Our AI policy.

Last updated Oct 1, 2026

Artificial intelligence can now write, design, edit video, manage social media, draft strategy and build entire campaigns. This is not a forecast. It is the working reality of our industry, and we are not going to insult you by pretending otherwise. The World Economic Forum puts as much as forty percent of communications jobs at risk of automation by the end of the decade. A press release that once cost a salaried week of someone's time now costs effectively nothing. Most agencies are quietly hoping their clients do not notice. We would rather start the conversation ourselves.

Here is why we are not afraid.

We have done this before. Cut to Black Media has adapted to every technological shift. In 2002, television content production provided more than half of our revenue. When online video arrived, we became a digital shop. When social media changed how the country talks to itself, we built the capability to work there.

By 2018 we were a fully integrated agency. In 2026 we are an AI-augmented newsroom. Twenty-four years, four reinventions, and a company that has come out of each one stronger. This is simply the fastest version of a thing we already know how to do.

And here is the argument itself.

When a machine can do the packaging, and increasingly the reach, for next to nothing, the obvious question is what is left that is actually worth paying for. The answer is the thing at the centre of everything we have shown you in this document. Judgement. The knowledge of this particular country, its history and its people, that lets you know which true thing to say and how to say it so it lands. The editorial instinct to tell when a story is right and when it is hollow. AI does not make that core less valuable. It makes it the only thing that is. As the outer rings get cheaper, the centre gets scarcer, and the centre is where we have always lived.

 

There is a legal dimension to this that we will not gloss over.

South Africa has no dedicated law for artificial intelligence. What governs our use of it, today, is the law we already have, chiefly the Protection of Personal Information Act. The law firm Bowmans recently warned South African businesses using ChatGPT, Gemini and Claude at work that POPIA already applies to how these tools are used, and that most companies are not treating it that way.

The sharpest edge of that warning is Section 71 of POPIA, which deals with automated decision-making. It prohibits any decision that has a legal or similarly significant effect on a person when that decision is based solely on the automated processing of their personal information, with no meaningful human judgement involved. Bowmans gave three examples that sit close to our own world of work: an AI system generating a performance rating, identifying someone for promotion or demotion, or recommending that a job candidate be rejected. Used that way, AI does not just carry legal risk. It becomes unlawful. Where a decision like this touches a colleague's employment, a candidate's prospects, or a member of the public on the receiving end of client communication we helped shape, that person must be able to see how the outcome was reached, challenge it, and speak to an actual human about it. A black box is not an acceptable final step in that chain, however well it writes.

The second risk sits with what we put into these tools in the first place. We are the responsible party for the personal information we hold, whether that is a colleague's HR record, a client's stakeholder database, or the contact details of a journalist we work with. Uploading any of that to a public AI tool, without the safeguards that tool actually offers, is our liability, not the platform's. This is why we use enterprise, no-train tiers of Claude, ChatGPT and Gemini rather than consumer versions, why client-identifiable or personal information never goes into an AI tool unless the client has approved it in writing and the tool's data handling has been checked, and why our automation runs through a self-hosted dashboard rather than an unvetted third-party platform wherever client data is involved.

In practice, this means three standing rules for everyone at Cut to Black Media.

THREE STANDING RULES

01

AI may suggest, summarise and accelerate, but it may never be the sole basis for a decision about a person's employment, a candidate's suitability, or a client's standing, without a named human who reviewed it and can explain how it was reached.

02

Personal or confidential information, whether it belongs to a colleague, a client or a member of the public, does not go into any AI tool outside our approved, enterprise environment, and never without checking first if you are unsure.

03

And every AI tool we bring into the business, for ourselves or proposed to a client, gets disclosed: what it does, what data it touches, where that data is hosted, and whether it is used to train the model. If you cannot answer those four questions about a tool, do not use it on live work. Ask first.

This is not caution for its own sake.
It is the same instinct that built this agency: know the terrain before you move fast on it.

So we use the tools, hard, and without sentiment. AI tests our campaigns, checks our outputs and optimises our schedules. It is, in our newsroom, a very fast and very willing junior: brief it well, check its work, and override it the moment it is wrong. The efficiency is real. So is the effectiveness. We are not cutting corners. We are cutting the time between a good idea and the people who need to hear it.

This is what we mean by the three phrases we keep returning to.

Newsroom-built

Because the standard came from journalism and never left.

AI-augmented

Because we refuse to fight the future with one hand tied behind our back.

Human-led

Because in the end someone has to decide what is true and what is worth saying, and that someone should never be a machine, and because the law agrees with us on that point.

We are not afraid of this moment. We are not backing away from it. We are building in it, the way we always have.