This article was supported by the Pulitzer Center’s AI Accountability Network. Rajsamand (Rajasthan): Ghisi Devi, of Vijaypura village in Rajsamand district’s Deogarh panchayat, turned 70 years old in 2006, says her voter ID card. That would make her just short of 90 years old today. A paper card bearing the details of her Aadhaar – her biometrics-linked ID number – however, records her as 92 years old. It states she was born on January 1,1933, over a decade before India became independent from British colonial rule.On a crisp November noon, Devi, who worked on a farm all her life but has been bedridden for the last year, tried to sit up on her cot covered with blankets. A local community activist was trying to carry out her facial authentication on a mobile device. Her room was bare, with a chakki or millstone to grind flour by hand in a corner, and a handful of metal pots and plastic cans on bare shelves. It was dark inside, though the noon sun shone bright outside.Nikesh Kumar, the activist, asked Devi in Marwari to blink her eyes at the screen of the mobile phone he held up in his hands. But she is hard of hearing, and could not follow what he said. She looked warily towards the phone camera. The screen showed her face encircled with a green line that would turn red every few seconds, implying a failed attempt. This went on for several minutes. The activists and her family members gathered around her repeated – louder each time – that she must blink to get her pensions. Devi got increasingly irritated and finally, gestured that everyone should leave the room.Though economically poor women such as Devi need to only be over 55 years old to qualify for an old age pension of Rs 1,500 a month, in 2016-17, the Rajasthan government made it mandatory for pensioners to authenticate their identity by enrolling in the biometrics database and authenticating their Aadhaar details. Earlier, the postman would bring a money order and deliver the pension in cash at pensioners’ homes, and pensioners’ thumbprints on paper would count as annual verification. Now, they continue to qualify for their pension for the next year only after they have digitally marked themselves as alive.For Devi, whose fingers are too bent with old age for fingerprint authentication, and who cannot receive a one-time password as she has no mobile phone linked to her Aadhaar, the government has offered facial authentication. But it was not working that day.Finally, Devi’s grandson Pooran, who works as a helper on a local construction site, lifted her feeble body on his shoulder and carried her outside. He placed her on the ground, crouched against a wall, and then activist Nikesh Kumar tried to authenticate her face one more time. Finally, in the bright sun in the courtyard, her face matched the data stored on government servers, and there was palpable relief. “Devi has been marked alive online for the next 12 months, but what after that?” Shubham Biswas, one of the young activists, was worried.Nikesh Kumar, a community activist with MKSS, tried to carry out facial authentication for Ghisi Devi at her home but it failed several times. Photo: Anumeha Yadav‘Lifetime spent on data updates’Since 2009, the Indian government has collected personal data – fingerprints, iris scans, photographs – of 1.4 billion residents in its digital identification project “Aadhaar”(meaning, foundation), which is the world’s largest digital ID programme. Senior bureaucrats presented it as voluntary to enroll in, arguing the programme would offer millions of poor who lacked identification a valid ID, and prevent corruption by ensuring social benefits reach the right people. The government has made access to over 300 schemes and public benefits – food subsidies, pensions, maternity benefits – conditional on enrolling in the programme. To access social benefits, citizens must verify their fingerprints against the biometric data stored under their unique ID in a central database, by scanning fingers, or most recently, by providing their facial scans.Since its G20 presidency in 2023, the government has positioned itself as an exporter of digital governance solutions which it terms “digital public infrastructure” (DPI). A digital ID such as Aadhaar is being considered a “core DPI”. India has entered into an agreement with neighbouring Asian countries such as Sri Lanka and even the military junta-led government in Myanmar to export e-governance software products based on Aadhaar.India is now set to host the AI Impact Summit, a global meet of governments and other bodies to discuss safety and governance of artificial intelligence (AI). The government has highlighted Aadhaar as a prominent example of deploying AI successfully in public schemes in events leading up to the summit to be held in February 2026. At one such event on December 5 in New Delhi, a senior bureaucrat in the Union cabinet secretariat, Saurabh K. Tiwari, who is additional secretary Direct Benefit Transfer Mission, asserted that the Indian government uses AI tools such as machine learning in Aadhaar to “prevent duplication, detect spoofing and strengthen biometric verification”. He said that in social benefits transfers, “nearly 99 percent of applications now use biometric authentication, and most of these rely on face authentication”. Since 2023, Aadhaar facial authentication transactions have grown at more than 50 million transactions a month, and are set to expand.To carry out facial authentication using Aadhaar, an individual’s live facial image on a mobile phone camera is compared to their image saved in the Aadhaar database, often from the time of their initial enrollment, for a 1:1 match. As per reports, Unique Identification Authority of India (UIDAI)’s proprietary algorithm uses a match score based on an analysis of facial contours and measurements, such as the distances between the eyes and the nose to the mouth, comparing a specific facial image to the previous data stored against a gallery dataset.Ghisi Devi was able to get her face authenticated only after being physically carried outside her home. Photo: Anumeha YadavUIDAI, the central agency managing the database, says facial authentication is carried out only after a resident has given explicit consent, and “consent may or may not be required to obtain in facial recognition”. In practice, consent is not effectively possible, as not being able to provide biometric authentication can lead to missing out on life-supporting social benefits such as pensions or food.Community activists say that when facial features and other biometrics change, the most vulnerable – elderly and those with disabilities who belong to marginalised castes – face obstructions in updating and correcting their data, and get cut off from essential social services for long, even years, in the time it takes for them to get data rectified.“The reality is the poorest who are deserving and eligible have to struggle for very long to go through the updating and correction processes, and they are deprived of their benefits,” said Ranjit Singh, an activist with Mazdoor Kisan Shakti Sangathan (MKSS), a union that struggled to bring a right to information law in Rajasthan in 2005.Singh, who in November 2025 took part in a two-week protest march demanding accountability in public services, said the switch to digital processes ought to have brought greater transparency – but that’s not what happened.“The government now says if fingerprints fail, use the residents’ faces as a biometric. But we have witnessed that by a certain age, their faces have changed a lot and it does not match, and then citizens feel helpless,” said Singh. “Governments will come and go in Rajasthan, and residents will spend a lifetime in the attempt to correct and update their data online. No reasons are provided for why their data is not working, or [clarification given on] when it will work.”In November, MKSS rural activists carried out a two week-long march through villages demanding accountability in public schemes in Rajasthan. Photo: Shakeel KhanSubjects of dataOver 250 km from Rajsamand village where Devi lived, in the Jhalana Doongri commercial area in the state capital Jaipur, the Bhamashah State Data Center, the western desert state’s largest data centre, stands amidst a clutch of new buildings. Its glass facade towers over the neighbouring international convention centre. Residents’ data is stored on three of the eight floors, while there are offices on floors above. At a short distance are visible the rocky outcrops of the Aravalli range, which have been mined intensively for marble, granite and sandstone over the decades. The data centre is a new nerve centre for the state, as the government plans to shift governance more and more to data mining and automation.State officials in the Jan Aadhaar Authority told The Wire that the state government has employed Deloitte Touche Tohamatsu India to create digital profiles of residents and families, and stated that it will use data to streamline all social security schemes using AI. The data centre “Bhamashah” is named after a royal minister of 16th-century Mewar associated with benevolent generosity. The government will rely on AI-led digital profiles, which it calls “golden records” as a “single source of truth”, to make vital decisions about residents, such as who gets welfare, whether they qualify for food subsidies or pensions, calculating a family’s income or assets, and verifying a family’s welfare claims. Additionally, as reported earlier, the government will operate a data exchange to make aggregate public data available at a price to private companies such as startups.Gathering and storing data has become cheaper over time. The Economic Survey 2019-20, while advocating for use of digital technologies citing the large increase in data and cheap technologies available to the public sector, noted that a gigabyte of data cost Rs 61,050 in 1981 and was several thousand times cheaper – less than Rs 3.48 – in 2019. Rajasthan is in process of building more data centres, and has announced a township of data centres along the Delhi Mumbai Industrial Corridor.A fundamental element of this data-led governance infrastructure is the information residents provide to enroll in Aadhaar or when they use it, and the Jan Aadhaar database in which the Rajasthan government used data from Aadhaar to create a family IDs.The Bhamashah centre in Jhalana Doongri is operated by the Rajasthan Department of Information Technology and Communication, which is registered as an Authentication Service Agency (ASA) with the UIDAI. To carry out Aadhaar authentications at high speeds, the department maintains dedicated infrastructure of two parallel leased lines between this building in Jaipur and the UIDAI’s data centres in Manesar, Haryana and Hebbal, Bengaluru, in Karnataka in southern India.Around 250 km from the villages, residents’ data is stored, processed and analysed at the Bhamashah Data Center in Jaipur. Photo: Anumeha YadavOfficials working at this data centre were able to access macro data of Aadhaar authentication requests from every corner of the vast desert state as they came in, and pointed to a tangled web of lines on a white dashboard. The centre carries out analysis and produces reports such as department-wise transactions, total transactions that are updated every couple of minutes, as well as details such as average response time to requests, which of the requests failed, what were the causes of errors.Though granular authentication data down to even hourly transactions is available with the department, it is not published anywhere. For example, for Department of Social Justice schemes (which includes pensions, but also other schemes such as scholarships), on this day, at the beginning of May, there had been 28,969 attempts at biometric finger authentication. Of these, 5,951 attempts, or 20.5%, had failed. There had been 747 attempts at Aadhaar facial authentication using the government FaceRD mobile app that day, of which 390 had worked.Overall, between January 1 and April 30, 2025, the Department of Social Justice recorded 59,23,168 total transactions. Of these, the data showed, 76.27% of all fingerprint authentication requests were successful and 69.11% of all facial authentication requests were successful. Officials say that this does not reflect true success rates as these may be attempts by the same person, but they admitted that residents frequently had to make multiple attempts.As per information from the data centre, the highest success rates, over 90%, were recorded in case of Aadhaar authentication using one-time passwords (OTPs) sent on a mobile phones registered against the Aadhaar holder’s data. But officials in the Department of Social Justice and Empowerment who administer the social security pensions said that the government did not consider this mode reliable enough. “It is possible that a pensioner may pass away, but their family members keep wrongly collecting the pension for additional months till annual verification comes up,” said Pooran Singh, a deputy director in the pensions division.Instead, Rajasthan government tried other avenues. It launched a RAJSSP app on February 13, 2023 to carry out facial authentication of pensioners using UIDAI’s FaceRD (Registered Device) App. As per officials, Jaipur has over six lakh pensioners, and they had tested the app on 15-20 persons locally before rolling it out. The new system required pensioners blink a few times to ensure that someone was not using a still photograph to deceive the FaceRD system.How facial authentication works in Aadhaar. Source: UIDAIThe data centre officials said that in meetings with the UIDAI staff, various difficulties and barriers in fingerprint authentication came up as the main reason to switch to doing more facial authentication.“UIDAI has suggested to migrate more to facial authentication as over time, residents’ fingers become coarse and fingerprint scanners stop working and detecting fingerprints,” said an official who spoke off the record as he was not authorised to speak to the press. “Fingerprint scanners run up more costs, need updating and also need monitoring. For example, fingerprint cloning had become rampant, even at rural sites. So, the government updated from L0 to L1 devices which would check for Liveness – is a real human finger being scanned or is it a cloned print.”In a UIDAI report titled ‘Unlocking Facial Authentication’ co-authored by MicroSave Consulting, the agency notes that the reasons for the switch from using primarily fingerprint authentication to the push to facial authentication included that “facial authentication provided the highest level of accuracy” (at 84%) compared to fingerprints and iris scans (both at 81%). Further, is said that facial authentication required only smartphone camera apps and this would reduce costs of fingerprint and iris scanners. It argued that facial authentication has multimodal integration, and the advantage that is has AI-and machine learning-readiness.It noted that authentication failures and rising fraud or “biometric spoofing” was a key reason to switch to facial authentication.The government has used Aadhaar to carry out cash transfers by using the Aadhaar Enabled Payment System (AePS). In AePS transactions, business correspondents (BCs) or contractual agents carry out financial transactions in rural areas primarily through fingerprint-based Aadhaar authentication on hand-held devices to withdraw cash etc. Over time, it states, AePS transactions had slowed down, and nearly 10-15% of all transaction failures were due to biometric mismatches. In January 2024, the Ministry of Home Affairs stated that the previous year, AePS-related fraud constituted 11% of the 1.13 million cyber financial fraud cases in India.Aadhaar Payment Systems frauds were 11% of over one million cuber fraud cases registered, as per the Ministry of Home Affairs.In New Delhi, legal scholar Dr Usha Ramanathan, who had supported those who challenged the Aadhaar project in the Supreme Court, said that it had been clear that Aadhaar would worsen social exclusion and exacerbate inequalities. “After the UIDAI found that neither fingerprint nor iris authentication works very well in the Indian environment and have high failure rates, they now want a switch to facial authentication,” said Ramanathan. “Since 2010, the UIDAI has had no clue if biometric authentication will work or not, it has failed to provide for those it itself said are particularly vulnerable, and ignored the digital and literacy divide.”She added, “During the arguments in the Supreme Court, the UIDAI had claimed that beyond the debate over privacy rights, the Aadhaar-based systems will give the poor dignity in accessing food and pensions, and yet, when it fails, they take zero accountability and provide little redress. After such unreliable results back home, the effort is on to export this technology to other countries.”The CEO of the UIDAI did not respond to requests for comments emailed on December 9.Gudiya Kumari a resident of Beawar who lost access to her pensions after she was not able to give biometric verification cannot get an Aadhaar number. Photo: Anumeha YadavIn a case over the constitutionality of the Aadhaar project, the Supreme Court in 2018 had ruled that not having an Aadhaar ID could not be a reason to deny any resident a service. In a dissenting judgment, Justice D.Y. Chandrachud had recognised that biometric authentication could be erroneous, and that authentication failures as well as false positives (wrongly identifying someone) and false negatives (failure to identify someone) will lead to exclusion of those who are already marginalised: “Even a small percentage of error results in a population of crores being affected,” he stated. He described the government’s claims of “better targeting” and plugging leakages as “purported”.Justice Chandrachud had highlighted the absence of consent, the expansive scope of the term “biometrics” within the Aadhaar Act of 2016, the burden placed upon the individual to update her own biometrics, and the lack of accountability mechanisms with respect to the UIDAI. Several of the structural and legal gaps he cautioned about have come to be borne by those who are now mandated to use these technologies.On December 9, six unions of angandwadi workers, representing 1.1 lakh grassroots workers who provide supplemental meals in community creches to 28 lakh children and to more than nine lakh pregnant and lactating women in the state of Maharashtra, took the Union and the state government to the high court over coercive use of facial authentication. They argued that the system is error-prone and causes great hardship to the workers, as well as the residents it is catering to. The petitioners contended that demanding facial authentication and Aadhaar verification every time before such women can access their or their family’s take-home food rations is “unnecessary” and “excessive”, and has led to denial of nutrition to those who are eligible to receive supplemental food from the government. The case is set to be heard in January 2026.Divij Joshi, a doctoral researcher studying the governance of digital societies at the University College London, said that a distinct concern is this “extensibility of Aadhaar”: “It starts with one thing, and the administrators layer more on top of it without political discussion or consent.” He said there was a greater need for accountability on authentication failures, and that external audits such as by civil society, as well as greater legislative scrutiny, would help. “No meaningful consent is possible and no agency exists for a resident when it comes to choosing or forgoing their social welfare.”He pointed to concerns that globally, the evidence is that probabilistic technologies such as facial recognition have not always worked very well, or without bias. “In case of Aadhaar, the actual proof comes only when the government deploys it at scale of a billion. When the technology fails, there is nothing for people to fall back upon.”Invisible and on the marginsIn interior rural areas, such as in a hamlet of Adivasi Bhil or indigenous families living in Chhapli village in Rajsamand district, Dali Devi, a Fellow with School for Democracy, an NGO that trains grassroots activists in democratic politics, said she had witnessed elderly pensioners struggle with facial authentication. “It is quite hard and it often goes through only after several attempts,” she told The Wire. “I wonder if it is because their face no longer matches their Aadhaar photo from 10 years back when they enrolled, or, if with age, their eyelids have drooped, or their face has changed?”Elderly pensioners in Rajasthan who worked in mines or on farms all their lives often face trouble in biometric authentication. Photo: Anumeha YadavSingh, the social activist in Beawar, added that an additional problem is that those who have physical disabilities in many instances cannot get an Aadhaar number, even now. “There are two young women who I am trying to get an Aadhaar enrollment done for since years. One of them, 22-year-old Anju Narayan, has cerebral palsy and cannot get an Aadhaar, and two, Gudiya Kumari who uses a wheelchair, even went to New Delhi as part of a national press meet we had organised, because without an Aadhaar she cannot continue to get her disability pension since 2022. We met Aadhaar ans home ministry officials in the national capital. They told us, ‘She has to enroll for it locally’, and yet, she just cannot get an Aadhaar number even after three years of trying.” A medical board certificate from Rajasthan government states Kumari is 80 percent disabled.The activists wished that instead of focusing on so called ‘upgrades’ to technology, the government had looked at increasing the pension amounts, which they said had remained stagnant [https://thewire.in/news/budget-for-national-social-assistance-programme-remains-stagnant-yet-again ] or low. The criteria too were outdated, such as only those elderly whose entire family earned below Rs 48,000 a year could qualify. “If we consider an old couple, that means, their income must be below Rs 4,000 a month, or Rs 2,000 per person per month ($22 a month, or less than a dollar a day),,” said Rukma Devi, a former sarpanch and member of MKSS and Pension Parishad. “They are arguing about how to use biometrics verify if a pensioner is alive or dead, but how do they expect a family to live on less than Rs 67 a day?”This article was supported by the Pulitzer Center’s AI Accountability Network.Anumeha Yadav is a journalist focusing on labour and rural policy.