Brazil’s AI bill copyright rules go further than the EU’s. PL 2338/2023, passed by the Federal Senate on 10 December 2024 and still before the Chamber of Deputies, would oblige AI developers to publish a summary of the protected content they use, let rights holders prohibit the use of their works, and require developers to pay for that use. The EU AI Act asks for the first two in a different form but contains no right to remuneration. This guide is for groups that develop, license or own content in both the EU and Brazil.
Key takeaways
- PL 2338/2023 is a bill, not law: the Senate approved it on 10 December 2024 and it has been with a special committee of the Chamber of Deputies since 2025. Any changes there send it back to the Senate.
- Both regimes require a public summary of training content, but only the Brazilian text adds a duty to remunerate rights holders for use in mining, training or development.
- Brazil’s text and data mining exception is narrow: research, educational and heritage institutions, non-commercial use only. Commercial training falls outside it.
- Remuneration would be owed only to holders domiciled in Brazil or in countries offering equivalent reciprocity, an open question for EU catalogues.
Where does PL 2338/2023 stand?
The bill was tabled in the Senate in 2023 and approved by the plenary in a symbolic vote on 10 December 2024, with Senator Eduardo Gomes as rapporteur, as DPL News reported on 11 December 2024. According to the Congress record of the bill, the Senate sent the approved text to the Chamber of Deputies on 17 March 2025. The Chamber assigned it to a special committee, where the Chamber’s bill record shows it awaiting the rapporteur’s report.
Copyright has been the sticking point. On 24 August 2026 the rapporteur in the Chamber, Deputy Aguinaldo Ribeiro, said the vote would take place after the October elections and that pending issues, copyright included, were being worked out with Senator Gomes, because the text will return to the Senate (Mobile Time, 24 August 2026). Everything below therefore describes the Senate text, which may still change.
What the Brazil AI bill says about copyright
The copyright rules sit in a section headed “Dos Direitos de Autor e Conexos”, Articles 62 to 66 of the text the Senate sent to the Chamber. For these purposes, “development” covers mining, training, retraining, testing, validation and application of AI systems.
- Transparency (Article 62): developers that use protected content must publish a summary of it on an easily accessible website, subject to trade secrets and to future regulations.
- Text and data mining exception (Article 63): automated use of protected content for research and AI development is not an infringement when carried out by scientific, research and educational institutions, museums, public archives and libraries, provided access was lawful, there is no commercial purpose and use is limited to what is necessary. Copies must be stored securely and only for as long as needed; the exception does not cover institutions linked to for-profit AI providers.
- Right to prohibit (Article 64): outside that exception, the rights holder may prohibit the use of its content in the development of AI systems. Prohibiting use after training does not exempt the AI agent from liability for damages.
- Remuneration (Article 65): an AI agent that uses protected content in mining, training or development must remunerate the holders, through collective management under the Brazilian Copyright Act (Law 9,610/1998) or direct negotiation, free or paid. The amount must be reasonable and proportionate, considering factors such as the size of the AI agent and the competitive effect of the outputs on the original works.
- Image and voice (Article 66): using images, audio, voice or video that identify natural persons must respect personality rights under the Civil Code.
The general sanctions regime would apply, with fines of up to BRL 50 million per infringement or, for private companies, up to 2% of the group’s gross turnover in Brazil (Article 50). The law would enter into force 730 days after publication, but the copyright section would apply after 180 days and Article 62 immediately (Article 80).
How does the EU AI Act deal with training data and copyright?
In the EU, the permission to train comes from copyright law, not from the AI Act. Article 4 of the DSM Directive (Directive (EU) 2019/790) allows reproductions of lawfully accessible works for text and data mining unless the rights holder has expressly reserved its rights in an appropriate manner, such as machine-readable means for content made publicly available online. Article 3 covers scientific research by research organisations and cultural heritage institutions, without an opt-out.
The AI Act then adds two obligations for providers of general-purpose AI models in Article 53(1)(c) and (d): a policy to comply with EU copyright law, including identifying and respecting reservations under Article 4(3), and a sufficiently detailed public summary of training content, using the template the Commission published on 24 July 2025. The Commission can fine providers up to 3% of worldwide turnover or EUR 15 million, whichever is higher (Article 101). There is no statutory remuneration for training: payment depends on licences.
Brazil vs EU: side-by-side comparison
| Issue | EU (DSM Directive + AI Act) | Brazil (PL 2338/2023, Senate text) |
|---|---|---|
| Legal status | In force; GPAI obligations apply since 2 August 2025 | Bill pending in the Chamber of Deputies |
| Who must be transparent | Providers of general-purpose AI models | Any AI developer that uses protected content |
| Form of transparency | Public summary on the Commission template | Summary on an accessible website, per future regulations |
| Research exception | Article 3: research organisations and heritage institutions | Article 63: research, educational and heritage institutions, non-commercial |
| Commercial training | Allowed for lawfully accessible content unless reserved | Holder may prohibit; if used, remuneration is due |
| How to opt out | Express reservation; machine-readable for online content | Right to prohibit; the text sets no technical format |
| Remuneration | None by law; contractual licences | Statutory duty, collective or direct negotiation |
| Foreign rights holders | Protection under general EU copyright rules | Paid only if domiciled in Brazil or with equivalent reciprocity |
What this means for your business
For EU-Brazil groups, the practical message is to build one compliance file that can serve both systems, while the Brazilian text settles.
- Map your training data now. A source-by-source record of what was used, how it was accessed and under which licence feeds the EU summary today and a Brazilian summary later.
- Review your opt-out practice. In the EU, reservations need to be machine-readable for online content; in Brazil, a written prohibition would also count. Use both.
- Check licences for Brazil. Content licensed in Europe for AI training may not cover a Brazilian remuneration claim; consider adding Brazil-specific terms.
- Rights holders in Spain, Portugal or elsewhere in the EU: check whether you have a Brazilian entity or collecting society mandate, because the reciprocity condition may decide whether you can be paid.
Our AI and digital assets team, covering copyright compliance for AI training, can prepare that dual file and the licence terms, and META Channel’s regulatory practice covers AI Act compliance within the same group.
Where companies get this wrong
- Treating the bill as final. Contracts drafted on the Senate text should include review clauses, because the Chamber may rewrite Articles 62 to 66.
- Assuming an EU opt-out works in Brazil. The legal basis is different, and a robots.txt line may not be read as a prohibition under Brazilian law.
- Ignoring domicile. Whether a European catalogue is owed remuneration may depend on reciprocity, which is untested.
- Splitting advice by country. A summary published for the EU that contradicts what is later disclosed in Brazil creates evidence for claimants in both.
Coordinating both regimes from one team keeps the data map, the licences and the public statements consistent; our licensing and IP disputes practice handles the contractual side.
Frequently asked questions
Is Brazil’s AI bill already in force?
No. PL 2338/2023 was approved by the Federal Senate on 10 December 2024 and sent to the Chamber of Deputies in March 2025, where a special committee is still examining it. If the Chamber amends it, the text returns to the Senate. Once published, the copyright section would apply after 180 days, except the transparency duty, which would apply immediately.
Would foreign rights holders be paid under the Brazilian bill?
Only in some cases. Article 65 limits remuneration to holders, Brazilian or foreign, domiciled in Brazil, and to people domiciled in countries that offer Brazilians equivalent reciprocity. Since EU law has no statutory remuneration for AI training, whether EU holders qualify is an open question that the final text or regulations may need to clarify.
Does the EU AI Act require AI companies to pay for training data?
No. The AI Act requires providers of general-purpose AI models to have a copyright compliance policy and to publish a training content summary, but it creates no right to remuneration. Rights holders who want payment must reserve their rights under Article 4 of the DSM Directive and negotiate licences.
Can IP Global Guard prepare our AI training compliance for the EU and Brazil?
Yes. We map your training data and licences, prepare rights reservations and licence terms for Europe, and coordinate qualified local correspondents in Brazil to follow the bill and adapt the documentation once it is approved. You work with a single point of contact and one strategy across the corridor.
How IP Global Guard can help with AI and copyright in Europe and Brazil
Whether you train models or own the content they learn from, the EU and Brazil are moving towards the same transparency but different economics. IP Global Guard, the IP services line of META Channel Corporation Limited, coordinates copyright, licensing and AI matters across more than 25 jurisdictions in Europe, Latin America and Africa, with one strategy and one billing relationship.
Send us a short description of your models or catalogue, the markets where you operate and the licences you already have. We will tell you where the EU and Brazilian rules bite and what to document now. Talk to our AI and copyright team.
This article is general information, not legal advice, and reflects the bill as approved by the Senate at the date of publication.
Sources
- DPL News, Brazilian Senate approves AI framework (11 December 2024)
- Congresso Nacional, PL 2338/2023 record (Senate approval 10 December 2024; sent to the Chamber 17 March 2025)
- Câmara dos Deputados, PL 2338/2023, text received from the Senate (17 March 2025)
- Câmara dos Deputados, PL 2338/2023 bill record (special committee)
- Mobile Time, AI framework vote expected at year end, says rapporteur (24 August 2026)
- Directive (EU) 2019/790, Article 4 (text as adopted, 17 April 2019)
- European Commission, AI Act Service Desk, Article 53 of Regulation (EU) 2024/1689
- European Commission, AI Act Service Desk, Article 101 of Regulation (EU) 2024/1689
- European Commission, Explanatory notice and template for the public summary of training content (24 July 2025)
