The division of house affairs is suspending two senior officers with instant impact after AI “hallucinations” had been discovered within the reference checklist appended to the revised white paper on citizenship, immigration and refugee safety – the second time in lower than per week {that a} cabinet-approved South African coverage doc has been caught with AI-fabricated sources.
A chief director within the unit accountable for the white paper was positioned on precautionary suspension on Thursday afternoon, with a director concerned within the drafting course of to observe on Monday, house affairs mentioned in an announcement.
The division has appointed two impartial legislation companies – one to handle the disciplinary course of and one to assessment each coverage doc it has produced since 30 November 2022, the day OpenAI’s ChatGPT was launched to the general public. Dwelling affairs has additionally dedicated to designing AI checks and declarations into its inside approval processes.
The revised white paper, championed by house affairs minister Leon Schreiber as essentially the most elementary reform of South Africa’s citizenship and immigration framework in a era, was accredited by cupboard on 3 April. The division mentioned its preliminary assessment suggests the suspect references had been generated and tacked on to the doc after the very fact – they aren’t cited within the physique of the textual content. The reference checklist has been withdrawn pending the end result of the impartial probe.
The physique of the white paper, DHA insisted, “continues to precisely replicate the federal government’s place” and is “not materially affected” by the obvious hallucinations. “The division nonetheless sincerely apologises for this unacceptable oversight,” it mentioned, describing the episode as “a chance to additional modernise our inside processes”.
AI coverage
The suspensions come 4 days after communications & digital applied sciences minister Solly Malatsi withdrew the total draft nationwide AI coverage within the wake of a News24 report that at the very least six of its 67 tutorial citations had been fabricated AI hallucinations – fictional articles attributed to actual journals and to authors who had by no means written on the matters in query. Cupboard accredited that coverage on 25 March, and it was gazetted for public touch upon 10 April.
Malatsi, in an announcement on his X account on Sunday, mentioned essentially the most believable clarification was that “AI-generated citations had been included with out correct verification”. He pledged “consequence administration” for these accountable for drafting and high quality assurance, calling the episode an “unacceptable lapse” that proved why “vigilant human oversight over using synthetic intelligence is vital”.
The parliamentary portfolio committee chair overseeing Malatsi’s division, Khusela Diko, recommended the division “skip utilizing ChatGPT this time” when redrafting the coverage.
Learn: ANC piles stress on Malatsi over AI coverage fiasco
The sample raises pointed questions on how extensively the follow has taken maintain throughout the general public sector. It’s presumably the explanation house affairs has now dedicated to reviewing each coverage doc it has produced over the previous three-and-a-half years. It additionally raises questions in regards to the high quality assurance processes that allowed two cabinet-approved paperwork in fast succession to clear inside scrutiny with manufactured references hooked up. — (c) 2026 NewsCentral Media
Get breaking information from TechCentral on WhatsApp. Enroll right here.
