Is AI-Generated Content Copyrightable in Brazil? What to Do

Is AI-Generated Content Copyrightable in Brazil? What to Do

Brazil's Copyright Law (Law 9.610) has no specific rule for works generated by AI. It treats the author as the individual who creates the work, so text or an image produced autonomously by a machine fits poorly within classic protection. Material with a relevant human creative contribution has stronger arguments. In practice, a company should assume that AI-generated content may not be exclusive to it and should check the risks before publishing.

Key takeaways

• Brazilian law defines the author as the individual who creates the work. It does not name AI as an author.
• The more human creative decisions involved (outline, selection, editing, composition), the stronger the argument for protection tends to be. That is still interpretation, not settled rule.
• Training models on third-party works, and outputs that resemble existing works, are the most disputed points and can create risk for whoever publishes.
• The tool's terms of use define what you can do commercially. That is contract, not copyright.
• The most useful precaution is operational: record who created what, review before publishing, and avoid AI where exclusivity matters.

What copyright law says today

The starting point is Law 9.610. It protects intellectual creations expressed in any medium, such as texts, drawings, photographs and compositions. It also says the author is the individual who creates a literary, artistic or scientific work.

Three practical consequences follow from that wording, all based on the text of the law and its most common reading:

• Protection is tied to human creation. An AI tool is not an individual and is not an author.
• Moral rights (such as being recognized as the author) and economic rights (using, reproducing, publishing, authorizing) belong to the author and can be transferred by contract, generally in writing.
• Using someone else's work requires prior authorization, except for exceptions provided by law.

What the law does not do is say expressly what happens when a person gives a model a short instruction and receives finished text. That gap is where the questions start.

What is still debated

It helps to separate honestly what is statutory text from what is interpretation or hypothesis. These are the most relevant debates.

Is a prompt enough to make you the author?

Some argue that writing an instruction is a sufficient creative act. Others argue that the instruction is an idea or a command, and that the final expression was chosen by the model. The more cautious reading, especially when the instruction is simple, is the second one. When there are many human steps, such as structuring the content, choosing among variations, rewriting passages and combining it with your own elements, the argument for protecting the final result gets stronger. Treat this as a working hypothesis, not a guarantee.

Does human editing create protection?

If a person rewrites, reorganizes and adds original content, the part that person created may be protected. The hard part is drawing the line between what is theirs and what came from the model. That is why keeping a record of the process (versions, drafts, decisions) matters so much: it helps show the human contribution if someone challenges it.

What about training the models?

The models were trained on large volumes of content, and many countries are debating whether that requires authorization from rights holders or fits within some limitation. In Brazil, the question also appears in the legislative debate on AI. Because texts and bill status change, follow the current source and see our summary of the Brazil AI regulation. For a company that simply uses the tool, the practical effect is a residual risk: the output may resemble an existing work without your noticing.

Who owns what the tool generated?

These are two different questions that often get mixed up. One is whether copyright exists in the output (the law, above). The other is what the tool vendor lets you do with it (the contract).

A tool's terms of use usually address things like whether the user can use the output commercially, whether the vendor takes on any commitment if a third party makes a claim, and whether the content you submit can be used to improve the service. This varies by vendor and by plan, so read the current document for the tool your team actually uses instead of relying on summaries.

Two simple precautions:

• Prefer business plans and settings, where the terms for commercial use and data handling are clearer.
• Do not submit third-party material, client material or personal data to the tool without a clear basis. This mixes copyright with data protection law (LGPD) and confidentiality. A company policy for employee ChatGPT use helps define what may go in.

What risks appear when you publish AI text and images

The risk of publishing is not only about who the author is. It usually shows up in four areas.

• Similarity to an existing work: the text or image may reproduce passages, highly recognizable styles or elements from third-party works. If you publish it, you answer to the rights holder, even though the tool produced it.
• Trademarks and visual identity: logos, characters and packaging that appear in generated images may be signs protected by rules other than copyright.
• People's image and data: realistic faces, real people and voices involve image rights and, when they identify someone, LGPD. The ANPD, Brazil's data protection authority, is the official reference for guidance on data protection.
• Lack of exclusivity: if the content may not be protected, a competitor can copy your material without you having a strong basis to object. This matters for a logo, mascot, manual, course and any asset that supports the brand.

There is also the question of quality and trust. Content published without review, just for volume, tends to be generic and to fail the reader. Google, for example, advises creating helpful, reliable content for people, regardless of how it was produced, as its Search Central guidance on helpful content explains.

Illustrative scenario: the store that wants a mascot

Imagine a building materials store that asks an image tool for a brand mascot and starts using it on packaging, uniforms and ads. After a while, a competitor adopts an almost identical figure.

What matters in this scenario, in order:

• The store may struggle to claim copyright in an image produced with few instructions and no human creative intervention.
• If the mascot resembles an existing character, the store is the one exposed to a claim, not the tool.
• If the brand was registered without checking for prior similar marks, there is a third problem, independent of copyright.

A safer approach would be to hire a designer to develop the mascot from sketches, use AI only for early exploration, formalize the assignment of rights in writing, and register the trademark after a prior-rights search. It takes more time, but the asset becomes defensible. This is an example for reasoning, not a real case.

How to organize AI content use in your company

The method we apply to other AI topics works well here: map, limit, pilot and hand over. This is our implementation recommendation, not a legal requirement.

• Map: list where the company already uses AI to create text, images, audio and video, and classify each use by the value of the asset (a blog post matters differently from a logo or a course you sell).
• Limit: define where AI may produce the final version, where it may only help with drafts, and where it should not be used. Identity assets and content that requires exclusivity belong in the highest-caution group. The article on what not to delegate to AI in business helps you think through that boundary.
• Pilot: pick one workflow (for example, blog articles) and run it for a short period with mandatory human review, version records and an agreed success criterion, such as the share of pieces approved without legal or editorial rework.
• Hand over: at the end, document the process, train the team and decide whether to expand, adjust or stop. Whoever runs the workflow needs to know what to record and when to involve legal.

If the content is produced by an agency or vendor, include the topic in the brief and the contract. See also what changes in AI for marketing and creative agencies.

Checklist before publishing AI-made content

Use this list as a quick filter, adapting it to your case.

• Do the tool and plan you used allow commercial use of the output under the current terms?
• Does the content require exclusivity (brand, mascot, teaching material, manual)? If so, was there relevant human creation and a written assignment of rights?
• Did someone review the text for copied passages, unsourced claims and factual errors?
• Does the image contain trademarks, characters, artworks or recognizable people?
• Was any personal data or third-party material submitted to the tool? Was there a basis for that?
• Is there a record of who requested, who edited and who approved, with versions saved?
• Is there a named person responsible for the publication?
• Do doubtful cases have a clear path to an intellectual property lawyer?

When content is produced by an automated agent, such as replies and materials generated in a workflow, review should be part of the design, with output guardrails and logs. To structure risk more broadly, the NIST AI Risk Management Framework is a useful public reference.

Frequently asked questions

Is AI-generated content copyrightable in Brazil? The law does not expressly cover this case and defines the author as an individual. So material produced autonomously by the machine has uncertain protection, while material containing relevant human creation has better arguments. For important decisions, consult an intellectual property lawyer.

Can I use AI text or images in a commercial campaign? In general, yes, as long as the tool and plan allow commercial use under the current terms and you review the material. The main risk is publishing something similar to a third party's work or an existing trademark, and responsibility to the rights holder usually falls on whoever publishes.

Do I need to disclose that the content was made with AI? We found no general rule in the copyright law that requires this disclosure from every company, but there may be sector, platform or contract requirements, and transparency helps maintain reader trust. Check the rules of the channel and follow the evolution of AI regulation.

How do I prove there was human contribution to the content? Keep the drafts, versions, the instructions used and the record of edits and approvals, with the name and date of whoever did each step. This history does not guarantee protection, but it makes it much easier to show what the team created if there is a challenge.

Sources

• Planalto: Lei 9.610 (Lei de Direitos Autorais)
• ANPD: Autoridade Nacional de Proteção de Dados
• NIST AI Risk Management Framework
• Google Search Central: conteúdo útil, confiável e centrado nas pessoas