A R370 Grant, an R624 Line, and a Constitutional Fight
On August 25, 2026, South Africa's Supreme Court of Appeal (SCA) sat in Bloemfontein to hear the state's appeal against a landmark ruling on how SASSA, the South African Social Security Agency, decides who qualifies for the R370-a-month Social Relief of Distress (SRD) grant. The underlying judgment, Institute for Economic Justice and Another v Minister of Social Development and Others [2025] ZAGPPHC 29, was handed down by Gauteng High Court Judge Leonard Twala on January 23, 2025. It found several of the regulations governing the grant unconstitutional, including the R624 monthly income ceiling, the online-only application requirement, and the treatment of once-off payments and gifts as disqualifying "income." The Institute for Economic Justice (IEJ) and #PayTheGrants, represented by the Socio-Economic Rights Institute, brought the original case; the Department of Social Development, SASSA, and National Treasury are now asking the SCA to overturn it.
What makes this appeal more than a routine welfare dispute is who else showed up: the Global Center on AI Governance, admitted as amicus curiae through Wits University's Centre for Applied Legal Studies, arguing that automated administrative decisions carry distinct risks to fairness, accuracy, and accountability that South African administrative law has not yet squarely addressed.
How the Algorithm Actually Works
Since an August 2022 amendment pegged the SRD means test to that year's food poverty line, SASSA has run an automated monthly bank-account sweep against records from SARS, the UIF, NSFAS, commercial banks, and Home Affairs. Applicants whose combined inflows from every source exceed R624 in a given month are disqualified — regardless of whether that money was a wage, a one-off gift, a loan, or funds a family member asked them to hold. The High Court found this aggregation logic, combined with error-ridden and outdated source databases, produced arbitrary exclusions: people flagged as employed when they were not, or penalized for a single month's snapshot that told the system nothing about their actual financial need.
The Case for Automation, Stated Fairly
Before condemning the system, it's worth taking the government's strongest argument seriously. SASSA processes SRD applications for a pool advocacy groups estimate at 18.3 million working-age South Africans with incomes below the food poverty line — while Treasury funds coverage for fewer than half that number. At that scale, manual, case-by-case income verification is not a realistic alternative; it would be slower, inconsistent across thousands of caseworkers, and arguably more vulnerable to the kind of discretionary abuse and corruption that has plagued South African grant administration before. Cross-referencing SARS and UIF records algorithmically is, in principle, a more consistent and harder-to-game check than a human reviewing a bank statement. National Treasury's separation-of-powers argument also has real weight: a court ordering the executive to raise a specific rand threshold arguably strays into budget-setting, a function the Constitution assigns to the legislature and executive, not the judiciary.
Where the Design Breaks Down
But the case against SASSA's system was never really about automation as such — it was about a specific set of design choices that happened to be automated. Counting gross, undifferentiated deposits as "income" without asking what the money was for; relying on a single month's data as a permanent verdict; drawing on databases the state's own evidence conceded were unreliable; and offering no meaningful individualized channel to contest a wrong result — those are due-process failures that would be just as unlawful if a human clerk made the same call. The scale of the fallout is not abstract: SRD beneficiary numbers fell from roughly 16 million at the grant's 2020 launch to about 5.6 million by 2022, once automated verification tightened. South Africa already has a legal hook for exactly this problem. Section 71 of the Protection of Personal Information Act (POPIA) restricts decisions based solely on automated processing that carry legal or substantial effects, and requires that subjects be able to make representations and receive enough information about the decision logic to respond meaningfully. SASSA's SRD process — an automated, solely-machine-driven eligibility call with a notoriously thin appeal record — sits uneasily with that standard, even though POPIA has rarely been tested against government welfare systems rather than commercial credit-scoring.
The Right Fix Is Narrower Than Either Side Wants
The SCA does not need to choose between a fully automated means test and a fully manual one, and it shouldn't try to set R370 or R624 by judicial decree either — that risks exactly the separation-of-powers overreach Treasury warns about, and courts are poor budget-setters. The more durable outcome is procedural: require SASSA to disclose its verification logic, fix the demonstrably faulty database inputs, and build a real human-review channel for applicants the algorithm flags, rather than a check-box appeal that functions as a rubber stamp. That standard — automate the screening, but keep a genuine off-ramp to a human — is the same one privacy regulators elsewhere have converged on for high-stakes automated decisions, and it would let SASSA keep the efficiency gains of algorithmic matching without exporting its errors onto the country's poorest applicants. However the SCA rules, this case will likely become the reference point for how South African and wider African administrative law treats government-by-algorithm going forward.