The assumption that doesn’t hold, what SA law actually looks for, and how to protect the parts that are genuinely yours.
“I made it with AI, so I own it” is the most common thing founders believe about this, and it isn’t how South African copyright works.
So what does give you ownership?
We asked Jacques Stemmet, a senior associate at Dommisse Attorneys who runs the firm’s IP and trademark practice, where the line actually sits…
The move: ownership follows human originality, not the prompt
For software, SA copyright generally points to the person who gave the instruction and controlled the system. But there’s a real difference between directing and controlling a build and typing a request, then taking whatever comes back. There’s no South African case law on this yet, so nobody can hand you certainty; the direction of travel requires human input and originality.
A prompt alone doesn’t give you ownership. Human originality does.
How to make sure you own the work your AI produces
1. Stop assuming the output is automatically yours
Start from the position that raw AI output may not be protectable by you at all. That matters most for the things you’d expect to defend: core product code, brand assets, written material you plan to license. If you couldn’t point to what a human contributed, assume you’re on thin ice until the courts say otherwise.
2. Put real human work into it, and keep the evidence
The safer ground is where a person meaningfully shapes, edits and builds on what the machine produced. The US takes a similar line, protecting bespoke human additions to the extent of those additions, and the EU broadly tracks the same approach. Keep some record of that human work, because if ownership is ever questioned, the contribution you can show is the part you can defend.
3. Read the terms of every tool you build with
Copyright is only half of it; the other half is what you agreed to when you signed up. Terms differ sharply between tools and change over time, with some stating plainly that the output is yours and others reserving rights you might not expect. Check each tool your team actually builds with, and check again when the terms update.
4. Protect the things that were always yours
The most valuable IP in an AI-era business usually isn’t the generated output. It’s your domain expertise, the judgement about what to build and how it should work, plus your prompt libraries, training data and fine-tuned models. Most of that lives in trade-secret territory rather than registered IP, which means it stays protected only as long as you take reasonable steps to keep it secret.
The big payoff
You stop treating AI output as an asset you automatically own, and start protecting the thing that actually gives you an edge: the human judgement behind what got built, and the know-how that made the machine useful in the first place.
You might also be interested in
Want the full playbook?
This is one piece of What You Own vs What You Think You Own, Jacques’s full masterclass inside the Founder Collab, which walks through the whole system for proving and protecting what your business owns:
The full IP Control Stack: the five layers that decide whether your ownership holds up
Why your architecture, data sets and fine-tuned models count as IP, and how to protect them
How free AI tools can quietly destroy your trade secrets, and the usage policy that prevents it
Open-source licence traps that can compromise your proprietary code
What “good enough” IP hygiene looks like at pre-seed, seed and growth stage
You’ll also get access to 40+ other masterclasses from SA founders and operators on sales, fundraising, UX, paid media and more inside The Founder Collab.
Get more SA tech and business news, tips and business-building workflows and subscribe to The Open Letter.
