South Africa pulls its draft AI policy after fake sources found in its references

South Africa had to pull its draft national policy on artificial intelligence. The reason: its list of sources included references that did not exist.

The draft was published for public comment in April 2026. On Sunday 26 April, the communications minister, Solly Malatsi, withdrew it. He said internal checks had "confirmed" that the draft "contains various fictitious sources in its reference list".

He did not play it down. "This failure is not a mere technical issue but has compromised the integrity and credibility of the draft policy," he said.

So how did fake sources get in? Malatsi gave his best guess, not a final answer. "The most plausible explanation is that AI-generated citations were included without proper verification," he said. AI chatbots can produce references that look real but are made up. That is a known weakness.

The timeline is short. Cabinet approved the draft in late March. It was published in the Government Gazette, the official record of government notices, on 10 April, with comments due by 10 June. It was withdrawn about two weeks later. In June, a formal notice in the Gazette withdrew the draft "in its entirety".

On 26 May, Parliament's communications committee called the department in to explain. It wanted answers on the withdrawal, on next steps, and on "consequence management" over claims that parts of the document had "AI-generated content with factitious and unverifiable sources".

The department had already acted. It placed two officials on precautionary suspension, SABC News reported. Neither has been named. The department's director-general, Nonkqubela Jordan-Dyani, told MPs that suspension "does not give the presumption of guilt", IOL reported.

The cause is still not settled. The director-general told the committee that "it has been said that ChatGPT was used in as far as the editing of the document itself", EWN reported. But she also said the fake references came from a document shared with the officials, according to IOL. She said the department had been told it was from Chile, and that the officials had translated it from Spanish. She said this was still being investigated. The committee chair questioned that account.

Now there is a newer twist. In October, TechCabal reported that one of the two suspended officials had resigned. Malatsi told TechCabal: "The ongoing processes now relate to the one official who remains an employee of the department." We found no official statement confirming the resignation.

He also told TechCabal the officials had not disclosed their use of AI. "Having such a major stain on an important policy is something that cannot just be allowed to slide," he said. He expects Cabinet to finalise a new policy before the end of March 2027.

There are two ways to see this. From the AI side, a tool that writes neat, confident references is not a fact-checker. It can invent sources that look perfect. From the human side, the failure was in the checks. Malatsi himself said the lapse "proves why vigilant human oversight over the use of artificial intelligence is critical".

The pattern is an invented source. A document meant to set the rules for AI was undone by references that were not properly verified.