaneously runs two institutional tracks: (a) GPAI/OECD-anchored multilateral AI governance since 2020, and (b) BRICS-anchored AI cooperation emerging in 2026 — requiring coordination across MeitY, MEA, and host-ministry logistics (Bharat Innovates Exposition run by Ministry of Education) [1][4].

6. Recent Developments (last 12–18 months)

• 3–4 July 2024: Renewed GPAI vision announced under India's Council chairmanship, merging with OECD [4][5].
• 11–12 September 2026: Bharat Innovates Exposition at Bharat Mandapam, New Delhi, showcasing Deep-Tech ahead of BRICS Summit [1].
• 12–13 September 2026: 18th BRICS Summit hosted by India in New Delhi; AI cooperation and tech "weaponisation" discussed [1][6].
• September 2026: India AI Impact Summit 2026 convened, with OECD.AI Observatory participation [3].
• September 2026: Xi Jinping proposes "BRICS AI open source community"; PM Modi cautions against weaponisation of tech/minerals at the Summit [6].

7. Prelims Hooks

• GPAI (Global Partnership on Artificial Intelligence) was launched in 2020 with 15 founding members, including India [4].
• India hosted the GPAI Summit in New Delhi, 12–14 December 2023, adopting the "New Delhi Declaration" [5].
• GPAI merged its functions with the OECD in July 2024 [4].
• GPAI currently comprises 46 member countries + the European Union [4].
• GPAI's four working-group themes: Responsible AI, Data Governance, Future of Work, Innovation & Commercialization [4].
• India hosted the 18th BRICS Summit in New Delhi on 12–13 September 2026 [1].
• The "Bharat Innovates Exposition" was organised by the Union Ministry of Education on 11–12 September 2026 at Bharat Mandapam [1].
• At the 2026 BRICS Summit, Chinese President Xi Jinping proposed a "BRICS AI open source community" [6].
• PM Narendra Modi warned against the "weaponisation" of technology and minerals at the same Summit [6].
• The India AI Impact Summit 2026 featured participation from the OECD.AI Observatory [3].
• India's stance on AI blocs is described as a "middle path" — not opposing US-led models, but resisting AI's capture by national rivalries [6].
• Anthropic's models, referenced in the article as "Fable" and "Mythos," were cited as examples of Global South access disruption via market pull-outs [6].

8. The UN Track the Bipolar Framing Leaves Out

• Membership arithmetic — the note's two tracks are both clubs: GPAI is 46 countries + the EU [4], BRICS is a bloc. UNESCO's Recommendation on the Ethics of AI applies to all 194 member states [10], and the Global Digital Compact was adopted by 193 states in September 2024 alongside the Pact for the Future [9] — the forum where the Global South actually has numbers is neither GPAI nor BRICS.
• An evidence body already exists, unused by the Summit rhetoric — the Compact created an Independent International Scientific Panel on AI of 40 experts, mandated to produce an annual evidence-based report and update the General Assembly up to twice a year [9]. Assertions about "weaponisation" [6] can be routed to a standing scientific assessment rather than a summit communiqué.
• The Panel's remit stops at the military line — its mandate expressly covers the non-military domain [9]. So the sharpest form of the weaponisation concern — autonomous targeting, AI in conflict — falls outside the only global scientific body dedicated to AI, and no BRICS or GPAI instrument fills it either.
• Facilitation sits with middle powers, not blocs — the Global Dialogue on AI Governance is co-facilitated by Costa Rica and Spain [9]. India's "middle path" [6] has a ready institutional home there; a third bloc-branded track adds a forum, not a rule.
• All of it is recommendatory — OECD 2019, UNESCO 2021 [10] and the GDC 2024 [9] create standards with no inspection, reporting or enforcement arm. A "BRICS AI open source community" [6] would be a fourth voluntary layer, which is why forum proliferation is itself the governance problem.

9. Why Open Weights Do Not Buy Autonomy Without Compute

• India's compute floor, stated precisely — the IndiaAI Mission's common compute pool reached 34,333 GPUs: 18,417 empanelled, plus 15,916 added with industry partners [8]. That is a subsidised shared facility for startups, MSMEs, academia and government users — sized for fine-tuning and inference, not frontier pre-training runs.
• Rented, not owned — the pool was assembled by empanelling cloud providers through an RFE issued 16 August 2024, drawing 19 bidders of which 10 reached commercial bidding [8]. The state buys subsidised access to privately held accelerators; an export-control or supply shock hits exactly the chokepoint PM Modi's "weaponisation" warning names [6].
• The BRICS proposal addresses the wrong bottleneck — open weights solve model access; training and serving them still consumes the same accelerators under the same export-control regime. A free model on unavailable hardware is not sovereignty [6][8].
• What India actually holds — adoption scale, data, and the convening chair (India AI Impact Summit 2026, with OECD.AI Observatory present [3]). A negotiating strategy built on compute parity fails on the numbers; one built on convening and market size does not.

10. The Strongest Case for a BRICS Open-Source Stack, and Where It Breaks

• Concede the grievance — the op-ed's example of frontier models being pulled from certain markets shows a Global South user can lose a deployed model by a vendor's commercial or legal decision, with no notice obligation [6]. Open weights, once released, cannot be recalled. That is a real resilience argument, not posturing.
• Concede the institutional gap — GPAI's four themes are Responsible AI, Data Governance, Future of Work, Innovation & Commercialization [4]. None is a model-access or compute-access pillar. The Western-anchored track has no answer to the withdrawal problem, which is precisely the opening Xi's proposal exploits [6].
• But open weights are not open governance — architecture, training corpus, tokenizer and safety filters are fixed upstream by whoever funds the training run. Recipients get free weights, not a vote on what the model encodes; dependence is relocated, not removed.
• And irrevocable release removes the only live safety lever — frontier developers' own call for a slowdown cited hacking and misalignment risks [6], and every existing framework — UNESCO 2021 [10], the GDC [9] — is recommendatory, relying on vendors' ability to restrict, rate-limit or withdraw. A stack built on permanent public release has no recall mechanism at all.
• Bloc-branding invites the mirror sanction — a stack identified with one bloc becomes a legitimate target for the other's export controls, converting a hedge into a second single point of failure.

11. Converting the Host Chair Into a Negotiated Deliverable

• MeitY: table the India AI Governance Guidelines as summit text — MeitY released them under the IndiaAI Mission as a framework for safe, inclusive and responsible AI adoption, explicitly framed as a reference to foster national and international cooperation [7]. The Impact Summit 2026 [3] is where that becomes a negotiating position instead of a domestic document.
• MEA: bank India's ask in the UN track, not a new bloc — support the Costa Rica/Spain-facilitated Global Dialogue and seek Indian representation on the 40-member Scientific Panel [9]. Precedent exists: India chaired the GPAI Council through the 2024 OECD integration [4][5], so it has held a governing seat in a technical AI body before.
• The one concrete Global South ask: a continuity obligation — a notice-period and data-portability commitment on frontier vendors before market withdrawal, which is the exact failure the pull-out example exposes [6] and which none of GPAI's four themes covers [4]. It is enforceable through procurement contracts even while treaties stay voluntary.
• Offer compute, not a communiqué — open a tranche of the 34,333-GPU empanelled pool [8] to partner-country researchers as a shared facility. It is a deliverable neither US proprietary licensing nor a BRICS weight-drop provides, and it makes India a supplier in the value chain rather than a venue for it.
• Sequence the norm before the bloc — India's Guidelines [7] plus the New Delhi Declaration [5] give it text to defend; endorsing an open-source community before those norms are multilateralised concedes the framing to the sponsor of the proposal [6].

12. Anchors for Answers

• Data: 34,333 GPUs in IndiaAI's common compute pool — 18,417 empanelled + 15,916 added with industry partners [8]
• Data: GPAI covers 46 countries + EU [4]; UNESCO's AI ethics standard covers all 194 member states [10]; the Global Digital Compact was adopted by 193 states [9]
• Report/Committee: India AI Governance Guidelines, MeitY/IndiaAI Mission, 2025 [7]; Independent International Scientific Panel on AI — 40 experts, non-military remit, annual report to the UNGA, 2024 [9]
• Law/Case: UNESCO Recommendation on the Ethics of Artificial Intelligence, 2021 — first global standard on AI ethics, recommendatory [10]; OECD Recommendation on AI, 2019; New Delhi Declaration, GPAI Summit 2023 [5]
• Comparison: the UN Global Dialogue on AI Governance is co-facilitated by Costa Rica and Spain — two middle powers, not bloc leaders — the template for India's "middle path" [9][6]
• Scheme: IndiaAI Mission — RFE of 16 August 2024, 19 bidders, 10 shortlisted, subsidised access to privately owned accelerators [8]

13. Mains Relevance

• GS-II (International Relations): India's role in multilateral groupings — BRICS, GPAI/OECD; India's foreign policy balancing act between competing power blocs.
• GS-III (Science & Technology): Awareness in fields of IT, AI, robotics; issues of intellectual property, AI governance, cybersecurity.
• Possible Mains question stems: 1. Discuss the significance of India's 'middle path' approach to global AI governance amid US-China technological rivalry. (GS-II, 15 marks) 2. Examine the institutional architecture of global AI cooperation, with reference to GPAI-OECD integration and BRICS-led alternatives. (GS-III, 15 marks) 3. AI development risks becoming a theatre for great-power rivalry. Critically analyse the implications for the Global South. (GS-II/GS-III, 10 marks)

14. Related Topics to Study Next

• GPAI (Global Partnership on AI) — India's founding-member role and 2023 New Delhi Declaration directly underpin this topic.
• BRICS grouping and NDB — institutional vehicle through which the "BRICS AI open source community" proposal was floated.
• India's National AI Strategy / IndiaAI Mission — domestic policy backdrop to India's international AI positioning.
• Export controls & critical minerals geopolitics — linked by PM Modi's "weaponisation" remarks.
• Data localization and digital sovereignty debates — underlying theme in AI governance disputes.
• UNESCO Recommendation on the Ethics of AI (2021) — global normative framework parallel to GPAI/OECD.
• US-China tech decoupling / semiconductor export controls — the rivalry backdrop driving AI cooperation debates.
• Global South digital divide — context for Anthropic model withdrawal example.

15. Common Errors / Trap Areas

• Do not confuse GPAI (Global Partnership on AI, founded 2020, merged with OECD in 2024) with GPAI's parent — GPAI is NOT an OECD creation; it merged functions with OECD later.
• Do not attribute the "BRICS AI open source community" proposal to India — it was proposed by China's Xi Jinping, not PM Modi.
• Distinguish India's stance: India has not joined any protest against US-led AI models, despite hosting BRICS — avoid assuming India has taken an anti-US position.
• The India AI Impact Summit 2026 and the 18th BRICS Summit (2026) are distinct events, though both held in India around the same period — do not conflate their agendas.
• Implementing ministry confusion: Bharat Innovates Exposition was organised by the Ministry of Education, not MeitY — verify ministry attributions carefully.