Can AI pass Gandhi’s talisman? Inside India’s Rs 25,530-crore SARTHAK-PDS gamble

How NIRMAL, SAKSHAM and ASHA will work?

In the closing days of his life, Mahatma Gandhi offered the leaders of a nascent India a moral compass, famously known as his Talisman. “Recall the face of the poorest and the weakest man whom you may have seen,” he advised, “and ask yourself if the step you contemplate is going to be of any use to him.”

For nearly eight decades, the Indian state has attempted to operationalise this empathy through the Public Distribution System (PDS) — a heavily subsidised logistical leviathan designed to ensure that no citizen starves. Yet, human administration is inherently with imperfection. Getting food from Government silos to a family’s plate is full of hurdles, like stolen grain, fake accounts, and tech failures.

On May 27, 2026, the Cabinet Committee on Economic Affairs fundamentally shifted how the Government handles these

inefficiencies.  Backed by a Rs 25,530-crore budget through March 2031, the SARTHAK-PDS scheme goes far beyond basic digitisation. By integrating AI, machine learning, natural language processing, and blockchain directly into the National Food Security Act, the Indian Government is attempting something historic: entrusting an algorithm to execute Gandhi’s Talisman and protect the nation’s poorest

As an AI strategist who has studied digital infrastructure rollouts across the globe, I view this not just as a policy update, but as the most consequential technological gamble of India’s chapter.

The economic and strategic logic behind SARTHAK-PDS is immaculate. The scheme smartly consolidates the financial margins of over 500,000 Fair Price Shop (FPS) dealers and intra-state transport costs into a unified framework. When a ration shop is financially viable, the economic incentive to divert Government grain plummets.

But the real game-changer is the technology itself. The Government is rolling out three core AI platforms: NIRMAL to automatically weed out fake accounts, SAKSHAM to track and optimise the supply chain in real time, and ASHA, a multilingual assistant for resolving citizen complaints. If successful, this Rs 25,530-crore investment will pay for itself. By optimising logistics and minimising theft, the state will save hundreds of crores annually while reducing the carbon footprint of its supply chains. India is poised to build a welfare architecture that dwarfs the United Nations’ localised blockchain aid programs or Brazil’s AI — driven social security nets.

But there’s a massive gap between a flawless algorithm in a Delhi server room and a glitchy fingerprint scanner in a dusty Jharkhand village. Code is clean; reality is messy. If SARTHAK-PDS fails, it won’t be because the technology is bad, but because it ignored the harsh, on-the-ground realities of India. To stop this massive upgrade from becoming a nightmare for the poor, we have to urgently fix the blind spots where the tech actually meets the people.

The tragedy of calloused thumb

The most glaring vulnerability in India’s tech-welfare stack is the biometric exclusion trap. Our distribution architecture is inextricably linked to fingerprint authentication via electronic Point of Sale (e-PoS) devices. Yet, the primary beneficiaries of the PDS — construction workers, agricultural farmers, and manual laborers-earn their bread through physical toil. Decades of hard labor literally erase their biometric footprints.

When an e-PoS device repeatedly rejects a calloused thumb, the digital system logs a “fraudulent attempt”. The blockchain ledger remains pristine, the system registers zero leakage, but a legitimate citizen is turned away empty-handed. We cannot allow an AI system to starve a citizen simply because their labor has worn away their digital identity.

To dismantle this trap, the SARTHAK-PDS rollout must mandate multi-modal authentication across the entire FPS network. e-PoS machines must be universally equipped with iris scanning capabilities and UIDAI’s face-authentication technology. More importantly, we must institutionalise a digital “Nominee Protocol”. If a beneficiary suffers from chronic biometric failure, they must have the unalienable right to digitally nominate a trusted community member — a neighbour or relative — whose biometrics can trigger the release of their rations.

The illusion of blockchain

The scheme places immense faith in Blockchain to secure the supply chain. Blockchain is heralded for its immutability; once a transaction is verified, it cannot be secretly altered. This is a brilliant defense against data tampering, but it has some problems too.  SAKSHAM’s GPS tracking and cryptographic ledgers can mathematically guarantee that a truck carrying premium grain traveled from a depot directly to the local dealer without unauthorised detours.

But what stops a corrupt intermediary from physically opening those QR-coded sacks upon arrival, siphoning off the high-quality rice, and replacing it with rotting, substandard grain? The blockchain will happily report a 100 per cent successful, untampered delivery, providing a high-tech smokescreen for a low-tech crime.

Bits cannot secure atoms. To bridge this gap, we must deploy the Internet of Things (IoT) at the grassroots. Every transit vehicle must be equipped with IoT-enabled Smart Locks that can only be unlocked via a localised digital handshake with the authorised receiving dealer’s device. Furthermore, AI vision tools must be integrated into the dealer’s workflow. A simple smartphone photograph of the received sacks should allow an AI to instantly read the batch details and assess grain quality, verifying that the physical asset matches the digital ledger before it ever reaches the public.

The algorithmic panopticon

Machine learning models, like those powering NIRMAL, thrive on identifying deviations from the norm. Under the brilliant “One Nation, One Ration Card” framework, a migrant worker from Bihar working in Mumbai might suddenly withdraw three months of accumulated rations for their family in one go.

To an improperly calibrated AI trained to hunt for welfare fraud, this sudden, high-volume cross-state transaction looks highly suspicious. If the algorithm is empowered to automatically freeze the dealer’s or the citizen’s account, it places the burden of proof on the destitute. They are forced into a Kafkaesque bureaucratic maze to prove the AI wrong, all while their food supply is severed.

Algorithms must be relegated to the role of auditors, never executioners. SARTHAK-PDS must enforce a strict “Human-in-the-Loop” (HITL) protocol. If an AI flags a transaction as anomalous, the account should be marked for urgent administrative review, but the food supply must never be interrupted. This is where the ASHA AI assistant must be weaponised for empathy. Instead of just answering queries, ASHA should proactively call citizens in their local dialects if their status is under review, offering an instant, one-click voice appeal to a human district officer.

The myth of ubiquitous connectivity

In remote tribal belts, mountainous terrains, or during frequent rural internet blackouts, an e-PoS machine without a signal becomes a useless brick. If the system defaults to “deny access when offline” in the name of security, the technology transforms from an enabler of survival into a direct barrier to it. The entire e-PoS network must be rebuilt on a “Store-and-Forward” cryptographic architecture. The machines must be capable of authenticating a beneficiary offline — using encrypted local caches or smart card chips — and disbursing the ration immediately. Once the dealer moves the machine back into a connectivity zone, the system can bulk — upload and synchronise the ledger. Resilience must always take precedence over real-time tracking.

The ultimate test of statecraft

Ultimately, this Rs 25,530-crore investment is about more than just buying new tech. It’s a test to see if a Government system can be both automated and compassionate.

Technology is just the plumbing; empathy is what actually matters. We won’t judge the success of SARTHAK-PDS by how fancy its code or blockchain is. We’ll judge it by what happens when a laborer with worn-out fingerprints tries to get their ration in a village with no internet. If the system still recognises them and hands over their food, it works.

If India can build AI that genuinely serves the most vulnerable, we won’t just feed 813 million people — we’ll show the world how to build technology with a conscience.

The writer is a physicist at the University of North Carolina at Chapel Hill and a columnist on AI, infrastructure and global systems; Views presented are personal.