Since July 2025, Himachal Pradesh has been the first Indian state to use Aadhaar face authentication to hand out subsidised grain at ration shops. It is not a national mandate and it has not replaced fingerprints: it was introduced precisely because the fingerprints of elderly manual labourers stop being readable. Every documented case of denied rations involves fingerprints or e-KYC verification, not faces.
On 16 May 2024, in the village of Aasan, Rajsamand district, Rajasthan, Chunni Bai Salvi pressed her finger onto the scanner of an ePoS device — the electronic point-of-sale terminal every Indian ration shop uses to release subsidised wheat and rice. She was eighty years old, with decades of construction and farm work behind her, including public works under the MGNREGA employment scheme. The scanner read nothing. The last ration she received that month was five kilos of wheat. By June her ration card had been cancelled, pending completion of electronic identity verification. The case was documented by the Indian magazine Down To Earth in an investigation published on 28 November 2024.
Just over a year later, on 5 July 2025, Himachal Pradesh announced it was the first Indian state to introduce Aadhaar face authentication — “FaceAuth” — for ration distribution. This is where artificial intelligence enters, and it is worth being precise about exactly where. The shopkeeper points the camera of his own smartphone at the beneficiary’s face; a dedicated app extracts a numerical representation of that face and sends it to the servers of UIDAI, the authority that runs India’s Aadhaar identity system, where a biometric recognition model compares it against the record linked to that twelve-digit number. The system produces no diagnosis and no forecast. It produces a yes or a no. On that yes or no depends whether a person collects the month’s grain. According to the announcement by the state’s Department of Digital Technologies and Governance and statements from Gokul Butail, principal adviser to the Chief Minister, the official rationale is twofold: the SMS carrying one-time passwords often fails to arrive, and fingerprints produce what UIDAI calls biometric mismatch.
Face authentication is not the cause of the problem. It is the patch.
This is the part of the story routinely told backwards. Face recognition in India’s Public Distribution System was not imposed from above on hundreds of millions of people, and it has not replaced fingerprints. It arrived state by state, as an alternative, and the reason it arrived is the failure of the layer beneath it. As things stand, there is no documented case of rations being denied because of a face recognition failure. The documented cases involve fingerprints, undelivered OTPs, and the mandatory e-KYC verification drive.
Revised ePoS device specifications published by India’s Department of Food & Public Distribution require hardware capable of “Biometric/Aadhaar based/Face Authentication” of the beneficiary against the UIDAI server, referencing earlier specifications from 2014, 2021 and early 2025. The phrasing is telling: three modes in the alternative, not a hierarchy of guarantees. There is also an official app built by the National Informatics Centre, “Mera eKYC”, which lets ration cardholders complete Aadhaar face authentication themselves. No private vendor sits in the middle of this: the matching engine is UIDAI’s own.
The scale, which is why error matters
The PDS is probably the largest food distribution system in the world. It covers roughly 204 million ration cards and around 806 million beneficiaries under the National Food Security Act. By July 2025, e-KYC verification had been completed for 658 million of 809 million beneficiaries — 82% — with pronounced lags in several states: Maharashtra at 54%, Haryana at 48%, Jharkhand at 74%, Bihar at 78%, Uttar Pradesh and Gujarat at 81%. In the course of the drive, 58 million ration cards — 5.8 crore, in Indian notation — were removed as fake or duplicate.
That figure is where the accounting breaks down. The government presents it as the cleaning of a register inflated by ghosts and duplicates, and some share of those cancellations certainly is that. Organisations working in villages document cancellations of genuine cardholders, and there is no independent public audit allowing anyone to say how much of the 58 million is error and how much is fraud. Anyone doing arithmetic on that number — in either direction — is guessing.
What actually went wrong in Rajasthan
The second case reported by Down To Earth is Mehna Ramlal, in her late eighties, a beneficiary under the Antyodaya category — the poorest tier, entitled to 35 kilos a month — in Rajsamand district. Card cancelled. Rations restored on 26 November 2024, after intervention by the Mazdoor Kisan Shakti Sangathan, a grassroots workers’ and farmers’ organisation, and its activist Shankar Singh. Her e-KYC was eventually completed with an iris scan. And here comes the most instructive detail: district official Om Paliwal confirmed that many shops have no iris-reading devices at all.
Formally, then, the alternatives exist: fingerprint, iris, OTP, and now in some states the face. Substantively, at the specific counter where a specific person shows up, only one may be available. That gap is the thing that matters and that no technical specification captures:
- the e-KYC deadline in Rajasthan slipped from September to November and then to 31 December 2024, yet cancellations happened before the deadline;
- OTP as a fallback for fingerprints was standard practice until May 2024;
- iris hardware is permitted but frequently absent;
- restoration, when it happens, runs through a civil-society group applying pressure, not through a procedure that triggers itself.
The vanishing-fingerprint problem among older manual workers is not a recent discovery: Indian local press has flagged it for years, for instance in Anantapur district, Andhra Pradesh. As far back as January 2018, Human Rights Watch documented grain being denied for lack of an Aadhaar number, failure to link it to a ration card, or fingerprint authentication failure, reporting starvation deaths alleged by local groups and the press — cases where the causal link was disputed by authorities.
The other programme, where faces are mandatory
There is one Indian setting where face recognition is not an alternative but a requirement, and it is not the PDS: Anganwadi centres, the village-level facilities of the Integrated Child Development Services, which handle nutrition and support for children, adolescents and pregnant women. Since 1 July 2025, face recognition has been mandatory for recording children’s attendance and for ration distribution there; since 1 August 2025 it also applies to new enrolments, with Aadhaar e-KYC and photo verification. The distinction matters and should be held firmly: ration shops and Anganwadi centres are two separate delivery chains with two different rulebooks.
On the Anganwadi side, a political row has already started. Jairam Ramesh of the Congress party accuses facial biometrics of excluding the poorest; Women and Child Development minister Annapurna Devi replied by citing growth in first-time beneficiaries of the PMMVY maternity benefit scheme, from 72.05 lakh in 2019-20 to 80.48 lakh in 2024-25. Those are enrolment figures, not authentication failure rates by age group — and failure rates by age, as far as can be verified, are not public.
The missing number is the interesting one
There is no official aggregate count of how many states currently offer face authentication at PDS shops, or what share of shops are equipped: the rollout is delegated to state departments, and the Himachal Pradesh reporting describes a launch, not full coverage. There are no public failure-rate statistics by age. There is no independent audit of the cancelled cards.
That is the part a European or American reader, accustomed to debating face recognition in terms of surveillance and vendor-claimed accuracy, probably does not expect. In the Indian PDS the decisive question is not how accurate the model is. It is whether anyone measures — and publishes — how often it says no to someone who was entitled. Until that number exists, adding one more biometric mode does not reduce exclusion: it moves exclusion to a different sensor and makes it harder to count, because each failure becomes attributable to the layer before it. In Himachal Pradesh, the face arrived as a remedy. A remedy whose effectiveness nobody is measuring is, from the outside, indistinguishable from a promise.