Serbia Social Card (Socijalna karta)

The Social Card (Socijalna karta) is a registry run by Serbia's Ministry of Labour, Employment, Veterans and Social Affairs. It cross-links roughly 130 to 135 categories of data from other state registers to check eligibility for social assistance. Its rules are undisclosed, and the sources show no evidence that it uses machine learning.

How it is used

The Law on the Social Card, adopted in 2021, gave the registry a statutory basis. It has been in force since 1 March 2022.

The registry draws on Tax Administration records, the Ministry of Interior's vehicle records, and property, pension, and banking data. The data cover each person receiving social assistance, their relatives, and so-called connected persons. The sources do not define connected persons. The registry uses them to verify eligibility and to detect undeclared income or assets.

The system is semi-automated. It flags discrepancies and sends notifications to social workers at municipal Centers for Social Work. The workers are meant to investigate before benefits are reduced, suspended, or removed. Over roughly two years, the Ministry processed more than 100,000 notifications of suspected increases in income or assets.

The registry's setup was financed as part of an 82.6 million euro World Bank public-sector loan. The World Bank also provided technical guidance.

What went wrong in documented cases

The Lab's case file, its written account of the sources, finds that the match turns informal, cash-based lives into a few sparse data points. The documented errors show how.

A one-off funeral donation of about 170 euro was flagged as income that disqualified a person. Two cars sold for scrap years earlier were still registered as assets. A collector of recyclable materials was recorded with income that implied selling far more material each month than was plausible. A single mother's seasonal jobs and vocational grant were read as regular income.

Amnesty International documented automation bias: workers deferring to the system instead of using their own judgment. Workers told beneficiaries there was nothing they could do, because the new system had decided.

Correcting a wrong flag

A removed person has 15 days to appeal. They must wait three months to reapply, whatever has changed in their circumstances. Removal letters often refer only to unspecified data from the electronic database.

Over roughly two years, beneficiaries filed 361 appeals against the rulings of the Centers for Social Work. In the same period, the Ministry processed more than 100,000 notifications. The two figures count different groups of people. So the gap shows how rarely flags were contested, not a measured appeal rate.

A11, the Initiative for Economic and Social Rights, alleges that social workers cannot correct errors recorded in the system, because it is semi-automated.

How many people lost assistance

Named sources report the caseload of financial social assistance falling by tens of thousands after the law. The figures depend on method and move over time.

A11, reading government data, records 211,266 recipients in February 2022 and 182,773 at the end of 2022. It records 168,307 by 1 February 2024, with at least 44,000 people having lost assistance since the law took effect.

Government figures cited by Amnesty International show about 176,000 recipients in August 2023. That is roughly 35,000 fewer than in March 2022. Earlier reporting by the Thomson Reuters Foundation put removals at about 27,000, some 15 percent of recipients.

The UN Working Group on Business and Human Rights visited Serbia from 6 to 15 October 2025. It reported that implementation of the law had left over 60,000 citizens without social assistance. Its full report is due to the Human Rights Council in June 2026.

These are largely net falls in the caseload, not audited counts of removals the system caused. The Serbian government attributes part of the fall to a stronger economy, so the cause is contested.

Who is most affected

The case file reports the system hitting the poorest, and Roma especially. The reporting says the informal earnings many Roma households rely on are misclassified as income. In 2023, Amnesty International reported that the system was worsening poverty, especially for Roma and people with disabilities.

The registry records no ethnicity. So the Roma figures are inferred from how many Roma households rely on benefits, and from case documentation. A 2019 UNICEF estimate, cited in the reporting, put the share of households in Roma settlements receiving some benefit at 84 percent.

Where the challenges stand

The system is operating and heavily contested. As of the latest reporting, no court or panel has ordered changes.

A11 filed a petition with the Constitutional Court of Serbia in April 2022. In late November 2022, a coalition from ESCR-Net, an international network for economic, social, and cultural rights, submitted a joint legal opinion to the court. It argued the law violates the rights to social security, equality and non-discrimination, privacy, due process, and remedy. The case remained pending.

On 31 March 2024, A11 filed a Request for Inspection with the World Bank Inspection Panel for six requesters, including Roma community members. It alleges that roughly 44,000 people were negatively and disproportionately affected. The Panel registered the request on 24 August 2024. It sent its Report and Recommendation to the World Bank's Board on 12 November 2024. As of 30 June 2025, the Board had not decided whether to authorize an investigation.

A11 has called for the law to be repealed. Amnesty International asks that the system be suspended, and made to respect rights if its risks cannot be prevented.

Where to look

Watch where correction is supposed to happen. Social workers are still part of the process. But the case file finds correction broken by the design of the system, not by individual neglect. A wrong flag counts again in the next match, because nothing pulls it back out.

The case file asks a live question. On what published basis, open to checking, may a state cross-link everyone's records and cut off the poorest? And who can correct it when the match is wrong?

What the available tools can and cannot address

A failure pathway is a link between two parts of the network where a mistake made by one part can be passed on to the other. A tool closes a failure pathway when mistakes stop passing along it. The work along it goes on.

Explore (No Targets) sets no targets. Under Service Targets Only, the targets can be met within this case's budget of 12 units. The cheapest way costs 3 units and uses one tool, Vet connections. Other ways pair two tools for 4 units, such as Mark AI-written records with Escalate checks.

Under Service and Safety Targets and All Governance Targets, the targets include closing every failure pathway. They can be met, but only just. Every combination of tools was checked. The cheapest way costs 11 of the 12 units and uses five tools: Understand the system, Escalate checks, Peer sharing rules, Vet connections, and Store less data.

Understand the system costs 4 units at these two levels. While it is on, Escalate checks, Peer sharing rules, and Vet connections each cost 1 unit less.

Under Service and Safety Targets, one more way costs all 12 units without Understand the system. It uses Mark AI-written records, Escalate checks, Peer sharing rules, Vet connections, and Store less data. Under All Governance Targets that way falls short. At that level, Store less data, Vet connections, Peer sharing rules, and Check copied records work less well without Understand the system.

Check copied records adds the check of a flag against its source. It closes none of the failure pathways, and it is in none of these ways. Upgrade model, Require sign-off, and Review on schedule close none either.

Using every tool at its strongest setting, at any cost, closes every failure pathway. But the added checks cost too much useful work to meet the service targets at any level that has targets.