Two officials at South Africa's Department of Home Affairs were suspended after AI-generated hallucinations were discovered in official government documents. The suspensions, reported in May 2026, are among the first publicly documented cases of government employees facing formal disciplinary action for submitting AI output containing factual errors.
Hallucinations - when an AI model confidently produces information that is simply false - are a known limitation of large language models. The models generate plausible-sounding text based on patterns, with no built-in mechanism for checking whether the facts they produce are accurate. Tools like ChatGPT are widely used in office environments, often without formal policies governing when and how employees can use them.
The Home Affairs department handles identity documents, immigration records, and citizenship applications. The specific documents involved and the nature of the false information weren't disclosed publicly.
Where Accountability Lands
The suspensions reflect a shift in how organizations are assigning blame for AI errors. Through most of 2023 and 2024, the default response to hallucinations was to treat them as a tool limitation rather than a human failure. That framing is changing.
Whether the suspended officials used AI tools without authorization, or operated in a policy vacuum where no guidance existed, matters significantly for the precedent this sets. If there was no policy prohibiting AI use, the department shares responsibility for the gap. If they violated explicit rules, that's a different case.
Either way, this is what government IT departments have been quietly dreading: a hallucination-related failure serious enough to trigger personnel action, and public enough to attract scrutiny. More cases like this will pressure organizations to get specific about what AI use is permitted, what human review is required, and who owns the output when something goes wrong.