Why South Africa’s Constitutional Court upheld fair use but struck down the education exceptions

Introduction

South Africa’s copyright reform has been stuck in Parliament for so long that the two bills at its centre now have a history of their own. The Copyright Amendment Bill was introduced in May 2017, passed by both houses in 2018 and 2019, and sent back by the President in 2020 (Ncube and Samtani, 2023). Parliament passed revised versions of it and of the Performers’ Protection Amendment Bill on 29 February 2024 (Oriakhogba and Erhagbe, 2024). Then the Constitutional Court got involved twice. On 7 May 2025 it read a disability exception into the existing Act in Blind SA v President of the Republic of South Africa ([2025] ZACC 9), and on 26 June 2026 it ruled on the President’s own referral of both bills in Ex parte President of the Republic of South Africa ([2026] ZACC 26).

This paper argues that the two rulings, read together, show a fairly consistent line, even if the Court never states it in one place. Exceptions to copyright seem to survive when they are tied to recognisable public-benefit purposes and fenced in by fairness factors or by narrow conditions. They fail when they let institutions copy at scale and push the cost of a public good, here education, onto individual authors. That line cuts across the academic camps. Forere (2025) argued that fair use was unconstitutional, while Beiter et al. (2022) defended both fair use and the education exceptions. The Court sided with the second group on fair use and with the first on the biggest education exception. The paper sets out the two decisions, looks at why fair use held, then at why section 12D(1) to (5) fell, and finally at what the Court left open, namely the royalty provisions and artificial intelligence.

Two rulings that asked different questions

Blind SA began as a plain access case. In Blind SA v Minister of Trade, Industry and Competition ([2022] ZACC 33) the Court held that sections 6 and 7 of the Copyright Act 98 of 1978, read with section 23, unfairly discriminated against people with visual and print disabilities, because every conversion into an accessible format needed the owner’s permission. It read in a temporary section 13A and suspended the declaration of invalidity for 24 months. Ncube and Samtani (2023) point out that this was the first copyright case the Court had decided in its 27-year history, and the first anywhere to treat a missing accessible-format exception as disability discrimination. Mupangavanhu (2024) reads it as a case about the equality, dignity, education and expression rights that copyright was getting in the way of, which is why she wants copyright law to lean towards users.

Parliament did not finish in time. The suspension ran out on 21 September 2024 and the reading-in lapsed, which left blind and print-disabled readers with no protection at all. Blind SA went back to the Court in October 2024, and on 14 October the President referred the two bills to the Court as well ([2025] ZACC 9). In its decision of 7 May 2025 the Court read in a section 19D based on the Bill’s own clause 19D, to apply until remedial legislation comes into force. That is a notable shift from 2022, when the Court found that the Bill’s version needed regulations and reached beyond the pleaded case, and so wrote its own narrower section 13A (Ncube and Samtani, 2023). The difference was that by 2025 Parliament had revised the clause and the President’s reservations did not touch it. The Court was, in effect, borrowing a text Parliament had already settled on, which looks like deference more than boldness.

The referral decision of 26 June 2026 was a different kind of exercise altogether. It was abstract review of bills that were not yet law, heard on 21 and 22 May 2025, with eight justices in the majority under Mhlantla J and Majiedt J dissenting on one provision ([2026] ZACC 26). The order declared the President’s referral incompetent as to sections 6A, 7A and 8A, upheld sections 12A, 12B, 12C and 19B and 12D(6) to (9), found section 19C neither vague nor an arbitrary deprivation, and struck down section 12D(1) to (5) as an arbitrary deprivation of property under section 25(1) of the Constitution. The same findings were extended to the matching provisions of the Performers’ Protection Amendment Bill. The upheld provisions were all declared constitutional only within the scope of the referral, and that qualifier matters later in this paper. For context, the reservations the President raised in 2020, as Beiter et al. (2022) list them, covered tagging, royalties, public participation on fair use, the library and education exceptions and international law, so the 2026 order answers a good part of that list but, as we will see, not the royalty part.

Fair use survives, but on a short leash

Fair use was the part of the Bill most people expected to fall. The publishing and creative sectors had fought it for years, and Tomaselli (2022) records how the publishing industry produced hundreds of pages of economic impact studies against it, while the pro-Bill lobby leaned on rhetoric in the popular media. Tomaselli and Pieterse (2023) go further and say that the state and large technology companies pushed an American-style exception at the expense of content creators. On the legal side, Forere (2025) used two WTO panel decisions to argue that the clause fails the Berne and TRIPS three-step test and amounts to an arbitrary deprivation of the reproduction and adaptation rights. Her view was that it was unconstitutional because copyright could be taken without a law of general application and without compensation.

The Court did not accept that, and it is worth being exact about how it did not. It treated copyright as property for section 25(1), accepted that an exception which cuts back the owner’s exclusive rights is a deprivation, and then asked whether the deprivation was arbitrary ([2026] ZACC 26). The majority found it was not. Sections 12A to 12D, on its reasoning, pursue public-interest goals like education and expression and qualify copyright rather than extinguishing it. It also treated fair use and fair dealing as sitting on a continuum, and said the open-ended wording was not fatal vagueness. Notice what the order does not say. It speaks of arbitrary deprivation under section 25(1) and nothing about expropriation or compensation, so Forere’s (2025) section 25(2) argument seems to have been met by the way the question was framed, not by a direct rebuttal. On the treaty point the majority said the question was constitutional compliance, not whether the Bill matches international law, though it kept treaty obligations in view.

What makes this less of a victory for fair use than it first looks is the way the Court read the clause down. The words “such as” do not allow any use at all; unlisted purposes must share the public-benefit or private-use character of the listed ones. The substitution-effect factor operates as a real constraint, and “personal use” covers the individual doing the copying, not copying for friends ([2026] ZACC 26). That is close to what Beiter et al. (2022) already said, since they noted that “such as” appears in about a dozen other jurisdictions and that the fairness factors mostly track existing South African case law. It also answers some of Miggels (2025), who argued that fair dealing is hardly the certain regime its defenders claim, given that its fairness test has been worked through in only one South African case, Moneyweb. In effect the Court has given South African fair use a narrower shape than the American original. Whether publishers find that comforting is another matter, and Tomaselli and Pieterse (2023) would probably say they have seen narrow readings erode before.

There is also the abstract-review point. The Court warned against striking down provisions on educated guesses about how they might work in practice and said it would act only on reasonably foreseeable unconstitutionality ([2026] ZACC 26). That sits awkwardly with an evidence-heavy campaign like the one Tomaselli (2022) describes, because a thick pile of economic studies counts for little when the Court is reviewing text and not outcomes. It also means the fair use provisions are cleared only for now. A rights holder who can show real harm after enactment could still bring a fresh challenge, and Forere’s three-step arguments would then have a factual footing they lack today.

Why section 12D(1) to (5) fell

The education exceptions were where the Court drew blood. According to the Court’s own media summary, the majority held that subsections 12D(1) to (5) went further than their purpose required and sat badly with international copyright standards (Constitutional Court of South Africa, 2026). Commentary on the ruling reports two further strands in the reasoning, that key terms such as commercial purpose were loosely defined and that the provisions moved the cost of public education from the state to authors and publishers, which is basically the worry Tomaselli and Pieterse (2023) had raised about the state leaning on creators to deliver access. Majiedt J, on the media summary’s account, thought the majority gave too little weight to education rights and to the treaty provision on educational use, and called a facial challenge to an unenacted bill too speculative.

The scholarly record shows how finely balanced this was. Beiter et al. (2022) thought section 12D was constitutional and a legitimate policy choice. Institutions had to try for a licence first, whole-book copying was confined to limited circumstances such as an out-of-print work, and they linked the whole thing to apartheid-era practice, when books were copied because censorship or boycotts kept them out of reach. On the other side, Tomaselli and Pieterse (2023) warned that the Bill’s contradictions would hurt educational publishing at every level, and Forere (2025) said textbook sales are exactly the sort of normal exploitation that an open exception threatens. So the safeguards that Beiter and colleagues relied on, such as the licence requirement, were not enough for the majority. That suggests the Court was less interested in the existence of a safeguard than in who carried the burden if it failed.

There is a real tension with Blind SA, and it deserves an honest look. In 2022 the Court said that those serving print-disabled readers should have the greatest latitude to produce accessible copies (Ncube and Samtani, 2023). Education rights were in play there too, since section 29(1) was one of the rights found to be infringed (Mupangavanhu, 2024). So why did one education-linked exception pass and the other fail? The best answer is probably market failure. The disability exception filled a gap where no functioning market existed, since authorisation was often refused or simply unobtainable, and it came with conditions: lawful access, non-profit supply, respect for the integrity of the work (Ncube and Samtani, 2023). Section 12D(1) to (5) would have operated where a licensing market does exist, for textbooks, and with looser conditions. That reading is an inference from the order and the reports, not something the Court says in so many words, and the dissent shows it is contestable.

Still, the distinction holds up better than the scholarly noise around it suggests. A country with South Africa’s inequality has good reasons to want cheaper access to learning materials, and nothing in the ruling denies that. What the Court insists on is that the answer cannot be an open-ended licence to copy, paid for by authors, particularly when the text is vague about who qualifies. Beiter et al. (2022) are right that apartheid-era copying grew out of real scarcity, but the state now has other levers, funding among them. That would explain why the 12D(6) to (9) provisions, which deal with smaller uses, could survive while the broad ones could not. It will frustrate the education sector, and fairly so, though the ruling does not turn against education rights themselves, and Ncube and Samtani (2023) show how much space the Court is prepared to give access where the conditions are tight.

What the Court left open

The first gap is remuneration. The Court refused to decide sections 6A, 7A and 8A at all, holding the referral incompetent because the President’s reservations to Parliament concerned retrospective subsections that Parliament later deleted, while the wider retrospectivity argument surfaced first in court ([2026] ZACC 26). So the fair royalty rules for authors, visual artists and performers are untested. Oriakhogba and Erhagbe (2024) note that the argument over exceptions overshadowed the Bill’s equitable remuneration, reversionary rights, orphan works licensing, resale royalty and collective management rules. Beiter et al. (2022) had already said that retrospectivity does not make the royalty provisions unconstitutional by itself, and that the real danger is arbitrariness, because sections 7A and 8A could force the renegotiation of licences that were perfectly fair.

That matters because creators were the intended beneficiaries of the whole exercise. Forere (2023) starts from the fact that well-known South African musicians keep dying poor, and she draws on industry reports showing global physical music sales falling from 22.3 billion dollars in 1999 to 4.6 billion in 2022 as streaming took over. She also argues that broad exceptions without a private copying levy could reduce artists’ income further. Fair use being constitutional does not answer her worry, and neither does the striking down of section 12D. If the royalty provisions are later challenged and fall, the Bill would have taken from creators with one hand and given little back with the other. That is why Oriakhogba and Erhagbe (2024) want the remuneration clauses examined as closely as the exceptions have been.

The second gap is artificial intelligence. Nothing in the order touches it. Okorie (2025) argues that researchers at public institutions should generally not need a licence, or be treated as infringers, when they use protected material for AI and machine learning research, and she tests that against both the existing fair dealing rule and the proposed fair use clause. Miggels (2025) says fair dealing does not clearly permit AI training unless the system is built for research, and that fair use with guardrails and remuneration for the works used could serve creators and the public better. Beiter et al. (2022) had suggested adding computational analysis to the research purpose in section 12A, which implies the text then on the table did not mention it.

The Court’s own narrowing of fair use makes the AI question harder, not easier. Commercial-scale training that produces outputs competing with the source works would struggle on the substitution factor, and it is neither personal use nor one of the named educational or research purposes, which echoes Miggels’s (2025) point that training sits uneasily inside fair dealing unless the system is built for research. Okorie (2025) leaves more room for non-commercial research use. That is inference. The Court did not decide it, and the Bill could still be amended. But it does mean that the constitutional green light for section 12A is not a green light for AI developers, and it means Parliament, which now has the Bills back, has a choice between dealing with the issue in the text and leaving it to litigation. At the time of writing in October 2026 no amended Bill appears to have been enacted.

Conclusion

The Constitutional Court’s two decisions suggest a rule of thumb that is easier to state than to apply. Exceptions to copyright pass when they have a defined purpose, built-in fairness limits and a narrow footprint, as with section 19D, the narrowed section 12A and the smaller uses in section 12D(6) to (9). They fail when they spread across institutions with loose definitions and leave authors to carry the cost of public education. On that line the academic predictions were each half right, with Forere (2025) closer on section 12D and Beiter et al. (2022) closer on section 12A, which is a fair reminder of how hard abstract review is to forecast.

The unresolved parts are where the real work now sits. Parliament has to repair section 12D, and a licensing-first structure that Beiter et al. (2022) thought sufficient was not enough, so a better route may be tighter definitions of who may copy and what counts as commercial, with funding from the state behind it, the kind of conditions that made the disability exception workable (Ncube and Samtani, 2023). It still has to face the royalty rules, which the Court never judged. And nothing in the order or in the sources discussed here shows the Bill dealing with AI directly (Okorie, 2025). The Court got the line roughly right, in this paper’s view, but its caution leaves Parliament with more to do than the headlines about fair use being upheld would suggest, and creators and educators are both still waiting (Forere, 2023).

Reference list

Beiter, K.D., Fiil-Flynn, S.M., Forere, M., Klaaren, J., Ncube, C.B., Nwauche, E.S., Rens, A., Samtani, S. and Schonwetter, T. (2022) Copyright reform in South Africa: two joint academic opinions on the Copyright Amendment Bill [B13B-2017]. Potchefstroom Electronic Law Journal, 25(1), pp. 1-45. doi:https://doi.org/10.17159/1727-3781/2022/v25i0a13880

Forere, M. (2023) The South African music landscape and the Copyright Amendment Bill. Journal of Legal Anthropology, 7(2), pp. 68-86. doi:https://doi.org/10.3167/jla.2023.070204

Forere, M. (2025) The compliance of the fair use clause in the South African copyright amendment bill with the three-step test and the Constitution of South Africa. Journal of Intellectual Property Law & Practice, 20(7), pp. 447-457. doi:https://doi.org/10.1093/jiplp/jpaf031

Miggels, A. (2025) The implementation of the fair use exception into South African copyright law in the age of artificial intelligence: a fever dream or a justified reform? South African Intellectual Property Law Journal, 13(2), pp. 45-72. doi:https://doi.org/10.47348/SAIPL/v13/i2a3

Mupangavanhu, Y. (2024) Copyright law and human rights: Blind SA v Minister of Trade, Industry and Competition and its potentially far-reaching implications. South African Journal on Human Rights, 40(3-4), pp. 313-326. doi:https://doi.org/10.1080/02587203.2024.2437025

Ncube, C.B. and Samtani, S. (2023) Copyright, disability rights, and the Constitution: Blind SA v Minister for Trade, Industry and Competition. Constitutional Court Review, 13(1), pp. 471-503. doi:https://doi.org/10.2989/CCR.2023.0016

Okorie, C.I. (2025) The fair dealing/fair use landscape for artificial intelligence innovation and computational research in Africa. International Review of Law, Computers & Technology, pp. 1-28. doi:https://doi.org/10.1080/13600869.2025.2557640

Oriakhogba, D.O. and Erhagbe, E.O. (2024) The Copyright Amendment Bill: a new vista for fair remuneration for South African creators and performers? GRUR International, 73(10), pp. 959-966. doi:https://doi.org/10.1093/grurint/ikae080

Tomaselli, K. (2022) The 2022 Copyright Amendment Bill: implications for the South African universities’ research economy. Communicare: Journal for Communication Studies in Africa, 41(2), pp. 14-33. doi:https://doi.org/10.36615/jcsa.v41i2.1479

Tomaselli, K. and Pieterse, H. (2023) Copyright Amendment Bill: contradictions to hit the South African education sector. Education as Change, 27, article 14325. doi:https://doi.org/10.25159/1947-9417/14325

Legislation and case law

Blind SA v Minister of Trade, Industry and Competition [2022] ZACC 33 (CCT 320/21), Constitutional Court of South Africa, 21 September 2022. Available at: https://www.saflii.org/za/cases/ZACC/2022/33.html (Accessed: 18 October 2026).

Blind SA v President of the Republic of South Africa [2025] ZACC 9 (CCT 300/24), Constitutional Court of South Africa, 7 May 2025. Available at: https://www.saflii.org/za/cases/ZACC/2025/9.pdf (Accessed: 18 October 2026).

Constitution of the Republic of South Africa, 1996. Available at: https://www.gov.za/documents/constitution-republic-south-africa-1996 (Accessed: 18 October 2026).

Constitutional Court of South Africa (2026) Media summary: Ex parte President of the Republic of South Africa, In re Constitutionality of the Copyright Amendment Bill and the Performers’ Protection Amendment Bill (CCT 306/24). Available at: https://www.concourt.org.za/index.php/judgement/647-ex-parte-president-of-the-republic-of-south-africa-in-re-constitutionality-of-the-copyright-amendment-bill-and-the-performers-protection-amendment-bill-cct306-24 (Accessed: 18 October 2026).

Copyright Act 98 of 1978. Available at: https://lawlibrary.org.za/akn/za/act/1978/98 (Accessed: 18 October 2026).

Ex parte President of the Republic of South Africa: In re Constitutionality of the Copyright Amendment Bill and the Performers’ Protection Amendment Bill [2026] ZACC 26 (CCT 306/24), Constitutional Court of South Africa, 26 June 2026. Available at: https://www.saflii.org/za/cases/ZACC/2026/26.html (Accessed: 18 October 2026).