Regulating the Use of Copyrighted Works in Artificial Intelligence Training: A Comparative Analysis of Nigeria and the European Union

Authored By: Abdullateef Ayodeji Ogunmuyiwa

Univeristy Of Ilorin, Kwara State, Nigeria

Introduction

Artificial intelligence models now underpin decision-making across law, commerce, and creative industries. This shift has been made possible by training these models on vast datasets that routinely include copyrighted works, often without the authors’ knowledge or consent.

In the Nigerian context, the Copyright Act 2022^[1] was drafted without AI training in mind, leaving it unclear whether the Act’s open-ended fair dealing provision extends to cover this use.

Its open-ended fair dealing provision under Section 20(1)^[2] could, in principle, extend to cover such use, but the absence of an explicit text-and-data-mining exception, of the kind found in the EU’s DSM Directive^[3], leaves the position interpretively uncertain and judicially untested, unlike the EU’s more explicit and transparency-anchored framework under the DSM Directive and AI Act.^[4]

This article first examines Nigeria’s fair dealing exception under S.20(1), then the EU’s text-and-data mining exceptions under the DSM Directive and the AI Act’s transparency obligations, before comparing both frameworks to identify the reforms Nigeria’s copyright regime should adopt.

Background

Article 2(2) of the DSM Directive^[5] defines text and data mining as “any automated analytical technique aimed at analysing text and data in digital form in order to generate information which includes but is not limited to patterns, trends and correlations”.

AI training fits squarely within this definition, as it entails developers assembling large volumes of texts, images, or other data, and exposing a model to that data so it can identify statistical patterns, and, on that basis, adjust its internal parameters.

Before an AI model can identify such patterns, the underlying works must first be copied (downloaded, stored and processed) by the system conducting the training. This step is what brings AI within the ambit of copyright’s reproduction right, i.e the right that governs whether and how a work may be copied at all.

Both Nigeria’s Copyright Act 2022 and the EU’s legal framework anchor this right in their respective domestic terms (examined in the sections that follow), but the underlying trigger is the same in either system —a copy is made and copyright law is engaged.

The doctrine of traditional copyright assumes that reproduction leaves behind an identifiable and retrievable copy (such as a photocopy, or a saved file) that can be compared to the original to prove that its an actual copy. However, AI training complicates this assumption, as the process does not preserve the work in any retrievable form within the resulting model.

Whether copyright’s reproduction right which was built around identifiable copies can meaningfully apply to this non-retrievable form of copying is precisely the interpretive difficulty this article now turns to examine, first under Nigeria’s Copyright Act 2022, then under the EU’s more explicit framework.

Nigeria’s S. 20(1) Position

However, S. 20(1) of the Act^[7] removes this control for specified fair dealing purposes. In specifying such fair dealing purposes, the Act uses the language “such as…”, rendering the list non-exhaustive.

Among the purposes listed under S. 20(1) is non-commercial research and private study under paragraph (c)^[8]. This article however maintains that most consequential AI training, such as the kind done by OpenAI, Meta, and other comparable developers, is commercial activity, meaning such use may uneasily fit within the paragraph (c) non-commercial requirement.

A closer mechanical fit may be found under paragraph (o), which exempts transient and incidental copies that are integral and essential parts of a technological process; also, the reproduction’s sole purpose must be to enable transmission in a network between third parties, by an intermediary, or another lawful use with no independent economic significance^[9]. AI training pipelines do involve copying works as data moves through ingestion and processing systems, making this provision a more plausible analogy than paragraph (c).

Yet, the qualifiers (transient and incidental) sit uneasily with what AI training actually does, as the reproduction is not fleeting or secondary to the process, but rather foundational to it, since the resulting model’s parameters are shaped by that reproduction. Similarly, the requirement that reproduction serve network transmission between third parties is a description of ISP caching and network routing, not what happens when a model ingests and is shaped by training data.

Additionally, training data is often the single biggest driver of a model’s commercial value, which carries independent economic significance.

No Nigerian case law has tested fair dealing against AI training, leaving the legal position open by omission, rather than design.

The Position of the EU

The DSM Directive creates two separate text and data mining exceptions, distinguished chiefly by who may rely on them.

Article 3^[10] permits reproduction and extraction for text and data mining, but only by research organizations and cultural heritage institutions acting for scientific research purposes. It is unavailable to commercial developers.

Article 4^[11], by contrast, extends the exception to any person and any purpose, which means commercial AI training falls squarely within its scope. This broader availability however comes with a significant condition: under Article 4(3)^[12], rightholders may opt out by expressly reserving their rights in an appropriate manner, such as through machine-readable metadata or terms of service for content made available online.Where such reservation exists, the exception does not apply, and the reproduction requires the rightholder’s authorisation.

Article 4 therefore does not grant AI developers unconditional permission to train on copyrighted works, it grants conditional permission which is defeasiblev at the rightholder’s election.

The AI Act, under Article 53^[13], creates two obligations for providers of general purpose AI models aimed at promoting transparency.

First, a detailed summary of the content must be drawn up and made publicly available.

Second, providers must maintain a policy for complying with the EU’s copyright law, specifically to identify and honour rights reservations made under Article 4(3) of the DSM Directive.

The second obligation serves as an enforcement backbone of Article 4(3) of the DSM Directive by creating a disclosure duty that allows rightholders to find out if their choices are honoured.

Comparative Analysis of the Two Jurisdiction

Since this article has established the respective positions of the two jurisdictions, this section intends to test each position against the other, to see whether authors are given actionable protection against unauthorized use of their work in AI training.

Both frameworks exclude commercial use from their research-specific exceptions, i.e S. 20(1)(c) of Nigeria’s Copyright Act^[14] and Article 3 of the DSM Directive^[15] However, the EU legislates further under Article 4 of the DSM Directive^[16] to create a broader exception that covers general commercial use, with a limitation that allows rightholders to opt out by restricting the use of their work under Article 4(3^[17]). Nigeria has no such further legislation to cover general commercial use, thereby implying that Nigerian rightholders retain full control over their works by default.
Article 53 of the AI Act^[18] under the EU’s framework creates a disclosure duty that demands that providers create a summary and make it publicly available to show how copyrighted works have been used, which allows authors to infer whether their works have been used. Under the Nigerian framework, however, no such disclosure duty exists. Thus, a gap exists in the Nigerian framework that has no counterpart under the EU’s regime.
On the EU side, the AI office is allowed necessary enforcement mechanisms, holding investigative powers, as well as sanctioning powers(to impose penalties where it establishes intentional or negligent breach), and powers to conduct inspections of providers’ premises^[19]. Also, non-compliance carries fines of up to fifteen million euros or 3% of global turnover, whichever is higher^[20]. On the Nigeria side, on the other hand, the NCC’s powers are substantial and arguably more aggressive than the EU’s enforcement mechanisms, in the sense that the commission is empowered to block or disable access to any content, link, or website, if it reasonably suspects it of infringing copyright^[21], and is also empowered under S. 78 of the Copyright Act 2022 to investigate, redress, and settle cases of copyright infringement directly, without waiting for rightholders to file the suit in court^[22]. A gap nonetheless remains between the two regimes, though not one rooted in institutional weakness on Nigeria’s part.
The NCC’s enforcement powers presuppose that a rightholder can identify an infringement to report in the first place. Yet, since the Copyright Act 2022 lacks a disclosure duty, as seen under Article 53 of the AI Act, the NCC’s capable enforcement apparatus has nothing to act on in this specific context.

As this article has now shown, both frameworks exclude non-research commercial use from their narrow exceptions.

However, the EU’s framework went further to create a path past the exclusion that allows commercial use generally, with a reservation for rightholders, and a duty to disclose the usage, which makes it possible for authors to infer whether their work has been used^[23].

It is, however, worth noting that the EU’s opt out mechanism has drawn criticism, with opinions holding that this might affect competitiveness of developers in other jurisdictions (such as the United States Of America), which allow more unrestricted access to training data for AI models^[24].

Yet, even though the EU’s balance is a debatable policy choice left open for argument and reform, there is no legislation to begin with under the Nigerian framework.

Recommendations

This article has shown that a significant gap exists in the Nigerian framework for which the EU’s framework has indicated a step in the direction of reform, which Nigeria can draw from and adapt.

The first recommendation is that Nigeria’s regime needs to adopt an explicit, albeit conditional, exception permitting text-and-data-mining (AI training) for any purpose, including commercial use. This would spare courts from having to determine, through strained interpretation, whether paragraphs ( c) or (o) can be read to cover AI training, or whether the non-exhaustive ‘such as’ language permits recognising it as an unlisted purpose in its own right.
The next recommendation is that any new exception that is created be paired with an opt-out mechanism that rightholders can trigger to disallow use of their work. However, it is also important to take cognizance of the debates surrounding the counterpart of such switch found under the EU’s regime^[25], particularly about being cautious of leaving developers of AI models dependent on rightholders’ independent choices, disadvantaging AI development in that part of the world in comparison to AI development in more tolerant jurisdictions of the world, such as the USA.
Another important recommendation is that Nigeria would need to create a disclosure obligation modeled after Article 53 of the EU’s AI Ac^[26]t, which requires developers of AI models to publish a summary of the content used to train the AI model. This would serve as a transparency measure, and at the same time create grounds for the enforcement mechanisms of the NCC to act on. Thus, a disclosure duty will not require the building of institutional capacity, but will simply give the NCC’s existing machinery grounds to act on.
Although legislative reform has been identified as a necessity, a full reform may take years, as it requires National Assembly action. In the meantime, this article recommends that the NCC exercise its existing powers to make interpretative regulations necessary to give effect to the purposes of Part II of the Act (Exceptions to Copyright), under S. 27(6) of the Act^[27], that will clarify how fair dealing applies to AI training.

These recommendations are not independent, but rather they function as a single package. The interim exercise of the NCC’s power under S. 27(6) to create regulations will buy time while legislative reform is pursued. The legislative exception, in turn, can only function safely if paired with both the opt-out mechanism and the disclosure duty, which will create a situation where the exception grants permission, the opt-out mechanism preserves rightholder control over the permission, and the disclosure duty makes the opt-out enforceable in practice.

Adopting any one reform in isolation would leave at least one of the three gaps this article has identified (permission, control, and enforceability) unresolved.)

Conclusion

This article has shown that Nigeria’s Copyright Act 2022 does not clearly regulate the use of copyrighted works in AI training because it has not legislated on the question at all. Neither paragraph ( c), which excludes commercial use, nor paragraph (o), whose ‘transient and incidental’ language sits uneasily with the foundational nature of training, was drafted with this use in mind, and no Nigerian court has yet been asked to determine whether either extends to cover it.^[28]

The EU, by contrast, has built an explicit architecture. Article 4 of the DSM Directive grants conditional permission, Article 4(3) preserves rightholder control through an opt-out mechanism, and Article 53 of the AI Act makes that control enforceable through mandatory disclosure.^[29]

The difference between the two jurisdictions can be seen to not be a difference in values, or in how each balances authors against developers, but a difference between a deliberate policy choice and the absence of one.

To close the gaps that have been identified, the article has recommended a three-part legislative reform (an explicit text-and-data-mining exception, a rights-reservation mechanism, and a disclosure duty), alongside an interim measure available to the Nigerian Copyright Commission under S. 27(6) of the Act^[30], which permits regulatory clarification, while full legislative reform is pursued.

The stakes of leaving this gap unresolved extend beyond doctrinal tidiness. Nigerian authors currently have no reliable way to know whether their works have been used to train AI models by developers operating within Nigeria, or those operating from jurisdictions far beyond its reach. Developers, in turn, operate under the same uncertainty, with no clear rule to design around, and no assurance that a court would not later find their activities infringing.

a framework that resolves this ambiguity through legislative or administrative reforms (as proposed by this article), or through judicial clarification in the interim, would serve both groups better than the silence that currently governs them both.

Table of Legislation

Nigeria

European Union

Directive (EU) 2019/790 of the European Parliament and of Council of 17 on copyright and related rights in the Digital Single Market, and amending Directives 96/9/EC and 201/29/EC [2019] OJ L130/92

Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence [2024] OJ L2024/1689

Bibliography

Journal Articles

Ziaja GM, ‘The Text and Data Mining Opt-Out in Article 4(3) CDSMD: Adequate Veto Right for Rightholders or a Suffocating Blanket for European Artificial Intelligence Innovations?’ (2024) 19 Journal of Intellectual Property Law & Practice 453

Secondary/Online Sources

European Commission, ‘The Enforcement Framework of the AI Act’ (Shaping Europe’s Digital Future) https://digital-strategy.ec.europa.eu/en/policies/enforcement-ai-act accessed 23 August 2026

^[1] Copyright Act 2022 (Nigeria).

^[2] ibid, s 20(1).

^[3] Directive (EU) 2019/790 of the European Parliament and of the Council of 17 April 2019 on copyright and related rights in the Digital Single Market and amending Directives 96/9/EC and 2001/29/EC [2019] OJ L130/92 (‘DSM Directive’).

^[4] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence [2024] OJ L, 2024/1689 (‘AI Act’).

^[5] DSM Directive (n 3), art 2(2).

^[7] ibid, s 20(1).

^[8] ibid, s 20(1)(c).

^[9] ibid, s 20(1)(o).

^[10] DSM Directive (n 3), art 3.

^[11] ibid, art 4.

^[12] ibid, art 4(3).

^[13] AI Act (n 4), art 53.

^[15] DSM Directive (n 3), art 3.

^[16] ibid, art 4.

^[17] ibid, art 4(3).

^[18] AI Act (n 4), art 53.

^[19] ibid, art 88; European Commission, ‘The Enforcement Framework of the AI Act’ (Shaping Europe’s Digital Future) https://digital-strategy.ec.europa.eu/en/policies/enforcement-ai-act accessed 21 August 2026.

^[20] AI Act (n 4), art 101.

^[22] ibid, s 78.

^[23] AI Act (n 4), art 53.

^[24] Gina Maria Ziaja, ‘The Text and Data Mining Opt-Out in Article 4(3) CDSMD: Adequate Veto Right for Rightholders or a Suffocating Blanket for European Artificial Intelligence Innovations?’ (2024) 19 Journal of Intellectual Property Law & Practice 453.

^[25] DSM Directive (n 3), art 4(3).

^[26] AI Act (n 4), art 53.

^[29] DSM Directive (n 3), arts 4, 4(3); AI Act (n 4), art 53.