Technology Law · AI & IP

AI-Generated Software & Copyright in South Africa

Who owns the code when AI helped write it — and how to protect software when copyright is uncertain. A plain-language guide for founders and business owners, with the actual words of the law beside every point.

16 min readMartin Kotze — Attorney

Written by

Martin Kotze

Attorney, Conveyancer & Notary Public

Quick answer

In South Africa, software you build with AI is usually still protected by copyright if a person meaningfully directs and shapes it — choosing the architecture, editing and rejecting output, testing and integrating the code. The Copyright Act 98 of 1978 gives copyright in a computer program to the person who “exercised control over the making” of it, and the courts (Haupt, Bergh) reward real, hands-on control over the code. Two traps catch business owners: paying an outside developer does not make you the owner (software is not on the Act’s “commissioned work” list — you need a signed written assignment), and fully AI-generated code, with little human input, may not be protected at all. As at June 2026 there is no AI-specific statute. The practical answer: get the IP in writing and protect the code with contracts and confidentiality, not copyright alone.

More and more South African software is written with help from AI tools. That raises a simple, high-stakes question for any business: do we actually own what we’re building or buying — and can we stop someone else from copying it?

The short version is that South African law has not “cancelled” copyright in software because AI was involved. But AI does shift the risk. Ownership now turns much more sharply on who did what, and on whether you have the right paperwork. This guide explains the position in plain English, and shows you the exact words of the law and the leading cases next to each point so you can see it for yourself. It is general information, not legal advice, and reflects the position as at June 2026.

1. The short answer

Think of AI as a very capable assistant. If a person is genuinely directing that assistant — making the design decisions, editing and rejecting its output, testing it and deciding what ships — then the resulting software is usually still protected, and the human (or their employer or client, depending on the deal) can own it. This is AI-assisted code.

But if a person just types a one-line instruction and ships whatever the model produces, with little real input, that is closer to fully AI-generated code — and South African copyright becomes much harder to rely on, because the law still looks for a human author who controlled the making of the work.

The practical takeaway: AI-assisted software is usually protectable; fully AI-generated software is legally fragile. Either way, paying for software does not automatically make you its owner — so get the IP in writing, and back it up with confidentiality and good records.

2. Is software even protected by copyright here?

Yes. Since 1992, a “computer program” is its own special category of work that can be protected by copyright in South Africa. You don’t register it — protection is automatic once two things are true: the work is original (it comes from a human author and isn’t just copied), and it has been written down or stored in some form (saving code to a file is enough).

“Original” in copyright doesn’t mean clever or novel. It means the work must come from the author’s own effort and not be copied. That’s the part AI puts under pressure: if almost everything came from the model, it gets harder to say the protected expression really came from a person.

3. Who counts as the “author”?

Here is the part that surprises people. For software, the “author” is not defined as the person who typed the code. It is the person who controlled the making of the program. A founder, product owner or technical lead who genuinely directs the build can be the author, even if someone else (or something else) does the actual coding.

There is a second, less-known rule that matters a lot for AI. The same Act has a special rule for works that are “computer-generated” — and, importantly, that rule expressly includes computer programs. For a computer-generated work, the author is the person who made the “arrangements necessary” for it to be created.

Why this matters for AI

South African software has two possible author rules: control over the making (for human-authored code) and “arrangements necessary” (for computer-generated code). No South African court has yet decided which one governs when an AI does most of the writing. The cautious, stronger position is to make sure a real person can show control over the making — because that is the test the courts have actually applied to software.

4. Trap 1: paying for software is not the same as owning it

This is the most expensive misunderstanding in South African software. For some kinds of work — a commissioned photograph, portrait, gravure, film or sound recording — the person who pays automatically owns the copyright. Software is deliberately not on that list.

So if you hire an outside developer, agency or “AI dev shop” to build your product, paying the invoice does not make you the copyright owner. To actually own it, you need a written assignment, signed by the developer. The Act is strict about this: an assignment (or an exclusive licence) is not effective unless it is in writing and signed.

If an employee writes the software as part of their job, the position flips in your favour: the employer owns it by default. The key is that they are an actual employee under a contract of service — not an independent contractor.

5. Trap 2: fully AI-made code may not be protected at all

If the human role shrinks to a high-level request — “build me an invoicing module with VAT” — and the model produces the code substantially on its own, a South African court could find that the person supplied only requirements and a quick review. On the reasoning in Bergh (below), that may not be enough “control over the making” to make anyone the author. If no one is the author, there may be no copyright to own.

That doesn’t mean the software is worthless — it means copyright is the wrong thing to lean on. You protect it with contracts, confidentiality, access control and good records instead (see section 8).

6. What the leading South African cases say

Two Supreme Court of Appeal decisions draw the line, and they map almost perfectly onto AI workflows.

Haupt — hands-on control wins

In Haupt, the person claiming ownership wasn’t the programmer — but he directed the work so closely that the court treated him as the author. He gave detailed instructions, stayed in constant contact, and checked and approved the work as it progressed.

Bergh — requirements and a review are not enough

In Bergh, an organisation argued it owned a program (BeefPro) it had commissioned. It lost. The developer had worked independently, and giving requirements plus checking the result was not “control over the making”. This is exactly the risk with one-shot AI prompting.

A third case, King v SA Weather Service, confirms the employment rule: programs an employee writes in the course of their job belong to the employer (on the facts of King, a state body, so they vested in the State). Arguments like “I wrote the prompts at home” won’t usually help if the work was made for the job.

7. How other countries treat AI-made work

South African courts can look to other countries for guidance. The pattern is clear: almost everywhere, copyright needs a human behind the work.

  • UKHas a specific rule for “computer-generated” works (the author is whoever made the arrangements for creation) — the closest match to South Africa’s wording. But even there, the courts still look for a real human contribution.
  • EUA computer program is protected only if it is the author’s “own intellectual creation”, and the author is a natural person — leaving little room for machine-only output.
  • USThe Copyright Office and the courts require human authorship; purely AI-generated material isn’t registrable.
  • AULike South Africa, a common-law system insisting on human “independent intellectual effort” — automated output has been refused protection.

8. How to protect AI-generated software

When copyright is uncertain, don’t bet everything on it. Build protection in layers — most of these don’t depend on proving who authored each line of code.

1

Contracts first

Every employee, founder, contractor and agency signs an agreement that assigns IP, allocates open-source and AI risk, and requires they keep prompt and edit records.

2

Trade secrets & confidentiality

Protect source code, prompts, configurations and know-how as confidential information — segment repositories, vault secrets, log access.

3

Escrow & continuity

Use source-code escrow with release triggers (supplier insolvency, breach) so you keep running if a vendor disappears.

4

Provenance & open-source

Keep an AI-tools register, a software bill of materials, and commit history showing human edits — proof of control and of clean licensing.

5

Protect what copyright can’t

Register trademarks early; use patents or registered designs selectively where there is a genuine invention or distinctive design.

9. Where the law is heading

South Africa is still working out its AI rules — slowly. A National AI Policy Framework was published for comment in October 2024. A Draft National AI Policy was then gazetted in April 2026, but the government withdrew it within weeks after discovering that some of its references were fabricated and apparently AI-generated. A revised draft is expected for comment around January 2027.

The Copyright Amendment Bill — which would change parts of this area — is still before the Constitutional Court; it was heard in May 2025 and, as at June 2026, judgment was still awaited. The message for business owners is simple: don’t wait for the law to catch up. Put the contracts and records in place now, on the assumption that some AI-generated code may attract weak or no copyright.

10. Frequently asked questions

Who owns software that AI helped write in South Africa?

It depends on who controlled the making of the code and on your contracts. Copyright in a computer program first belongs to its "author" — the person who "exercised control over the making" of it, not necessarily whoever typed it. Employee work belongs to the employer; work by an outside developer needs a signed written assignment before you own it; and code generated almost entirely by AI may not be protected at all.

If I pay a developer or AI dev shop, do I own the code?

Not automatically. The Copyright Act gives the payer automatic ownership only for commissioned photographs, portraits, gravures, films and sound recordings — software is not on that list. You need a written assignment signed by the developer (s 22(3)); without it you may only have a licence to use the software.

Does using GitHub Copilot or ChatGPT mean I own the output?

Between you and the provider, broadly yes — GitHub does not claim ownership of your output and OpenAI assigns you whatever rights it has. But those terms cannot create South African copyright where the law would not, cannot cure a defective written assignment, and do not remove third-party or open-source risk.

Can an AI be the author or owner of copyright?

No. South African law has no rule making a machine an author, and no reported case has recognised one. The question is whether a human or company behind the AI controlled the making of the code enough to be its author.

Is there a law specifically about AI in South Africa?

Not yet, as at June 2026. A National AI Policy Framework was published in October 2024; a Draft National AI Policy was gazetted in April 2026 but withdrawn after fabricated references were found, with a revised draft targeted for around January 2027. AI software is governed by the existing Copyright Act and case law.

Sources & authorities

  1. 1.Copyright Act 98 of 1978 — definitions, ss 2, 3(2), 5(2), 11B, 20, 21, 22, 25 (WIPO Lex consolidated text)
  2. 2.Copyright Amendment Act 125 of 1992 (the amendment that added computer programs and the computer-generated rule)
  3. 3.Haupt t/a Softcopy v Brewers Marketing Intelligence (Pty) Ltd [2006] ZASCA 40; 2006 (4) SA 458 (SCA)
  4. 4.Bergh and Others v Agricultural Research Council [2020] ZASCA 30; [2020] 2 All SA 637 (SCA)
  5. 5.King v SA Weather Service [2008] ZASCA 143; 2009 (3) SA 13 (SCA)
  6. 6.Thaler v Perlmutter, No. 23-5233 (D.C. Cir., 18 March 2025)
  7. 7.Copyright, Designs and Patents Act 1988 (UK) s 9(3) — computer-generated works
  8. 8.Directive 2009/24/EC (EU Software Directive) Arts 1(3), 2(1)
  9. 9.U.S. Copyright Office, Copyright and Artificial Intelligence, Part 2: Copyrightability (Jan 2025)
  10. 10.DCDT — Draft National AI Policy withdrawal (Reuters, 26 May 2026)
  11. 11.CIPC — Copyright information (eligibility & term of computer programs)

Every authority above was checked against its primary source in June 2026. This page is general information about South African law, not legal advice.

Get your AI & software IP in writing

We draft IP assignments, AI-aware development and SaaS agreements, and review the contracts you already have — so you actually own what you build.

Why you can trust this: Martin Kotze has been an admitted Attorney of the High Court of South Africa, registered Conveyancer, and Notary Public since 2014, practising from Pretoria. The firm is regulated by the Legal Practice Council under firm registration F17333.

This guide is general information, not legal advice for your specific matter.