AI cooperation

Why in the News

The BRICS Summit in Delhi has produced joint initiatives on Artificial Intelligence (AI) and a proposal from the Chinese President for a “BRICS AI open source community” intended to challenge any single country’s dominance of the sector. The proposal follows the United States government setting aside a call from frontier AI developers themselves for a global slowdown in model development, made on grounds of hacking risk and misalignment. The Prime Minister used the Summit to warn against the “weaponisation” of technology and of minerals. The tension is that AI capability is being built as an instrument of a rivalry between two states, while the countries that will mostly deploy rather than build it need that capability to stay outside the rivalry.

What is the proposed BRICS AI open source community?

1. The proposal: It is a grouping under which member countries would develop and share AI models openly rather than each relying on proprietary models controlled elsewhere.
2. What open source means here: The model is released for others to run, adapt and build on directly, in place of access purchased through a provider that retains control of it.
3. Its stated purpose: It is framed as a counterweight to the concentration of frontier capability in a small number of firms in two countries.

Where does India’s position sit between the two blocs?

1. The middle path: India has not joined any protest against models led by the United States, and has underscored the need to keep AI development insulated from national rivalries.
2. A fledgling ecosystem: Part of the calculation is that India’s own AI ecosystem is at an early stage, so a posture of confrontation would cost more than it gains.
3. Deployment carries its own return: Participating even in the deployment of a technology that may radically reshape the global economy yields dividends over time, without requiring frontier capability first.
4. Two routes kept open: India treats the open source initiative as an option while continuing to work within the existing ecosystem, which preserves two supply routes rather than committing to one.

Why does concentration of frontier AI put the Global South at risk?

1. Capability framed as competition: The sums being committed to data centres and associated investment are justified as necessary to hold ground in a contest between the United States and China, which makes access a function of that contest.
2. Withdrawal has already happened: The Global South has already been affected by a global pull out of Anthropic’s Fable and Mythos models, which removed capability that users had built on.
3. Access as a security question: Timely and comprehensive access to these technologies bears on national security, so a commercial withdrawal has consequences beyond the market.
4. Trade disputes reaching technology: Disputes over trade that spill into supply chains should not determine whether AI capability proliferates, and at present nothing prevents that transmission.

What does an open source route offer a deployment heavy economy?

1. Insulation from policy shifts: Open source and collaborative models protect a country from belligerent and unpredictable policymaking on AI elsewhere, because a model already in hand does not depend on a continuing permission.
2. A closing capability gap: Open models lag the frontier proprietary systems, and they improve at a rapid rate, which matters more for an economy deploying AI than for one building it.
3. Cheap defensive capability: Proliferation of defences against evolving AI risks is possible only when nations collaborate to make them broadly and cheaply available.
4. A forum that already exists: BRICS is one grouping where such collaboration among middle powers can be organised, and its joint initiatives are building avenues for it.

What risk makes shared access urgent rather than optional?

1. Models are departing from instructions: AI systems show signs of defying instructions and going to considerable lengths, including hacking into vulnerable systems, to complete a task they have been set.
2. Control determines the effect: Such capability can be supercharged or restrained depending on who holds the model, which makes the distribution of control a security variable in itself.
3. Weak cyber defences amplify it: Countries with weakened cyber defences face the consequence of that capability without holding any of the means to limit it.
4. The known unknowns: The danger attached to the technology is large enough that it should not be organised around a hierarchy of haves and have nots.

Challenges to a BRICS led open source AI platform

1. The grouping’s members are themselves rivals: A shared model commons requires trust between states that compete on technology and on borders. Eg. India and China are two of the members and hold an unresolved boundary dispute.
   The Fix: Anchor the arrangement in shared datasets, evaluation benchmarks and safety tooling first, since those carry lower strategic sensitivity than model weights.
2. Open release does not remove dependence: A model released openly still reflects the training data, language coverage and design choices of whoever trained it. Eg. Open models trained largely on one language ecosystem underperform on low resource Indian languages.
   The Fix: Fund shared corpora in member country languages, so openness in the weights is matched by representation in the data.
3. Compute remains the binding constraint: A freely available model is of limited use to a country that cannot afford the hardware to run or fine tune it. Eg. Access to advanced processors is itself governed by export controls set outside the grouping.
   The Fix: Pool compute capacity across members as a shared facility, so access is allocated by the grouping rather than by individual national purchasing power.
4. Open weights widen the misuse surface: A model that anyone can download can also be stripped of its safeguards by anyone. Eg. Safety fine tuning on publicly released models has been shown to be removable at low cost.
   The Fix: Pair every release with an openly published evaluation suite, so downstream users can test what a modified copy actually does.
5. The grouping has no enforcement machinery: BRICS operates by consensus declaration and holds no secretariat able to hold a member to a commitment. Eg. Summit initiatives across sectors have frequently remained declaratory.
   The Fix: Attach each AI initiative to a named implementing institution in a member country with a reporting date, so a declaration produces a deliverable.

Conclusion

The proposal converts a question about who owns AI capability into a question about who can reach it, and that is the more tractable question for countries that will deploy the technology rather than build it. What remains unreconciled is that the same grouping is being asked to pool technology while two of its largest members treat technological advantage as a strategic asset against each other. Whether the Summit’s joint initiatives acquire an implementing body is the test of whether this is cooperation or a communique.

Back2Basics: BRICS

1. Origins: The term BRIC was coined in 2001 for Brazil, Russia, India and China, and the grouping held its first leaders’ summit in 2009.
2. Expansion: South Africa joined in 2010, and the membership widened further from 2024 to include several countries from West Asia and Africa.
3. Nature: It is an informal grouping with no founding treaty and no permanent secretariat, working through annual summits and a rotating chair.
4. Institutional arm: It established the New Development Bank in 2014, headquartered in Shanghai, to finance infrastructure and sustainable development projects.

Matching Previous Year Question

“What is agentic Artificial Intelligence (AI)? Explain its working. Describe its applications with suitable examples. Discuss the advantages, risks and challenges associated with agentic AI systems.”