AI and the national security calculus

AI and the National Security Calculus

UPSC Study Note | GS-III (Internal Security, S&T) | GS-II (International Relations)

1. At a Glance

• AI has become a dual-use strategic technology: simultaneously driving civilian innovation and reshaping military doctrine, autonomous weapons, and intelligence operations. [S1]
• The "kill chain" acceleration — from target identification to legal approval to strike — via AI models marks a qualitative leap in warfare, raising sovereignty, legality, and accountability questions. [S4]
• Model distillation by adversary states (extracting intelligence from frontier AI systems) mirrors nuclear proliferation concerns, making AI governance a core national security issue. [S4]
• India is an active stakeholder: it launched 75 AI defence products in 2022 and has demonstrated indigenous AI-enabled military systems in Operation SINDOOR (2025). [S1][S2]

2. Why in the News

• March 2026: Anthropic (US AI lab) alleged that three Chinese AI firms — DeepSeek, MoonshotAI, and MiniMax — conducted industrial-scale model distillation attacks using ~24,000 fraudulent accounts and 16 million exchanges to extract knowledge from its Claude model. [S4]
• Anthropic called for these firms to be designated national security threats under US law. [S4]
• The U.S. Pentagon paradoxically labelled Anthropic itself a "supply chain risk" (a designation normally applied to foreign adversaries) after Anthropic raised concerns about its models being used in US military operations against Iran, reportedly to fast-track kill-chain approvals. [S4]
• Operation SINDOOR (May 2025): India demonstrated Aatmanirbhar (self-reliant) AI-enabled defence innovation, including the Akashteer AI-integrated air defence command system. [S2][S3]

3. Background & Evolution

4. Core Static Facts

Key Definitions: - Kill Chain: Military targeting sequence — Find → Fix → Track → Target → Engage → Assess (F2T2EA). AI compresses the human decision loop within this sequence. - Model Distillation: Training a smaller/weaker AI model on outputs of a stronger frontier model, effectively transferring its knowledge without direct access to training data. - Dual-Use Technology: Technology with legitimate civilian applications that can also be weaponised (analogous to nuclear, chemical, biological tech). - LAWS (Lethal Autonomous Weapon Systems): Weapons that select and engage targets without meaningful human control ("killer robots"). - Supply Chain Risk Designation: US national security tool labelling an entity as a systemic threat to defence procurement or technology chains.

Indian Institutional Architecture: - Defence AI Council (DAIC): Apex body; chaired by Raksha Mantri [S1] - Defence AI Project Agency (DAIPA): Executive agency for AI defence projects [S1] - DRDO: R&D for autonomous object detection, tracking, drone AI [S5] - ADITI Scheme: Ministry of Defence funding for deep-tech startups in AI, autonomous weapons, semiconductors, quantum tech [S1] - Akashteer: AI-integrated Air Defence Control & Reporting System (ADCRS) — automated real-time air picture [S3]

AIDef 2022 — 75 AI Products across 8 categories: [S1] 1. AI Platform Automation 2. Autonomous/Unmanned/Robotics 3. Blockchain-based Automation 4. C4ISR (Command, Control, Comms, Computers & Intelligence, Surveillance, Reconnaissance) 5. Cyber Security 6. Human Behavioural Analysis 7. Intelligent Monitoring Systems 8. Lethal Autonomous Weapon Systems

5. Multi-Dimensional Analysis

Geopolitical / Strategic

• The US–China AI rivalry has produced an informal "AI Cold War": export controls on chips (NVIDIA H100), restrictions on frontier model access, and now allegations of covert model distillation at industrial scale. [S4]
• AI enabling asymmetric warfare: smaller states or non-state actors can potentially access distilled versions of frontier military-grade AI. [S4]
• India's Akashteer system demonstrated credible multi-domain deterrence — a force multiplier in grey-zone conflicts — during Operation SINDOOR. [S3]
• The Pentagon's paradox (labelling its own AI partner a security risk) exposes governance gaps in AI procurement within democratic states. [S4]

Legal / Constitutional

• No binding international treaty on LAWS exists; the UN CCW debates remain inconclusive, with major military powers resisting binding constraints.
• Anthropic's court challenge to the Pentagon's "supply chain risk" designation raises questions about due process and civil-military AI governance in liberal democracies. [S4]
• The IT Act, 2000 and draft Digital India Act do not yet address autonomous weapons or AI-enabled military systems in Indian domestic law.
• Article 51 (UN Charter) — right to self-defence — is being reinterpreted in the context of AI-enabled pre-emptive strikes and accelerated kill chains.

Scientific / Technological

• Model distillation (the core technique alleged against Chinese labs) is a well-established ML method (e.g., Hinton et al., 2015); its weaponisation for intelligence extraction is the novel threat. [S4]
• DRDO is developing AI-based autonomous object detection and tracking for beyond-line-of-sight drone operations with optional man-in-the-loop architecture. [S5]
• Akashteer uses AI for real-time, automated fusion of air-defence data across the theatre — reducing human reaction time from minutes to seconds. [S3]
• The analogy to nuclear proliferation is imperfect: unlike fissile material, AI model weights are software (infinitely copyable), making containment structurally harder. [S4]

Ethical / Governance

• Meaningful Human Control (MHC) over lethal decisions is the central ethical demand from civil society and international bodies (ICRC, UN Secretary-General).
• The Anthropic case reveals a dual accountability gap: private AI labs are neither bound by military law nor subject to democratic oversight when their models are used in warfare. [S4]
• Algorithmic bias in target identification systems risks unlawful civilian casualties — a potential violation of International Humanitarian Law (IHL) principles of distinction and proportionality.
• India's AI ethics framework (NITI Aayog's Responsible AI for All, 2021) does not address military applications.

Economic

• The ADITI scheme (Ministry of Defence) funds deep-tech AI startups, creating a defence-industrial AI ecosystem under Aatmanirbhar Bharat. [S1]
• AI defence market globally projected at $38 billion by 2028 (SIPRI estimates); India's iDEX (Innovations for Defence Excellence) is a key incubation vehicle.
• US export controls on AI chips (Oct 2022, Oct 2023) directly impact India's own AI capacity-building; India has sought exemptions as a strategic partner.

Administrative

• India's DAIC → DAIPA two-tier structure mirrors the US JAIC (Joint AI Centre) model but lacks a published National AI Defence Strategy document.
• The Raksha Mantri's workshop on AI in National Security (PIB, 2019) was an early signal of institutional intent, but inter-ministerial coordination (MoD–MeitY–DRDO–ISRO) remains a bottleneck. [S6]
• Operation SINDOOR revealed strengths (Akashteer integration) and surfaced questions about AI-human interface protocols under combat stress. [S2][S3]

6. Recent Developments (Last 12–18 Months)

• May 2025 — Operation SINDOOR: India deployed Akashteer (AI air defence) and AI-enabled loitering munitions; PIB described it as "Aatmanirbhar Innovation in National Security." [S2][S3]
• 2025 — ADITI Scheme operationalised: Defence Ministry funding for AI, autonomous systems, quantum, semiconductors under deep-tech startup framework. [S1]
• Year-End Review 2025 (MoD): AI integration highlighted across Army, Navy, and Air Force logistics, predictive maintenance, and ISR domains. [S1]
• March 2026 — Anthropic vs. Pentagon: Pentagon designated Anthropic a "supply chain risk"; Anthropic challenged this in US federal court. [S4]
• March 2026 — US–Iran strikes: US military reportedly used AI models to accelerate kill-chain approvals; first publicly acknowledged instance of AI in an offensive strike decision loop. [S4]
• March 2026 — Chinese lab distillation scandal: Anthropic alleged DeepSeek, MoonshotAI, MiniMax distilled frontier US AI models via 24,000 fraudulent accounts, 16 million exchanges. [S4]

7. Prelims Hooks

1. India's first 'AI in Defence' (AIDef) symposium was held on 11 July 2022; 75 AI products/technologies were launched. [S1]
2. The apex body for AI in Indian defence is the Defence AI Council (DAIC), chaired by the Raksha Mantri. [S1]
3. The executive agency implementing AI defence projects in India is DAIPA (Defence AI Project Agency). [S1]
4. Akashteer is India's AI-integrated Air Defence Control & Reporting System; it was operationally demonstrated during Operation SINDOOR (May 2025). [S3]
5. ADITI scheme funds deep-tech startups in defence including AI, autonomous weapons, semiconductors, and quantum technology — under Ministry of Defence (not MeitY). [S1]
6. Model distillation involves training a weaker AI model on outputs of a stronger frontier model — the technique allegedly used by Chinese labs against Anthropic's Claude. [S4]
7. Anthropic alleged ~24,000 fraudulent accounts and 16 million exchanges were used to distill its AI models by Chinese labs. [S4]
8. The Pentagon labelled Anthropic a "supply chain risk" — a designation normally applied to foreign adversaries. [S4]
9. Three Chinese AI labs named by Anthropic as alleged distillers: DeepSeek, MoonshotAI, MiniMax. [S4]
10. The F2T2EA kill chain (Find-Fix-Track-Target-Engage-Assess) is the military targeting sequence that AI is compressing in modern warfare. [S4]
11. DRDO's Technology Development Fund (TDF) has an EOI for AI-based autonomous object detection and tracking for drone operations with/without man-in-the-loop. [S5]
12. AI in defence is treated analogously to nuclear technology in proliferation debates — though AI model weights are software, unlike fissile material, making containment structurally harder. [S4]
13. NITI Aayog's "Responsible AI for All" (2021) is India's primary AI ethics document — it does not cover military/defence applications.
14. India's task force on AI in Defence was established in 2018; it submitted recommendations within 3 months. [S1]

8. Mains Relevance

Plausible Mains Questions: 1. "The use of Artificial Intelligence in military kill chains raises fundamental questions about accountability, legality, and the erosion of meaningful human control. Critically examine in the context of India's national security doctrine." (GS-III/GS-IV, 250 words) 2. "Model distillation by adversary states represents a new vector of AI proliferation. How should India calibrate its technology export control framework and plurilateral commitments in response?" (GS-II/GS-III, 250 words) 3. "Operation SINDOOR demonstrated the potential of Aatmanirbhar AI in national security. Assess the institutional architecture India has built for AI in defence and the gaps that remain." (GS-III, 150 words)

9. Related Topics to Study Next

10. Common Errors / Trap Areas

1. DAIC vs DAIPA confusion: DAIC (Defence AI Council) is the apex policy body chaired by Raksha Mantri; DAIPA is the executive project agency — they are distinct.
2. ADITI scheme is under MoD, not MeitY: Aspirants often conflate defence-tech funding with MeitY's Digital India/startup schemes.
3. "Supply chain risk" designation: In the Anthropic case, the Pentagon applied this label to a domestic US company — not a foreign entity. This is the novel and examinable twist.
4. Model distillation ≠ hacking: Distillation is a legitimate ML technique; the violation alleged is terms-of-service fraud and export-control circumvention, not a cyberattack per se.
5. Nuclear analogy is imperfect: AI model weights are software — infinitely replicable at near-zero marginal cost. Unlike fissile material, physical containment is impossible. Exam questions may test whether students understand why AI proliferation governance is harder than nuclear non-proliferation.
6. Akashteer is air defence, not a weapon: It is a command & reporting system integrating radar and sensor data — not a kinetic weapon system itself.

11. Sources

• [S1] "Raksha Mantri launches 75 Artificial Intelligence products/technologies during first-ever 'AI in Defence' symposium" — https://www.pib.gov.in/PressReleasePage.aspx?PRID=1840740 — (Tier 1)
• [S2] "Operation SINDOOR: The Rise of Aatmanirbhar Innovation in National Security" — https://www.pib.gov.in/PressReleasePage.aspx?PRID=2128746 — (Tier 1)
• [S3] "Akashteer: The Unseen Force Behind India's New War Capability" — https://www.pib.gov.in/PressReleasePage.aspx?PRID=2129132 — (Tier 1)
• [S4] "AI and the national security calculus" — The Hindu, 11 March 2026, p. 8 (International), author: Bharath Reddy, Takshashila Institution — https://www.thehindu.com/todays-paper/2026-03-11/th_international/articleGPMFMSB82-13814016.ece — (Tier 4 / Article Primary Source)
• [S5] "Artificial Intelligence based futuristic system for autonomous object detection and tracking" — DRDO Technology Development Fund — https://tdf.drdo.gov.in/eoi/artificial-intelligence-based-futuristic-system-autonomous-object-detection-and-tracking-3429 — (Tier 1)
• [S6] "Raksha Mantri Inaugurates Workshop on AI in National Security and Defence" — https://pib.gov.in/newsite/PrintRelease.aspx?relid=179445 — (Tier 1)