Suspended officials and fake references: Inside SA’s draft AI policy fiasco

Two officials at the Department of Communications and Digital Technologies (DCDT) have been placed on precautionary suspension after fabricated sources were discovered in the reference list of the Draft South Africa Artificial Intelligence (AI) Policy, and the department withdrew the draft while investigating the matter.

What happened

The DCDT published the draft AI policy for public comment, with a closing date in mid-June 2026. After news emerged that the policy’s integrity was compromised because it contained fictitious sources in its reference list, Minister Solly Malatsi withdrew the draft and asked the director-general, Nonkqubela Jordan-Dyani, to launch an internal investigation.

Suspensions and the investigation

As a result of internal engagements, the director-general placed two officials on precautionary suspension. Malatsi told the Portfolio Committee on Communications and Digital Technologies that this step was taken to allow a thorough investigation while protecting the officials’ rights under the rules of natural justice and public service and labour regulations.

“Precautionary suspension, I just want to highlight, it does not give the presumption of guilt. It is a process that we need to adhere to, and therefore, we needed to ensure that once the investigation was taking place, we needed to make sure that we could do it without the necessary involvement of the officials as well,”

Jordan-Dyani said the matter is still under investigation but is in its final stages and that the department will update the committee on progress.

How the fictitious references entered the draft

The committee pressed the minister and department about how the error occurred. Jordan-Dyani said: “It has been said that ChatGPT was used in as far as the editing of the document itself, but the issue related to the fictitious reference, we’ve also been able to locate and pinpoint where it emanated from.”

She later told the committee that officials had used an “actual document” said to be from Chile, which had been translated from Spanish into English, and that the fictitious references in the draft policy traced back to that document.

“There’s an actual document that was shared with the officials, which they used as a context with the drafting of the draft AI policy. This document, we’re told, that it’s from Chile and they had translated it as well from Spanish into English, and that is the source from which the information emanates in terms of those references that have been contained in the draft policy,”

Committee chair Khusela Sangoni-Diko summarised the committee’s understanding during the meeting: “Just for clarity and as we step off this point, so you’re saying the suspended officials took a document from Chile, which had fictitious references, and then translated it from Spanish to English, and it then found itself in our national AI policy?” Jordan-Dyani responded that the point was part of the ongoing investigation.

Policy overhaul and expert review panel

To address the oversight failures that enabled the “AI hallucinations” in the policy, Malatsi said the department has created a seven-member advisory body, the National AI Expert Review Panel, to strengthen the integrity of the policy document.

The panel will be chaired by Professor Benjamin Rosman (Wits Machine Intelligence and Neural Discovery Institute) and includes:

• Professor Vukosi Marivate – Associate Professor at University of Pretoria and Director of African Institute for Data Science and Artificial Intelligence (AfriDSAI)
• Professor Alison Gildwald – Executive Director of Research ICT Africa
• Heather Irvine – partner at Bowmans
• Dr Tshepo Feela – Commissioner at the National Planning Commission
• Dr Jabu Mtsweni – Head of Cyber and Information Security Centre at the CSIR
• Advocate Lufuno Tshikalange – legal professional at Tshikosi Attorneys

Committee members questioned who would fund the advisory panel and why external experts were needed when the department has internal expertise.

Accountability and next steps

Malatsi acknowledged shortcomings in the department’s internal processes for authenticating and verifying AI use and said internal guardrails are needed for cases where AI dependence is not disclosed. He told the committee the department fell short in these processes.

“So when we reflected on this in terms of where the root cause emanates from… It has been said that ChatGPT was used in as far as the editing of the document itself,”

The committee ended the session with more questions than answers, and the DCDT said it will update Parliament as the investigation concludes.