Introduction: Brazilian Copyright Framework and the existing L&Es
Brazil has signed most Intellectual Property (IP) law Treaties, but notably it has not adhered to the WIPO Copyright or the WIPO Phonogram and Performance Treaties. The main national legal instrument for copyright is the Brazilian Copyright Law (Law n. 9.610/98), which is complemented by the software protection legislation.
Statutory limitations and exceptions (L&E) for copyright in Brazil are notoriously precarious. However, under the Constitutional framework, the courts have come to regard their interpretation as extensive and not limited to the situations set forth in the copyright law. As it stands now, they must be considered as a flexible set (see here). Such judicial understanding is summarized by Statement 115 set by the Federal Justice Council: “The limitations of copyright established in articles 46, 47 and 48 of the Copyright Law must be interpreted extensively, in accordance with the fundamental rights and the social function of the property established in article 5, XXIII, of CF/88”.
Despite the flexible character of L&Es in Brazil, none of them directly addresses uses for the purpose of research or text and data mining (TDM). Nonetheless, policy discussions (both within the legislature and the executive) have started to address the issue, particularly in relation to innovation and the training of Artificial Intelligence (AI) systems.
IP, Innovation and AI national strategies
In Brazil, actions and objectives for technological development, especially the intersection between data-intensive technologies and IP, can be found in various policy documents, usually referred to as “strategies”, three of which are of relevance here: the Brazilian Artificial Intelligence Strategy (EBIA) (English summary here), the National Strategy for Intellectual Property (ENPI) and the National Strategy for Innovation (ENI).
Supported by WIPO, in 2020 the Brazilian Federal Government carried out a review of the National System of Intellectual Property (SNPI), which concluded, among other things, that the SNPI is not fit to promote development and foster creation and innovation, one of the causes being the “imbalances in the uses of the IP system related to underuse and overuse of IP rights” (ENPI 38-39). In order to overcome this, the National Strategy for Intellectual Property (ENPI) proposes guidelines, goals, objectives and actions divided into strategic thematic axes (ENPI 38).
Even though there is no direct recommendation to include TDM-related L&Es, some are related to its practices, for example (authors’ own translation):
- “Automate the extraction of statistical data from the Offices’ databases;”
- “Improve, update and prepare a Bill at the initiative of the Executive Branch, for the reform of the… Copyright Law, mainly considering the new technologies and business models on the Internet and the responsibility of internet service providers in relation to violations of [IP] rights;”
- “Improve regulation related to [IP] Rights of emerging sectors including, but not limited to: internet plus, e-commerce and big data;”
- “Support the Government’s initiatives to create a favorable environment for innovation, and, based on the knowledge of the need for the development of new technologies, promote prospection, technological monitoring and induce the generation of IP assets”.
Focused on AI, the Brazilian Artificial Intelligence Strategy (EBIA – English summary here) is the first national effort to follow international trends towards developing a regulatory framework for AI. It only briefly mentions the importance of a TDM exception, so it misses the chance to address IP-related concerns, especially those present in the ENPI.
When it comes to public policies, it is also worth mentioning the most recent National Policy on Innovation (PNI) formalized in Decree n. 10.534/2020. Within its objectives, there is the encouragement of “research, development and innovation by companies, ICT and non-profit private entities, with the aim of increasing the productivity and competitiveness of the economy, generating wealth and social well-being” (art. 6, I). As seen in the EBIA and ENPI, the PNI is also divided into different thematic axes, which relate to education, funding, markets, IP and a culture of innovation (art. 5).
One of the instruments of the PNI is the National Strategy for Innovation (ENI), which provides more specific initiatives. Here, again, there is no specific mention of TDM or the need to (re) consider IP rights in light of AI and data-intensive technologies. The Annex to the PNI brings some guidelines that were designed to promote strategic actions listed in the ENI, and, despite not providing any specific comment on L&Es and/or TDM, it reinforces the need for it.
The legislative debate
The House of Representatives, in an effort to regulate AI, discussed and approved Bill 21/2020 on September 29, 2021. Under art. 5, the Bill institutes some guidelines for a TDM exception, which reads as follows:
The following are principles for the development and application of artificial intelligence in Brazil: (…) VIII – availability of data: non-infringement of copyright by the use of data, databases and texts protected by it, for the purpose of training artificial intelligence systems, provided that the normal exploitation of the work by its owner is not impacted.
Although the provision lacks precision and clarity, it is notable as the first legislative initiative stating the need for a TDM exception to copyright. Following its approval it was sent to the Senate for discussion and to complete the legislative process. Once in the Senate, it met two other proposed Bills, and in order to agglutinate, restructure and improve the content and text, a specific commission was created.
The public hearings carried out by this Commission addressed several issues, including TDM and copyright.
The Commission was open to contributions and held public hearings (see the Brazilian Copyright Institution presentation and Annex), some of which were specifically on the topic of TDM. Research institutions and civil society organizations mostly supported the existence of L&Es for TDM, especially for research and innovation purposes, recognizing them as a tool for promoting development, research and innovation. Traditional copyright industries, with a long history of opposing L&Es, claimed without evidence that it would economically harm them and authors alike, as well as that it would be in violation of the three-step-test. Tech companies, on the other hand, supported its adoption in order to provide more clarity to organizations working in the development of AI technologies. Parts of the Government contended it should be regulated within the IP legal framework and not in the AI Bill and that the approach should be more principle-oriented (all statements available here).
The Final Report came out on December 6, 2022. It contains an amended text, justifications and also a summary of the contributions made by different stakeholders (Private Sector, Government, Academia, and Civil Society). When it comes to TDM, the new text is more precise, reasonable and functional, and reads as follows:
Art. 4. For the purposes of this Law, the following definitions are adopted:
[…] VIII – text and data mining: process of extracting and analyzing large amounts of data or partial or full excerpts of textual content, from which patterns and correlations are extracted that will generate relevant information for the development or use of artificial intelligence systems.
Art. 42. The automated use of works, such as extraction, reproduction, storage and transformation, in data and text mining processes in artificial intelligence systems, in activities carried out by research and journalism organizations and institutions, and by museums, archives and libraries, is not copyright infringement provided that the use:
I – does not have the objective of simply reproducing, displaying or disseminating the original work itself;
II – takes place to the extent necessary for the purpose to be achieved;
III – does not unjustifiably harm the titleholders’ economic interests; and
IV – does not compete with the normal exploitation of the works.
- 1 Any reproductions of works for the data mining activity will be kept under strict security conditions, and only for the time necessary to carry out the activity or for the specific purpose of verifying the results of the scientific research.
- 2 The provisions of the caput apply to data and text mining activities for other analytical activities in artificial intelligence systems, subject to the conditions set out in the caput and paragraph 1, provided that the activities do not communicate the work to the public and that access to the works was given legitimately.
- 3 The text and data mining activity involving personal data will be subject to the provisions of… [the] General Law for the Protection of Personal Data.
As for the TDM text itself, some innovation-friendly wording stands out, especially when compared to the modern TDM L&Es. One example is the scope of uses allowed under this provision, which can be seen as comprehensive, since it admits the “use of works, such as…”. Another positive aspect in the proposed clause is that it is not limited to research but directs its use for the TDM automated processes in AI systems. Also, it allows for such activities when carried out by research and journalist institutions, museums, archives and libraries.
On the other hand, the clause is not as open and flexible as it could be, since the four factors listed in the provision seem to be cumulative. Finally, when providing in paragraph two for uses other than those specified by the caput (commercial organizations, for example), the restriction to only using works ‘legitimately’ accessed can be subject to further discussion on the potential outcomes of imposing such restrictions.
In the last few years, there has been a growing body of legal regulation of TDM (see here). Since 2018, Japan, the EU, Singapore and others have promoted changes to their copyright law and included specific limitations for TDM. In the Global South and the developing world, such changes have been slow, although urgently needed (see here). In Brazil, as we have seen above, the many ‘strategies’ developed under the prior Government are nothing more than general guidelines for public policies but, nonetheless, they add consistency and direction to the political and public debate and influence the construction and approval of effective norms.
In Brazil, TDM came to be included in the Legislative Bill on AI as one specific topic. After the approval in the House, the Senate-formed Commission gathered views from multiple stakeholders, and the final report is a real advance to the status quo. The Brazilian Copyright Institute and other organizations and scholars contributed extensively to the debate and as far as we can see, there is a high possibility that a TDM limitation will get voted and approved in the Senate. But, above all, it is certain that the case has been laid out, and it encountered a fertile political and social environment.
If and when this Bill passes, Brazil would finally join the growing number of countries adjusting their laws to foster innovation, provide a safer environment for data-intensive projects and businesses and, in our case, fulfil the Constitutional mandate of balancing IP rights with other relevant social interests, such as innovation.