AI Transparency in South African Public Services: Building an Accountable Digital State

Overview

South Africa’s withdrawn Draft National Artificial Intelligence Policy created an important policy window. As the country reworks its national AI framework, it has an opportunity to establish practical requirements for transparency, auditability, human review and accountability before high-impact AI becomes deeply embedded across public services.

Artificial intelligence can help public institutions process information and support decisions. However, citizens may struggle to determine whether AI was used, how it influenced an outcome, who remains accountable and how the decision can be challenged. This becomes particularly important when AI affects access to public services, social programmes, employment, education, healthcare or other consequential government processes.

SADGF’s position is that AI should improve public administration without weakening citizens’ ability to understand, question or hold the state accountable. Governance requirements should increase according to the risk and consequences of the system’s use.

Public-service context

South Africa’s Draft National Artificial Intelligence Policy was published for public comment in April 2026. The draft emphasised ethical AI governance, including transparency, accountability, human-centred deployment and proposed governance structures.

The draft was subsequently withdrawn after an internal review found that its reference list contained fictitious sources. The Minister of Communications and Digital Technologies stated that this compromised the integrity and credibility of the document, and Cabinet approved its withdrawal so that the policy could be reworked.

The withdrawal should not be viewed only as a credibility failure. It also creates an opportunity to strengthen South Africa’s emerging AI-governance architecture by translating broad ethical principles into enforceable operational safeguards.

AI governance must remain connected to South Africa’s existing constitutional and administrative-accountability framework. Relevant protections include the Constitution, the Protection of Personal Information Act, the Promotion of Access to Information Act, the Promotion of Administrative Justice Act, administrative-law principles and public procurement requirements.

The policy challenge is not to replace these protections with an entirely separate AI regime. It is to translate them into clear requirements for an increasingly automated public sector.

Transparency, oversight and accountability

Transparency is a foundation of public trust. Citizens should not be expected to accept decisions merely because they were produced or supported by technology. Efficiency cannot replace accountability, and technological complexity should not prevent citizens from understanding how public decisions are made.

AI transparency means ensuring that citizens understand:

• When AI is being used
• What role the system plays in a process or decision
• Which public institution and officials remain accountable
• How they can request an explanation
• How they can seek human review or challenge an outcome

Transparency requirements should account for differences in language, disability, digital access and digital literacy. Information that is technically available but inaccessible to the affected citizen does not amount to meaningful transparency.

Human and institutional accountability must remain central to AI-assisted public decision-making. Government institutions remain responsible for decisions made with AI support, including when the underlying system is supplied or operated by a private technology provider.

Introducing AI into a government process must never reduce the accountability, explanation, review or redress that would have existed if a human official had made the decision. SADGF describes this principle as preventing an automated accountability gap.

Public accountability must follow the public function. Outsourcing technology must never amount to outsourcing the state’s constitutional and institutional responsibilities.

What counts as high-impact public-sector AI?

SADGF proposes that high-impact public-sector AI should include systems that materially influence decisions involving:

• Access to public services, benefits or social programmes
• Government assistance or the allocation of public resources
• Public-sector employment and recruitment
• Healthcare and education
• Law enforcement, security, immigration or identity
• Administrative penalties or enforcement
• Other legal, constitutional or socioeconomic interests

The definition should remain technology-neutral. Oversight should be determined by the consequences of the system’s use rather than the sophistication or branding of the technology.

Building trust in the digital state

South Africa’s digital transformation presents an opportunity to create a more responsive, efficient and accessible public sector. However, technological advancement must be matched by strong governance.

AI should not become a black box between citizens and the state. Citizens should understand how AI affects them, know who is accountable and have meaningful opportunities to question decisions and seek redress.

As South Africa reworks its National AI Policy, SADGF calls for practical mechanisms supporting transparency, human accountability, auditability, public oversight and redress. Public-sector AI must place the public interest at the centre of South Africa’s digital transformation.