As India positions itself as a global voice on the responsible use of 'Artificial Intelligence in governance', the real test of that ambition lies in everyday systems that shape citizens’ lives. Few systems are as consequential or as complex as the public distribution system. Anchored in the National Food Security Act, 2013, the PDS reaches over 800 million people, making it one of the largest food security programmes in the world. When it works, it is a quiet pillar of economic stability and social protection. When it fails, the cost is immediate and human.
Over the past decade, India has invested heavily in digitising the PDS. Aadhaar-linked authentication, digitised beneficiary databases, electronic point-of-sale devices, and portability through One Nation One Ration Card have reshaped delivery. These reforms have improved transparency and mobility; according to government data, over 90 crore cumulative portability transactions have been recorded since the rollout of ONORC. Yet digitisation alone has not guaranteed reliability or dignity at the last mile. Authentication failures, connectivity gaps, and rigid exception handling continue to result in exclusion, especially for migrants, the elderly, and persons with disabilities.
As India looks ahead, AI offers an opportunity to move from digitised delivery to intelligent delivery, provided it is applied carefully, end-to-end, and with rights at the centre.
The PDS is best understood not as a single scheme, but as a value chain. It begins with estimating demand, moves through procurement, storage, and logistics, and ends at the fair price shop, where the State meets the citizen. Most failures occur not because a single node collapses but because the links between nodes are weak. This is where AI’s potential lies: not as a tool of surveillance or automated denial, but as a decision-support layer that improves coordination, anticipates stress, and responds faster when things go wrong.
An often-overlooked stress point in this chain lies even before grain enters storage, during procurement at mandis. During peak rabi and kharif seasons, delays or disruptions in mandi operations, whether due to late opening, administrative bottlenecks, or poor coordination, result in immediate grievances for farmers. AI-enabled forecasting, using historical procurement data, weather patterns, and sowing trends, can improve preparedness by anticipating crop arrivals, aligning procurement capacity, and synchronising transport and storage. Used responsibly, such tools can reduce uncertainty for farmers while preventing upstream stress from cascading into downstream failures within the PDS.
At the upstream end, demand estimation remains a vulnerability. Allocations are still largely anchored in static population estimates, even as migration, urbanisation, seasonal labour movement, and climate shocks reshape consumption patterns. ONORC has expanded beneficiary choice but has also introduced volatility at the local level. AI-enabled forecasting, using historical offtake patterns and anonymised transaction data, can help predict demand at the district or FPS level, reducing stock-outs and excess inventory that hurt beneficiaries and frontline dealers alike.
Further along the chain, procurement and storage account for persistent inefficiencies. Government audits and parliamentary standing committee reports have repeatedly flagged grain losses due to inadequate storage, delayed movement, and handling gaps. AI-supported warehouse management systems can help identify spoilage risks, optimise grain rotation, and flag anomalies in storage conditions. These are not headline reforms, but they matter. Grain lost upstream ultimately manifests as denial or delay at the last mile.
Logistics is another critical link. The movement of grain from depots to FPSs is vulnerable to route inefficiencies, delays, and leakages. Predictive analytics can help anticipate bottlenecks, optimise transport scheduling, and flag unusual stoppages for inspection. Used well, such systems improve predictability for administrators and dealers while reducing costs for the state. Used poorly, they risk becoming opaque monitoring tools. The distinction lies in design: AI should support timely delivery, not merely generate red flags.
The last mile remains the most visible and politically sensitive part of the PDS. While biometric authentication has reduced some diversion, it has also created new frictions. Government data and independent studies show that authentication failures remain a leading cause of exclusion, particularly in low-connectivity areas. Here, AI must assist service delivery, not harden denial. Decision-support tools can help FPS dealers anticipate peak demand, manage exceptions, and resolve failures locally. Crucially, AI must never become the final arbiter of entitlement. Human override and offline fallbacks are safeguards consistent with a rights-based law.
Perhaps the most underused opportunity lies in grievance redressal. Most states operate helplines and online portals, but grievances are often treated as static records rather than management intelligence. AI can change this by classifying complaints, prioritising severe cases, routing them to the correct authority, and tracking the quality, not just the speed, of resolution. When grievance data is systematically analysed, it can feed back into demand planning, logistics oversight, and dealer support.
Any serious conversation about AI in the PDS must also confront its risks. Automated systems can amplify bias against migrants, women, persons with disabilities, and those in low-connectivity regions. Predictive models can misread anomalies as fraud, and data collected for service delivery can expand into profiling. These risks demand clear guardrails: explainable models, audit trails for AI-assisted decisions, bias testing across social groups, adherence to the principles of data minimisation and purpose limitation, and strict limits on data use. There must also be clarity on where AI should not be used, including eligibility determination or automated fraud scoring without meaningful human review.
The PDS is a federal system, and AI deployment will reflect uneven state capacity. Central standards, shared infrastructure, and open technical frameworks will be essential to prevent a widening gap between high-capacity and low-capacity states. Equally important is investment in frontline capacity. FPS dealers and local officials must be trained to use AI-enabled tools as aids, not burdens. AI cannot replace administrative judgment, political accountability, or frontline capacity. But, if deployed with restraint and safeguards, it can make complex systems more responsive, anticipate stress before it becomes denial, and support officials and dealers in delivering services with greater reliability and dignity.
As India prepares to shape global conversations on AI in governance, the PDS offers a compelling test case for what this should mean in practice. If India can demonstrate that AI strengthens the PDS, from procurement to grievance redressal, while upholding rights and accountability, it will offer the world a governance model grounded in dignity, service quality, and democratic responsibility.
Sasmit Patra is a member of Parliament (Rajya Sabha) and advocate, Supreme Court. Suryaprabha Sadasivan is Senior Vice-President, Chase Advisors