Bismillah Bee is a 67-year-old widow living in an urban slum in Telangana, India. An algorithm denied her family subsidised food grains. The reason: the system confused her late husband, a rickshaw puller named Syed Ali, with a car owner named Syed Hyder Ali. The state accepted the algorithm’s output without question. Between 2014 and 2019, Telangana cancelled over 1.86 million food security cards and rejected 142,086 fresh applications without issuing any notice. The government attributed these exclusions to fraud detection. A joint investigation by Al Jazeera and the Pulitzer Center’s AI Accountability Network found thousands of these cancellations stemmed from faulty data and flawed algorithmic decisions made by a system called Samagra Vedika.

Posidex Technologies Private Limited built Samagra Vedika for the Telangana state government. The system pulls citizen data from over 30 government databases and creates “360-degree profiles.” Machine learning algorithms then compare records across these databases through a process called “entity resolution” to verify welfare eligibility and flag suspected fraud. Officials must consult the system before approving any welfare application. They retain the formal authority to override algorithmic outputs. In practice, they defer to the machine. An Amnesty International technical investigation published in April 2024 confirmed that the system raises serious human rights concerns. Neither Posidex nor the Telangana government has released any source code or accuracy data.

Article 11 of the ICESCR protects the right to adequate food. India’s National Food Security Act 2013 guarantees subsidised food grains to roughly two-thirds of the population. The Supreme Court of India, in PUCL v Union of India (2001), recognised the right to food as part of the right to life under Article 21 of the Constitution. An opaque algorithm now strips citizens of this entitlement. The state offers no explanation and no notice. Affected individuals receive no meaningful opportunity to contest the decision. Instead, excluded beneficiaries carry the burden of proving the algorithm wrong. This reverses the constitutional duty the state owes its citizens.

Other countries have confronted similar failures. In the Netherlands, the Hague District Court struck down a welfare fraud algorithm in the SyRI case (2020) under Article 8 of the ECHR, holding the system’s opacity violated proportionality. In Australia, courts ruled the Robodebt automated debt recovery scheme unlawful. The government repaid AUD 1.8 billion to over 400,000 people. In Serbia, Amnesty International’s 2023 research showed algorithmic welfare decisions disproportionately excluded Roma communities and persons with disabilities from social assistance. Each case carries the same lesson: opaque algorithms making high-stakes decisions about vulnerable populations demand transparency, oversight, and enforceable remedies.

India has no legal framework for algorithmic accountability in welfare delivery. The Digital Personal Data Protection Act 2023 does not address algorithmic discrimination. The law does not mandate impact assessments. Section 17 exempts government agencies on grounds of national security and public order. Welfare recipients have no statutory route to challenge algorithmic exclusions. Meanwhile, the Samagra Vedika model spreads. Haryana, Maharashtra, Tamil Nadu, and Uttar Pradesh have adopted or are developing similar platforms. In Haryana, a pension algorithm declared thousands of living pensioners dead, terminating their old-age and widow pensions. These individuals, many elderly and illiterate, were forced to prove they were alive.

Three reforms are urgent. First, India needs legislation requiring algorithmic impact assessments before any automated system enters welfare delivery. Second, the government must guarantee affected individuals a right to explanation and a meaningful process to challenge algorithmic decisions, consistent with Article 14 of the ICCPR. Third, private developers of welfare algorithms must face independent audits and transparency obligations. The Council of Europe’s Framework Convention on AI (2024) and the EU AI Act’s classification of welfare-related AI as “high-risk” provide working models. Without these safeguards, India’s expanding digital welfare infrastructure will deny food, pensions, and housing to the people who need them most.