Subject: Science and Technology

Mentor’s Comment

The United States government directed Anthropic to suspend foreign national access to its Fable 5 and Mythos 5 AI models on national security grounds, and is separately considering equity stakes in leading AI companies. At the same time, India lacks frontier AI capability of its own and must rely on foreign models to remain competitive. This dependence carries geopolitical risk that neither market competition nor inter-ministerial coordination alone can resolve.

What explains the global turn toward sovereign AI policymaking, and why does India need a coordinated response?

1. US export controls: The US suspended foreign national access to Anthropic’s Fable 5 and Mythos 5 models on national security grounds and created a voluntary mechanism for federal government access up to 30 days before trusted partners.
2. Equity stake consideration: The US administration is considering taking equity stakes in leading AI firms to capture a share of the supernormal profits expected from the technology.
3. Global pattern: Governments are increasingly shaping AI policy around national advantage rather than leaving diffusion purely to markets.
4. India’s structural gap: India is a large IT services economy without its own frontier AI systems (Frontier AI: AI systems requiring upwards of ten septillion floating-point operations to train).
5. Reason for urgency: Policy decisions made elsewhere increasingly determine the terms on which India can access frontier technology, making a coherent domestic response necessary now.

Why is India’s AI policy discourse trapped in a false binary, and why must this framing be rejected?

1. The dependence dilemma: India’s IT and app companies must use the best available foreign AI to remain competitive, yet this use deepens dependence on models built abroad.
2. Sequencing logic: Using foreign AI today builds the economic surplus needed to depend on it less in future. Diffusion and dependence-reduction are sequential goals, not opposed ones.
3. Limits of firm-level action: Firms can outcompete rivals using foreign AI. Firms cannot manage the geopolitical risks that accompany dependence on it. That risk-management role falls to public policy.
4. False binary named: India’s discourse frames globalisation and industrial policy as mutually exclusive. Indian industry must benefit from both at the same time.
5. Pharma precedent: Indian pharmaceutical manufacturing shows the limits of industrial policy alone. A Production-Linked Incentive (PLI: a government scheme offering incentives tied to incremental domestic manufacturing output) promoted domestic bulk drug production. India still sources 65% of critical ingredients from China, per NITI Aayog’s latest assessment.
6. Implication: Industrial policy creates footholds. It does not create instant resilience. This sets the correct expectation for AI policy as well.

What institutional architecture should India build to benefit from frontier AI without deepening strategic dependence?

1. Scale of the gap: India spends 0.6% of GDP on research and development, of which the private sector accounts for a third. OpenAI alone projects $50 billion in compute spending this year, over six times India’s annual private R&D spend.
2. Strategic implication: India cannot outspend frontier AI investment. India must instead deepen backward linkages to frontier AI while strengthening forward linkages for its own products and services.
3. Whole-of-government approach: Ministries of external affairs, commerce, and information technology must coordinate closely. Coordination should extend to defence, energy, and telecom where relevant.
4. Objective of coordination: The architecture secures continued access to frontier AI inputs. It simultaneously builds global market access for Indian AI-enabled products and services.

Since coordination alone cannot manage geopolitical risk, what role must the state play in underwriting it?

1. Limits of firm-level risk management: Firms can manage commercial risk through contracts and diversified supply chains. Firms cannot insure themselves against geopolitical risk or concentrated technological dependence.
2. Sovereign risk-bearing role: Underwriting such risk is a function only the state can perform. Private capital cannot efficiently bear this risk alone.
3. Export credit analogy: Export credit mechanisms insure firms against risks they cannot shoulder independently in international trade, offering a template for AI-related risk underwriting.
4. Hybrid-annuity analogy: The Hybrid-Annuity Model (HAM: an infrastructure financing structure where the state funds part of a project and makes fixed payments over time) reduces the share of risk borne by private capital in long-gestation infrastructure. A comparable approach could apply to frontier AI dependence.

What do the available global examples suggest about alternative sovereign AI strategies? 

1. Europe: Shifted from a “regulate first, ask questions later” approach to investing directly in AI compute capacity and promoting “Buy European” public procurement to support its domestic AI industry.
2. Argentina: Is positioning itself to attract AI investment by offering a regulatory safe harbour under an accommodative regulatory posture.

Why must India’s technology industry itself close the competitiveness gap, and what does this reveal about the limits of policy alone?

1. Government’s limits: Government action can create conditions for success. Competitiveness must ultimately come from firms themselves.
2. Export benchmark: The Philippines generates $40 billion in IT exports, nearly a sixth of India’s IT exports, and is growing faster than the global industry.
3. App market underperformance: No Indian app features among the top 10 globally by downloads, in-app purchase revenue, or monthly active users.
4. Fragmented industry voice: Incumbent IT firms remain focused on visas and market access. Startups remain consumed by regulatory friction and fundraising. Both share a common interest in India’s continued connection to global AI ecosystems alongside growing domestic capability.
5. Core stakes: The central contest in AI is not only over who builds the best models. It is over who captures the economic and strategic advantages the models create.

Conclusion

India’s AI strategy must reject the false choice between global integration and domestic capability building. The objective is to remain deeply integrated with global AI ecosystems while steadily reducing the strategic vulnerabilities such integration creates. This requires backward linkages secured through whole-of-government coordination, forward linkages built through competitive Indian products and services, and state-backed risk underwriting on the export-credit and hybrid-annuity model. Without matching ambition from industry itself, government action alone cannot close the gap.

Mentor’s Comment

AI debates have centred on job losses and concentration of power among a few firms and nations. A third, less discussed risk is emerging: AI is being treated as a substitute for human cognition, even though it produces information, not knowledge. The conflation of AI output with genuine knowledge has no such precedent and currently has no accountability structure attached to it.

Why are labour displacement and power concentration considered the more manageable AI risks?

1. Historical precedent on labour: Technology has automated specific tasks, not entire professions; the steam engine displaced labour into new industries rather than eliminating it.
2. Expected AI trajectory: Some occupations will shrink, others will expand, and new professions will emerge, mirroring past transitions.
3. Transition cost is real: The shift will require substantial investment in reskilling, but is not existential.
4. Capital-intensive economics of AI: Frontier models require massive investment in computing infrastructure, energy, talent and data, restricting ownership to a few firms and countries.
5. Concentration risk has known parallels: Concentrated control of strategic resources such as gold or oil has historically produced geopolitical leverage and coercive behaviour.
6. Institutional tools already exist: Legal institutions, international treaties and negotiated frameworks have managed comparable concentration risks before.

What is the curse of “artificial wisdom” and why is it the most dangerous AI risk?

1. Core misconception: AI enthusiasts position AI as a substitute for human cognition, leading society to internalise the belief that AI generates knowledge.
2. What AI actually does: An AI system is trained on data to learn patterns and statistical relationships, and predicts the most probable next step in a sequence.
3. Knowledge versus information: Information is what AI produces; Knowledge: understanding that requires context, judgment, experience and an understanding of consequences.
4. Verification requires expertise: Only a human mind with domain expertise can judge whether AI-generated output is useful and appropriate for a given problem.
5. Why this risk is least understood: It is structurally different from labour and power risks because it changes how truth itself is assessed, not just who holds resources or jobs.

How does the information-knowledge conflation translate into systemic harm?

1. Synthetic information advantage: AI-generated content can be more persuasive, accessible or appealing than genuine information.
2. Erosion of fact-fabrication distinction: Individuals and institutions struggle to separate fact from fabrication, creating conditions for manipulation and misinformation.
3. Organisational dependence: Organisations increasingly use AI for research, coding, legal drafting and financial analysis.
4. Unverifiable decision-making: This creates systemic risk because decisions are influenced by intelligence that nobody is qualified to verify.
5. Paradox of expertise: The AI age makes genuine domain expertise more valuable, since the rarest skill becomes determining whether machine-generated answers are correct.

Why does AI’s accountability gap require a new governance architecture?

1. Existing liability model: Manufacturers of harmful pharmaceutical products can be held accountable under established liability law.
2. AI’s liability gap: AI systems have largely operated without comparable clear liability.
3. Emerging accountability signal: Meta Platforms has faced lawsuits alleging that its platform design contributed to harm among young users, indicating accountability boundaries are beginning to be redrawn for digital platforms.
4. Proposed safeguard structure: The response requires both technical and institutional safeguards, backed by a global non-proliferation agreement on disruptive AI.
5. Containment objective: Such an agreement must allow humans to limit or shut down AI systems operating outside their intended boundaries.
6. Precedent for restraint: Humanity has avoided nuclear catastrophe for eight decades; AI governance is framed as a comparable challenge of sustained, deliberate restraint.

Conclusion

The defining AI risk is not job loss or concentrated ownership, both of which have historical management precedents. It is the unchecked substitution of AI-generated information for genuine knowledge, compounded by the absence of liability and verification structures. Closing this gap requires a global governance architecture combining technical safeguards, institutional accountability, and a non-proliferation framework for disruptive AI capabilities, built before reliance on unverified AI output becomes irreversible.