The Department of Home Affairs has moved swiftly to suspend two senior officials after AI-generated “hallucinations” were discovered in the reference list of its revised white paper on citizenship, immigration and refugee protection, deepening concerns about how artificial intelligence is being used inside government. The move comes just days after another cabinet-approved South African policy document was pulled over fake citations, putting the spotlight on the public sector’s quality control systems and the growing risk of unchecked AI use in official work.
According to the department, a chief director in the unit responsible for the white paper was placed on precautionary suspension on Thursday afternoon, while a director who was also involved in the drafting process is expected to be suspended on Monday. Home Affairs said the disciplinary process will be handled by one of two independent law firms appointed to deal with the matter, while a second legal team will review every policy document produced by the department since 30 November 2022 — the date OpenAI’s ChatGPT was released to the public.
That timeline is telling. It suggests the department is not only reacting to a single embarrassing document, but is now trying to determine whether the problem stretches far wider across its policy pipeline. In practical terms, that review could expose just how deeply AI tools may have been used in drafting or editing official material, and whether the department’s internal approval systems were simply not built to catch fabricated references before they reached Cabinet.
The revised white paper itself has been one of Home Affairs Minister Leon Schreiber’s flagship reforms, billed as the most significant overhaul of South Africa’s citizenship and immigration framework in a generation. Cabinet approved the document on 3 April, but the department now says its initial checks suggest the suspicious citations were not part of the body of the white paper. Instead, they appear to have been generated and added later in the reference list, which has now been withdrawn pending the outcome of the independent probe.
Importantly, the department insists the core policy content remains intact. It said the body of the white paper “continues to accurately reflect the government’s position” and is “not materially affected” by the hallucinated references. Even so, the department has apologised and acknowledged that the oversight was unacceptable. In its own words, the episode is being treated as an opportunity to modernise internal processes — but for many observers, it also exposes a much bigger administrative weakness.
AI policy under fire after Home Affairs suspends officials over fake references
The Home Affairs suspensions come in the same week that South Africa’s broader AI policy debate was rocked by a near-identical scandal. On Sunday, Communications and Digital Technologies Minister Solly Malatsi withdrew the draft national AI policy after it emerged that at least six of its 67 academic citations were fabricated. Those references reportedly pointed to real journals and real authors, but the articles themselves did not exist — a classic case of AI hallucination slipping into an official government document.
That policy had already cleared Cabinet on 25 March and was published for public comment on 10 April, making the withdrawal particularly awkward for the department and for government more broadly. In a statement posted on his X account, Malatsi said the most likely explanation was that AI-generated citations were included without proper verification. He promised “consequence management” for the officials involved in drafting and quality assurance, and called the incident an “unacceptable lapse” that showed why human oversight remains essential when using artificial intelligence.
The political fallout has been swift. Parliamentary portfolio committee chair Khusela Diko did not mince her words, suggesting the department should “skip using ChatGPT this time” when redrafting the policy. The comment may have carried a touch of irony, but the underlying message was serious: public institutions cannot afford to rely on AI outputs without rigorous human checking, especially when those outputs shape laws, regulations and national strategy.
For SA Report readers, the bigger story is not just that two departments were caught out. It is that both incidents happened within days of each other, and both involved documents that had already climbed the ladder of official approval. That is a red flag. If fake citations can make it into Cabinet submissions twice in one week, then the problem is no longer a one-off drafting error. It becomes a governance issue.
The fact that Home Affairs has now committed to review every policy document it has produced over the past three-and-a-half years is a remarkable admission of risk. It suggests the department itself is unsure how far the practice may have spread, or whether other documents may also contain unverified AI-generated material. That kind of sweeping review is not something government departments announce lightly.
It also raises practical questions about who signed off these documents, what verification steps were followed, and whether civil servants were given clear guidance on the use of AI tools. In the private sector, companies are increasingly building policies around AI disclosure, source verification and human review. In government, especially where policy documents carry legal and constitutional consequences, the standards should arguably be even stricter.
There is also the reputational damage to consider. Leon Schreiber has positioned Home Affairs as a department trying to modernise and improve service delivery, while Solly Malatsi has pushed the AI conversation as part of South Africa’s digital future. But the recent disclosures have turned what should have been a forward-looking policy discussion into a cautionary tale about haste, convenience and weak oversight.
What makes the Home Affairs case particularly awkward is that the department says the suspect references were appended after the main text was completed, meaning the policy’s substance may still stand. Yet even if the body of the document is sound, the presence of fake sources in an official white paper is enough to damage confidence. Reference lists are not decorative; they are part of the credibility chain that allows lawmakers, stakeholders and the public to trust what government is saying.
The episode also lands at a time when South Africans are becoming more familiar with AI tools in everyday life, from writing support to research assistance. But as these incidents show, the line between using AI as a helper and letting it quietly undermine accuracy is dangerously thin. In public administration, that line must be policed properly, because once a fabricated source makes it into a policy document, the integrity of the entire process is called into question.
Home Affairs says the independent probe will now determine exactly how the hallucinated citations were introduced and who is responsible. Until then, the department has effectively conceded that its processes were not strong enough to catch the error before Cabinet approval. And with another AI policy scandal still fresh in the news cycle, the pressure on government to tighten controls around artificial intelligence has rarely been greater.