Artificial intelligence (AI) is already here. Not just in business, in the political and electoral landscape too. And right now, the referee does not understand the rules of its own game. Days before telling political parties they must label their AI-generated content, South Africa’s own Electoral Commission (IEC) posted an AI-altered image to promote voter registration.

Phil Pelucha, revenue architect and AI governance commentator, outlines why this moment matters far beyond politics.

Nobody flagged it internally. Nobody labelled it. If the body writing the rulebook cannot follow its own draft rules, what hope does anyone else have? I said something close to this on national television earlier this year.

Mandatory labelling of AI-generated campaign content was not optional, I said, it was where this had to go. I was not guessing. I have watched this exact pattern play out in industries with far less on the line than an election. So, when the IEC’s own draft regulations surfaced, requiring parties to label AI-generated material and take responsibility for misinformation on their own platforms, I was not surprised. I was relieved. Not because I called it – being right is not the point.

The point is that the gap between what should happen and what does happen in AI governance is usually measured in years. Here, it is being measured in weeks. That is actually a good sign for South Africa, even if the Commission’s own conduct shows the habit has not caught up with the policy yet.

When trust becomes the cost

There is a reason this matters more than it appears to. This year, South African government ministers were caught relying on AI-generated content that turned out to be hallucinated – plausible-sounding, confidently stated, entirely false. My first reaction was amusement. It is not the first time I have heard this one. A man once tried to build his own AI lawyer and stood in front of a judge citing legal precedent that simply did not exist. Funny, until you think about what it actually costs.

Because here is the political version of that same mistake.

Political parties reaching voters through AI, often for the first time, are not just risking an embarrassing headline. They are risking the thing they can least afford to lose: trust.

A message that was previously solid, credible, worth hearing, becomes diluted the moment it is built on something false. And trust does not rebuild at the same speed it breaks.

It is the electoral equivalent of walking out of a job interview and only then thinking of the answer you should have given. Except in an election, there is no walking back in. AI can genuinely help a party reach people faster, understand what actually matters to them, and communicate with a precision that simply was not possible five years ago. That is real, and it is valuable.

Foundation before the decoration

But hand it the microphone without checking what it says first, and it will not just cost you one bad answer. It will cost you the room. Which brings me to the only real instruction I would give anyone reading this: a party, a regulator, or a voter. AI can be an extraordinarily powerful tool across the political landscape.

But right now, too many parties are trying to decorate the house before the concrete foundations have even set.

You do not win an election the way you post a viral image. You win it the same way you build a house – one brick at a time, verified, tested, load-bearing – before the next one goes on top.

An unlabelled image from the body that is supposed to set the standard. Ministers citing facts that never existed. It is the same mistake, in a different room. Skipping the foundation to get to the decoration faster. AI is not the risk in this election. Impatience is.

ALSO READ: New York City bans AI in public schools for students up to eighth grade