India's welfare stack is getting bigger, and the database behind it is starting to bite

On a Tuesday morning in late June, Aadhaar-enabled payment machines in village ration shops across Madhya Pradesh briefly stopped authenticating fingerprints for pensioners whose hands had roughened with seasonal agricultural work. The outage, reported locally and resolved within hours, surfaced a question that has migrated from technocratic memos into the everyday texture of Indian welfare delivery: what happens when the database that decides who eats, who gets paid, and who receives a pension also decides, by its own logic, who is real?

The Indian welfare stack has thickened faster than almost any comparable state apparatus in the democratic world. Above it sits Aadhaar, the twelve-digit biometric identity issued by the Unique Identification Authority of India to more than 1.3 billion residents, layered with the JAM Trinity (Jan Dhan bank accounts, Aadhaar, and mobile) that was supposed to let subsidies flow directly to citizens without leaking into the hands of middlemen. Below that sit the welfare schemes themselves: the public distribution system for food grains, the Mahatma Gandhi National Rural Employment Guarantee scheme, the National Social Assistance Programme for old-age and disability pensions, and the more recent direct-benefit transfers for cooking gas, fertilizer, and farm income support.

Each of those schemes now runs, to varying degrees, through a database. And each database makes exclusion cheap.

The friction of being legible

Two stories reported by Indian Express this month, sitting side by side in the newspaper's coverage, illustrate the texture of the shift. The first involves pension payments stalled or rejected because biometric authentication fails for elderly recipients whose fingerprints have worn down, whose eyes have changed, or whose records were seeded from older enumeration drives that did not align with Aadhaar enrolment. The second involves food-grain ration claims that returned errors at point-of-sale ration machines, sometimes because the household had migrated between states and the local ration database had not caught up, sometimes because the head of household had died and the ration card had not been transferred to a surviving family member's Aadhaar number.

Taken individually, these are administrative irritants. Taken together, they describe a state whose welfare delivery is now hostage to a system that can refuse service in milliseconds, without appeal, and without a clerk to argue with.

What the stack actually decides

The official theory of direct-benefit transfer is clean: money or in-kind entitlements move from the Treasury to a beneficiary's authenticated account, the leakage that once fed ghost beneficiaries and corrupt intermediaries is squeezed out, and the poor receive more of what was meant for them. The empirical literature on DBT leakage, much of it from the early years of the JAM rollout, did show reductions in the diversion of subsidy flows, particularly for cooking gas, where the bank-account-plus-Aadhaar link ended a well-documented market for ghost cylinders.

But the theory assumes a clean handoff between the database and the citizen, and that handoff is rarely clean. Biometric authentication fails for the elderly, the manual labourer, the disabled, and the infant. Database deduplication, the process of making sure no one appears twice, can remove the legitimate second claim of a migrant worker who registered in two places at different times. Address records lag the reality of where families actually live. And when a scheme's eligibility rules change, the database inherits the new rules and quietly stops paying the people who no longer match them.

This is not a glitch in the welfare stack. It is the welfare stack, working as designed, from the perspective of its designers.

The politics of who gets edited

The stack's administrators are insulated from the citizen by an authentication layer that does not negotiate. A pension officer in a block office can, in principle, override a failed biometric match with a manual exception. A point-of-sale ration machine, deciding in real time whether a transaction clears, cannot. The shift of authority from a discretionary human to an automated check has been a slow and under-reported transfer of power inside the Indian state, and the people on the wrong end of it are exactly those who have the least recourse to challenge a machine: the illiterate widow, the migrant labourer, the disabled pensioner.

Opposition parties, principally the Indian National Congress and a constellation of regional parties in Tamil Nadu, West Bengal, and Kerala, have begun to frame Aadhaar-linked exclusion as a governance failure of the current central government. The governing Bharatiya Janata Party's response has generally been that authentication failures are operational, not structural, and that grievance redressal windows at Common Service Centres and bank mitras are sufficient. Both framings are incomplete: the operational-versus-structural distinction misses the fact that, for the beneficiary, an operational failure that ends a pension and a structural design that ends a pension are indistinguishable.

What the database cannot apologise for

There is a temptation, in writing about this, to lean on the idea that the Indian state is becoming an algorithmic state in the manner of a Chinese social-credit system. The comparison is lazy and partly wrong. India's welfare database does not score citizens or rank them. It authenticates and routes entitlement. But it does share one feature with more intrusive systems: the difficulty of correcting it. When a household is removed from a ration database because of a deduplication sweep, the burden of proof falls on the household to demonstrate that they are not a duplicate, not on the database to demonstrate that they are. The default direction of error is exclusion.

That asymmetry is the policy choice sitting underneath the technology, and it has not been seriously debated in Parliament. The Aadhaar Act of 2016 established UIDAI as a statutory authority but did not create a robust, citizen-facing grievance mechanism for authentication failures inside welfare schemes; subsequent rulings by the Supreme Court of India, including the 2018 Puttaswamy-aligned verdicts on privacy, have placed limits on Aadhaar's compulsory use but have left the operational question of what happens when authentication fails at a ration shop largely to scheme-level circulars.

The stakes, in plain terms

The Indian welfare stack will keep growing. The 2024 expansion of PM-KISAN income support, the continuing rollout of the One Nation One Ration Card scheme for migrant portability, and the slow migration of MGNREGS wage payments into Aadhaar-linked bank accounts all point in the same direction: more entitlements, more databases, more exclusion-by-default.

The next round of political pressure will not be about whether to digitise welfare, which has already been decided. It will be about who carries the cost of authentication failure, and whether the Indian state is willing to build, at the same scale as its databases, an apparatus for admitting when they are wrong.