Wisdom of Solomon: Hallucination-free (hopefully) — new draft AI policy will be ready by March, says Malatsi

It’s all fun and digital sovereignty until someone uses AI to draft a new national AI policy and you’re the minister who has to withdraw it.

There were many definitions of digital sovereignty thrown around on the GovTech 2026 stage, but only one man had to withdraw what he thought was a landmark moment for the country and his political legacy.

As it turns out, it’s harder to etch your name into the annals of South Africa’s democracy than it looks, and even harder when you’re doing it on a group project.

“From my side, we were very, very robust in the feedback to the department from the very first draft [of SA’s National AI Policy],” the minister of communications and digital technologies, Solly Malatsi, reveals to Daily Maverick at GovTech, which opened on Monday, 21 September at Durban’s ICC.

“And that’s why it was very, very disappointing when the AI hallucinations were discovered .”

It’s a full-circle moment because the draft policy was presented during a session at GovTech 2025, which is why we led with the question about his experience and what it taught him. His response:

“You hope – and it’s an informed hope, right? – that everyone in the execution of their responsibilities does the bare minimum ... because there are officials that have this responsibility and we deal with the substance of the policy, right?”

Then he drops the hammer of the shared experience and real future fear that anyone scrutinising policy now has to live with:

“And we didn’t interrogate the bibliography, because you take it as a given that anyone performing their functions ethically and honestly will not have an instance where there’s undeclared reliance on these technologies.

“Relying on AI in research, in developing of documents, is only unethical when it’s not declared, right?”

He’s not wrong, which is why it’s important to declare that this interview was transcribed from the audio recording using AI tools.

But the incident highlighted a critical, almost comical paradox: a policy designed to mandate human oversight for national AI deployment failed precisely because of a lack of human oversight in its own drafting.

When asked about the lessons from the incident, he said, “You should never drop your vigilance guard — go to every detail, a comma, and all of that. And it makes the job just increasingly difficult.”

To his credit, when the fabrications came to light, Malatsi chose immediate public retraction over deflection, establishing a high bar for executive accountability: