110 incidents on record · 2026 Headlights Incident reports by Ellie Harris · Melbourne
10 new this week Library last updated 27 July 2026
← The incident library
HD-INC-100
Government · India · 2022 · Automated-decision harm

Telangana used an algorithm to cancel millions of food-ration cards, and India's Supreme Court ordered all cancelled cards to be re-verified

By Ellie Harris · Filed Ration cards cancelled 2014 to 2019

Alleged: Government of Telangana; Posidex Technologies developed or deployed the AI system implicated in this incident. Details are drawn from public reports; parties are presumed innocent of any wrongdoing not established by an official finding.

Telangana used an algorithm to cancel millions of food-ration cards, and India's Supreme Court ordered all cancelled cards to be re-verified

What happened

It was reported that the Indian state of Telangana used a data-matching platform called Samagra Vedika to cancel the food-security ration cards of large numbers of families, and that many of those cancellations were later said to be wrong. An investigation by the Reporters’ Collective, published with Al Jazeera and the Pulitzer Center, found that the platform, built by a company called Posidex Technologies, pulled together data from dozens of government databases, covering things like electricity and water connections, property and vehicle ownership, and pension records, and used machine learning and entity-resolution techniques to build profiles of residents and flag the ones it judged ineligible. Between 2014 and 2019, the investigation found, the state cancelled more than 1.8 million existing ration cards and rejected 142,086 fresh applications, often with no notice to the families involved. Thousands of those exclusions, the journalists found, came down to faulty data and bad matches: people wrongly recorded as owning a car, paying property tax, or working for the government. The burden then fell on the families cut off to prove they were eligible.

In April 2022 the matter reached the Supreme Court, in a petition brought by the activist S Q Masood. According to reports of the hearing, the Court directed the state to carry out field verification of all the ration cards cancelled under a 2016 directive and to deal promptly with any aggrieved cardholder’s representation, noting that seventeen parameters had been used before cancellation and that the state “should have been more cautious before cancelling the cards.” Verification results afterwards, as the journalists described them, showed only a small fraction of reviewed applications being approved, which they took as a sign that the share of people wrongly cut off was higher than the government’s stated error rate. The state’s IT department defended the system as more efficient than the manual checks it replaced, and later reopened applications, the journalists noted, using the same technology.

What an auditable version would have shown

A ration card is meant to guarantee a family food, so behind every cancellation there is a simple question: what data said this family was not eligible, and was it right? An auditable version would answer that before the card is cut, not after the family turns up at the shop and is turned away. Each cancellation would come with a signed note of the match that caused it, the databases and fields it used, and the exact thing the family supposedly failed on. Add those up and you can see how often the matching gets it wrong, as a real number, the figure the journalists could only guess at from the verification that came later. Families being wrongly cut off would show in that number early, instead of only coming to light when a court steps in and orders a re-check.

Where the gap was

A family’s entitlement was withdrawn on the strength of an automated match, and any checking came afterwards, often, the journalists reported, with no notice to the family at all. A VerificationGate would hold that match to be confirmed, and the family given a chance to respond, before something as serious as a food entitlement is taken away. A ConductRecord would keep the basis of each cancellation, the data it rested on and the test it failed, so a wrongly excluded family could see why and argue back from the record instead of being left to prove a negative. The failure was not that data was used, but that an automated match was treated as the last word, with no accessible record tying each cancellation to the data behind it in a form the family could challenge.

What governance should have looked like

When a system decides who keeps the right to food, speed is not the test that matters, but getting it right is. Before an automated match takes away an entitlement, a person should check it and the family should be told, and the process should keep track of how often it gets things wrong rather than trust that it is efficient. Where the data cannot be shown to support a cancellation, it is the cancellation that should wait, not the family’s next meal. That is close to what the Court itself reached for: caution before cancelling, and a record that can tell each person, plainly, why they were cut off.

If this content raises concerns about access to food or essential support, help is available through local community and welfare services.

Failure Pattern: an entity-matching algorithm withdrew a statutory entitlement on the strength of imperfect data, frequently without prior notice or verification, and left affected people to prove their eligibility after being cut off.

Governance Principle: before an automated system withdraws a statutory entitlement, the match it relies on should be verified and the person notified, and each cancellation should carry a record of its basis that the person can contest.

The reference implementation of VerificationGate and ConductRecord is open source. It lives at github.com/saffronandindia/headlights-oss, Apache 2.0 licensed and free to install. The repository is public now.

Sources

The mailing list

Fresh incident reports every week. One email to match.

We add new incidents to the library regularly, and send a single short email each week with what's new. The library stays free and open; this is just how you keep up with it.

No tracking. Unsubscribe in one click.

The record

An auditable system would have produced a signed, tamper-evident record the moment this happened: what the system did, the version that did it, the basis it acted on, and the action taken, and Government of Telangana; Posidex Technologies could have produced it on demand.

This is the record the system as deployed did not produce in a signed, auditable form.

What this teaches
Capture what happened when it happens
What the system did, the version that did it, the basis it acted on, and the action taken, recorded at the moment, not reconstructed after.
Sign it, so no one has to trust the record-keeper
A tamper-evident entry. Edit it later and the signature breaks. The record does not ask for the benefit of the doubt.
Make it verifiable by anyone
A court, a regulator, a customer's lawyer can check the record themselves, without taking the company, or us, at our word.

Headlights summarises publicly reported AI incidents. All summaries are independently written, attributed to their original sources, and intended for research and educational purposes. Allegations are identified as such until established through official findings.

Last reviewed June 2026. This report is based on the sources listed above and reflects information available at the time of review; later developments may not be captured. Where a person is described as charged with or alleged to have done something, that allegation is unproven unless a conviction or a court or regulatory finding is stated. Headlights publishes journalism and commentary, not legal advice.

Want to write back?

Direct to my inbox.

ellie@useheadlights.com →