Recent legislative and regulatory developments in Brazil suggest an emerging consensus: AI may assist the inventive process, but inventorship remains human. The difficult question is not who invented, but what must be disclosed when AI contributes to the invention.
On May 13, 2026, Brazilian Congress’ Committee on Science, Technology, and Innovation replaced the previous wording of Bill #3,936/2024 and recommended its approval with the proposed amendments. The legislative direction preserves inventorship for human actors and treats AI as a tool in the inventive process, rather than as an inventor.
This approach aligns, at least in broad terms, with the position currently adopted by the BRPTO. In August 2025, the BRPTO launched Public Consultation #3/2025 on draft Examination Guidelines for AI-related inventions. The proposal classifies such inventions into three categories: (i) AI models and techniques; (ii) AI-based inventions; and (iii) AI-assisted inventions. Regardless of the category, the invention must solve a technical problem, present a technical solution, and produce a technical effect. The draft guidelines also make explicit something that has long been implicit in Brazilian practice: inventions generated autonomously by AI, without human intervention, are not eligible for patent protection, as inventorship must ultimately be attributed to a natural person.
While the debate on AI inventorship has attracted significant attention, a more complex question is how patent offices will assess sufficiency of disclosure, written description, and enablement in AI-assisted inventions.
This is particularly relevant in biotechnology and gene therapy. AI tools are now routinely used to predict biological targets, identify and rank candidate molecules, and optimize constructs or combinations. However, the use of AI does not diminish the fundamental requirements of patent law and may, in practice, reinforce them. The key question remains the same: Can a skilled person in the art reproduce the invention without undue experimentation when AI-engineered tools have been used?
The BRPTO’s draft Examination Guidelines emphasize that patent applications involving the use of AI tools must include sufficient technical detail to allow reproducibility. Depending on the nature of the invention, this may require a relatively detailed level of disclosure, including the dataset used (or at least sufficiently described), the relationship between input data and outputs, the model or algorithm employed, relevant parameters, as well as training and validation methods supporting the claimed technical effect.
The draft guidelines also caution against over-reliance on so-called “black box” systems. In other words, the BRPTO current understanding is that simply stating that an AI model generated a result is not enough to fulfill the enablement requirement if the underlying technical contribution cannot be meaningfully reproduced.
These issues become more serious in the context of genus-type claims, which remain central in biotechnology practice. In the US, for instance, the discussions have gained more traction lately after the U.S. Supreme Court’s decision in Amgen v. Sanofi. The US precedent serves as a clear example of tension between broad functional claims and the enablement requirement. Even though we cannot anticipate the same interpretation in Brazil, a similar debate may arise if the BRPTO adopts a stricter interpretation of disclosure requirements, as recent draft Guidelines and the Public Consultation appear to suggest.
AI systems may identify or predict large classes of potentially functional molecules or biological entities, but patent protection does not automatically extend to that entire predicted class.
As a result, applicants may need to be more deliberate. It is often necessary to show that the specification does more than simply point to a result; it must provide a coherent technical rationale that supports the full scope of the claim. In some cases, that may mean identifying a unifying structural feature; in others, it may mean providing enough representative examples to make the claimed scope credible from a technical standpoint.
From a drafting perspective, this is where many AI-assisted applications will succeed or fail. For companies operating in biotechnology and life sciences, the implications are relatively immediate. It is no longer enough to document that AI was used. What matters is how the invention is framed and disclosed.
From a practical standpoint, applications should clearly describe: (i) the human role in defining and validating the invention; (ii) the technical pathway from AI output to a concrete solution; and (iii) the elements of the disclosure that support the full scope of the claims.
In practice before the BRPTO, these aspects are likely to be scrutinized more closely as examination of AI-related inventions becomes more structured.
In conclusion, Brazil’s legislative developments and the BRPTO’s draft guidelines reflect a broader international trend: the patent system is adapting to artificial intelligence by treating it as an enabling tool, rather than a source of inventorship.
That is not particularly controversial. What is less settled, and likely to be more consequential in the short term, is how patent offices will apply disclosure requirements to AI-assisted inventions, especially in technically complex and less predictable fields. In biotechnology and gene therapy, this distinction may ultimately determine how robust and defensible patent protection will be in practice.
The next phase of this debate will depend not only on how the BRPTO applies these standards during examination, but also on how Brazilian courts interpret and review such administrative decisions, particularly as these questions are likely to arise as matters of first impression in patent litigation.