Two Department of Home Affairs (DHA) officials in South Africa have been suspended after AI-generated 'hallucinations'—fictitious references—were discovered in the revised white paper on citizenship, immigration, and refugee protection.

The suspensions, announced by DHA, involve a Chief Director and a director tied to the policy draft, with the department citing embarrassment and pledging to implement AI checks and declarations in future processes (Citizen, 2023). A parallel incident at the Department of Communications and Digital Technologies (DCDT) saw a National AI Policy draft withdrawn due to similar fictitious citations, highlighting a systemic issue in public sector AI adoption ( TechCentral, 2023). Beyond these incidents, the use of unverified AI outputs points to a critical gap in training and oversight, as large language models like ChatGPT, released in November 2022, often prioritize fluency over factual accuracy—a risk well-documented in academic studies ( Nature, 2023).

What mainstream coverage misses is the broader pattern of accountability deficits in AI deployment across government contexts globally, as seen in cases like the UK’s Horizon scandal where algorithmic errors led to wrongful convictions ( BBC, 2023). South Africa’s response—suspensions and independent reviews—mirrors reactive measures elsewhere but lacks proactive frameworks for AI ethics, unlike the EU’s AI Act which mandates risk assessments ( European Commission, 2023). This incident underscores a deeper vulnerability: without standardized vetting, AI tools can erode trust in public policy, especially in sensitive areas like immigration.

Finally, the DHA’s assertion that the policy content remains unaffected by the hallucinations is questionable, as unverified references could signal broader data integrity issues not yet addressed ( Citizen, 2023). The public sector’s rush to 'adapt' to AI, as DHA states, risks prioritizing speed over scrutiny—a trend evident in global AI adoption where hype often outpaces governance. South Africa’s case is a microcosm of a larger challenge: balancing AI’s transformative potential with the rigor needed to prevent systemic errors.