India’s use of AI-based facial recognition in welfare delivery improves efficiency but raises concerns about exclusion when systems fail beneficiaries
India’s digital welfare infrastructure is among the most ambitious governance projects in the world. India’s welfare delivery system is currently built on the JAM trinity, i.e. Jan Dhan bank accounts, Aadhaar identity infrastructure, and mobile connectivity. The state today delivers subsidies, pensions, food support, healthcare, and employment benefits to hundreds of millions of citizens through this digital system. In recent years, this digital architecture has begun incorporating an Artificial Intelligence (AI) tool which uses facial recognition as a new method for beneficiary verification, particularly in schemes such as the Poshan Tracker for Anganwadi services and, more recently, under PM e-DRIVE, where Aadhaar-linked face eKYC is presented as a more seamless mode of authentication.
At the scale at which India operates its welfare programmes, even small leakages arising from duplicate records, ghost beneficiaries, and identity fraud can impose high fiscal costs and weaken public trust in service delivery. Facial recognition is being introduced as a more convenient and efficient alternative to earlier authentication methods, allowing beneficiaries to verify their identity through a simple smartphone-based scan. In principle, this promises faster verification, reduced dependence on physical documents, and more seamless delivery of benefits.
The promise of greater efficiency, however, also entails a higher degree of responsibility. When technologies such as facial recognition are deployed in welfare schemes, they do not merely streamline administrative processes; they become part of the mechanism through which people access food, nutrition, healthcare, and other essential services. In such contexts, technical failures can have consequences that extend far beyond administrative inconvenience. The challenge, therefore, is not whether technological innovation should be used in public service delivery, but how it can be implemented in ways that are inclusive, reliable, and responsive to beneficiaries’ lived realities.
Facial recognition importantly alters this arrangement. Eligibility alone is no longer sufficient; beneficiaries must also be successfully authenticated each time they seek to access a benefit.
Traditionally, access to welfare benefits was based on a relatively straightforward process. Individuals established their eligibility by submitting identity and supporting documents, and once enrolled in a scheme, the primary responsibility of the administration was to ensure that the promised benefit reached them. Technology was used to maintain records and improve administration, but it did not generally determine whether a person could access an entitlement at the point of delivery.
Facial recognition importantly alters this arrangement. Eligibility alone is no longer sufficient; beneficiaries must also be successfully authenticated each time they seek to access a benefit. In other words, entitlement becomes conditional on real-time verification. A pregnant woman may be fully eligible to receive nutritional support under the Integrated Child Development Services, yet still face delays if the system cannot match her live image to the photograph stored in the Aadhaar database.
The question is no longer only whether a person is entitled to a benefit, but whether the technology is able to recognise them at the moment they seek to receive it.
While this approach offers clear administrative advantages, it creates a digital audit trail, reduces reliance on specialised hardware, and can provide an alternative where fingerprint-based authentication is difficult. At the same time, it introduces a new dependency, i.e. access to welfare increasingly rests on the smooth functioning of cameras, phones, software, connectivity, and backend systems. As a result, the question is no longer only whether a person is entitled to a benefit, but whether the technology is able to recognise them at the moment they seek to receive it.
The limitations of facial recognition become most apparent when the technology is deployed in complex and uneven conditions in which welfare is delivered. These failures are not isolated events; instead, they arise from predictable features of both the technology and the environments in which it operates.
The first challenge lies in recognition accuracy. Facial recognition systems are sensitive to variables such as lighting, camera quality, ageing, illness, and changes in appearance over time. These factors can affect whether a live image matches the photograph stored in Aadhaar records. As a result, a beneficiary may be entirely genuine and fully eligible, yet still be rejected by the system.
A second challenge is infrastructural dependence. Facial authentication requires stable internet connectivity, functioning servers, and smartphones with adequate cameras. In many welfare settings, particularly in rural and low-resource areas, these conditions cannot be assumed. Beneficiaries have little control over whether the network is operational or whether the device being used is capable of capturing a usable image. A recent survey by the Pulitzer Centre found that three-quarters of Anganwadi workers under the ICDS scheme reported frequent network failures during facial recognition, with each attempt taking over three minutes and adding more than three hours to workers’ daily workload.
Occlusion of the face is a third major challenge, with masks being a notable example since the COVID-19 pandemic. Sunglasses, scarves, and even hairstyles can have similar effects by blocking key visual cues. Further, demographic bias is another critical challenge in implementing facial recognition technology. Studies have consistently shown that recognition systems often perform better for certain groups, particularly lighter-skinned males, than for others. These disparities are often caused by imbalances in training datasets, where some demographic groups are underrepresented, leading the models to learn less accurate representations for them.
Taken together, these limitations reveal that exclusion is not merely a technical failure. It is a foreseeable consequence of making access to welfare contingent on a technology that cannot be expected to function perfectly in every circumstance.
The most visible example of these challenges is the use of facial recognition in the Poshan Tracker under the Integrated Child Development Services (ICDS). Designed to improve monitoring and ensure that nutrition benefits reach intended recipients, the system requires beneficiaries to complete Aadhaar FACE authentication before receiving take-home rations and other nutritional support through Anganwadi centres. In practice, it has been noticed that in various states, authentication failures have delayed or disrupted access to food for pregnant women and lactating mothers.
Even children who are not eligible for Aadhaar cards, i.e. children up until the age of six years, are being denied benefits of the scheme. The system uses a parent’s identity, usually the mother’s. There have been cases in Bihar, Jharkhand, and Karnataka where facial verification failed to match the mother’s current face to her database photograph, and her child was dropped from the welfare rolls. Government data shows that only 52.7 percent of eligible beneficiaries received rations by the end of 2025. The Ministry of Women and Child Development has not disclosed how many of those excluded are children.
Similar concerns arise in the Public Distribution System (PDS), where facial recognition is increasingly being introduced as an alternative authentication mechanism to earlier methods like OTPs, fingerprints, and iris scans. In July 2025, Himachal Pradesh became the first state to implement Aadhaar-based facial authentication at fair price shops across the state. While the initiative is intended to make ration distribution more accessible, particularly for senior citizens and persons with disabilities, it continues to depend on the same underlying infrastructure, i.e. smartphones, internet connectivity, and UIDAI servers, that has historically contributed to authentication-related disruptions in the PDS.
The experience of schemes such as the Poshan Tracker and the Public Distribution System demonstrates that technology cannot be treated as an infallible gatekeeper to essential entitlements.
The adoption of Aadhaar FACE-based eKYC under PM e-DRIVE illustrates that this model is expanding beyond traditional welfare programmes into broader subsidy administration. Under the scheme, beneficiaries must complete an e-KYC Aadhaar FACE-authenticated e-Voucher to access incentives for electric vehicle purchases. Although this programme serves a different policy objective, it relies on the same underlying assumption that successful facial recognition can function as a reliable gateway to public benefits.
Taken together, these examples suggest that facial recognition is becoming a standard mode of state verification across increasingly diverse schemes, making the consequences of implementation failures correspondingly more significant.
The growing use of facial recognition in welfare delivery does not call for a rejection of technological innovation. At India’s scale, efforts to reduce leakages, improve targeting, and strengthen administrative efficiency are both legitimate and necessary. Facial recognition, when deployed appropriately, can simplify beneficiary verification and contribute to more streamlined delivery of public benefits.
At the same time, the experience of schemes such as the Poshan Tracker and the Public Distribution System demonstrates that technology cannot be treated as an infallible gatekeeper to essential entitlements. Policy discussions have rightly emphasised that AI should not become the final arbiter of eligibility and that human oversight, offline alternatives, and time-bound grievance redress mechanisms must remain integral to the system. These safeguards preserve the rights-based character of welfare programmes by ensuring that technical failures do not translate into denial of benefits.
The task, therefore, is not to choose between efficiency and inclusion, but to design systems that advance both. As facial recognition becomes more deeply embedded in India’s digital public infrastructure, transparency, accountability, and responsible system design will be critical to ensuring that technology strengthens welfare delivery while keeping beneficiaries, not algorithms, at the centre of public policy.
Tanusha Tyagi is a Research Assistant at the Centre for Digital Societies at Observer Research Foundation.
The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.
Tanusha Tyagi is a research assistant with the Centre for Digital Societies at ORF. Her research focuses on issues of emerging technologies, data protection and ...
Read More +
PREV
