Policy on the Use of Artificial Intelligence
Revista de Direito FAE – RDF. This section forms part of the Guidelines for Authors.
Statement of reasons
Revista de Direito FAE recognises that generative artificial intelligence (GenAI) tools — writing assistants, machine translators and data analysis systems — have become part of everyday legal research. The use of such tools is not prohibited, but it must be ethical, transparent and responsible, in accordance with the Scientific Integrity Policy established by CNPq Ordinance No. 2,664 of 6 March 2026, and with international editorial best practice, in particular the IBFD Policy on the Use of Artificial Intelligence in Content Creation (International Bureau of Fiscal Documentation).
These guidelines are consistent with the artificial intelligence policies adopted by leading international scientific publishers and journals, among them Springer Nature, publisher of Nature; Elsevier, publisher of The Lancet and Cell; Taylor & Francis; and also Wiley, SAGE and IEEE; as well as with the recommendations of the International Committee of Medical Journal Editors (ICMJE, 2026 revision), followed by journals such as JAMA and The New England Journal of Medicine, and the guidance of the Committee on Publication Ethics (COPE). There is international convergence on the pillars adopted here: the prohibition on attributing authorship to AI systems, the requirement of a transparent declaration of use at submission, the maintenance of human oversight and responsibility at every stage, the prohibition on reviewers and editors entering manuscripts into AI tools, and restrictions on AI-generated images. In the specific international context of legal scholarship, these guidelines further follow the AI policy of the International Bureau of Fiscal Documentation (IBFD), an editorial reference in international tax law publishing.
1. General principles
1.1 Human oversight must be maintained at every stage of the preparation of the work. AI is an auxiliary tool; legal reasoning, argument, interpretation and conclusions must be entirely human.
1.2 Authors are fully and exclusively responsible for the accuracy, originality and integrity of the content submitted, including any errors, inaccuracies, bias, non-existent citations ("hallucinations") or plagiarism arising from the use of AI tools (CNPq Ordinance No. 2,664/2026; IBFD policy).
1.3 It is prohibited to enter confidential content, third-party personal data or copyright-protected material without authorisation into open-access AI tools that store data or use it for training, in compliance with the General Personal Data Protection Law (Law No. 13,709/2018) and the Copyright Law (Law No. 9,610/1998).
2. Permitted uses
Provided they are declared (item 4), the following uses of AI by authors are permitted:
a) research support: bibliographic searching, organisation of ideas, exploratory data analysis;
b) linguistic improvement: grammar and spelling revision, style suggestions, support in preparing the abstract and in formatting references;
c) paraphrase or summary of the author's own content.
2.1 Paraphrase or summary of the author's own content
AI may be used to rephrase, condense or summarise text previously written by the author within the submitted work — in which case the ideas, the analysis and the original expression are the author's, and the tool acts only on form. This permission covers, by way of example: rephrasing paragraphs of the author's own manuscript for clarity or concision; summarising the author's own sections to draft the introduction, conclusion or abstract; and adjusting the author's own passages to meet the journal's length limits.
This permission does not extend to:
a) using AI to paraphrase third-party texts (doctrine, annotated legislation, judgments, other authors' articles) in order to incorporate them into the work without proper citation of the source; paraphrasing another's work, with or without AI, requires credit to the original author, and failure to give it constitutes plagiarism (CNPq Ordinance No. 2,664/2026; Law No. 9,610/1998);
b) paraphrasing the author's own already-published work in order to present it as unpublished, which constitutes self-plagiarism or duplicate publication and breaches the originality requirement of the journal's submission guidelines;
c) using paraphrasing tools to disguise textual similarity and circumvent plagiarism-detection systems; the IBFD policy expressly warns that it uses tools to detect the use of AI writing and paraphrasing tools in submitted work, and this journal may adopt an equivalent practice (item 6).
Authors must check the result of any paraphrase or summary in full, since automatic rephrasing may alter the technical legal meaning of concepts, institutions and statutory quotations. Such use remains subject to the declaration under item 4, stating the tool and the purpose.
In every case, the output produced by the tool must be verified, edited and rewritten in the author's own voice, with full checking of the sources cited.
3. Prohibited uses
Authors are prohibited from:
a) attributing authorship or co-authorship to AI systems. AI tools do not meet the requirements of authorship: they cannot take responsibility for the content submitted and, as non-legal entities, they cannot attest to the existence or absence of conflicts of interest, hold copyright or enter into licence agreements (CNPq Ordinance No. 2,664/2026; COPE position, followed by WAME and the JAMA Network; IBFD policy; Law No. 9,610/1998, art. 11);
b) submitting as their own work any text created or substantially written by AI, as defined in item 3.1;
c) using AI in place of the intellectual activities proper to scholarly work, such as reasoning, the construction of argument and the formulation of conclusions;
d) using AI to generate images, illustrations or artistic figures for the article (only charts drawn from real data and basic technical image adjustments are permitted);
e) omitting or falsely declaring the use of AI tools.
3.1 What counts as "text created or substantially written by AI"
Text is regarded as created or substantially written by AI where its intellectual origin — the conception of the ideas, the structure of the argument, the analysis and the expression — comes from the tool rather than from the author, even if the author subsequently edits it. The test is not the number of words altered (the percentage of AI use reported by common online detection tools), but the origin of the intellectual contribution: if, once the AI's contribution is removed, the text does not stand as the author's own creation, the writing is substantial. This test follows from the Copyright Law (Law No. 9,610/1998, art. 11), under which the author is the natural person who creates the work, and from CNPq Ordinance No. 2,664/2026, which prohibits submitting AI-generated content as human production.
The prohibition covers, by way of example:
a) using AI to generate entire sections of the article (introduction, substantive sections, conclusion) from prompts and incorporating them into the text, even with subsequent editing or rephrasing;
b) asking AI to formulate the central thesis, the line of argument, the analysis of doctrine, legislation or case law, or the conclusions of the work (intellectual activities that the IBFD policy expressly reserves to the author);
c) incorporating an AI-produced literature review without the author having read and verified the works cited;
d) using AI to turn notes, outlines or bullet points into continuous prose presented as the author's own writing ("algorithmic ghost-writing");
e) submitting an AI translation of a whole text as the final version, without specialist human revision (the IBFD policy does not accept full AI translation of works and recommends human translators).
The following do not amount to substantial writing and remain permitted, provided they are declared (item 4): assistive uses applied to text previously written by the author — grammar, spelling and style revision; targeted suggestions for clarity and concision; paraphrase or summary of the author's own passages; support in preparing the abstract and in formatting references. In every case the output must be checked and rewritten in the author's own voice, and the author remains fully responsible for the final content.
Where there is doubt as to classification, the author must describe the use made in the Declaration on the use of artificial intelligence (item 4), and the Editorial Board will assess its compatibility with this policy.
4. Mandatory declaration of AI use
Under CNPq Ordinance No. 2,664/2026, any use of generative AI, at any stage of the development of the research and of the writing, must be declared, specifying the tool used (name and version) and the purpose of its use.
The declaration must appear in a specific note at the end of the article, before the references, under the heading "Declaration on the use of artificial intelligence", following this model:
"In preparing this work, the author(s) used the tool [NAME AND VERSION] for the purpose of [PURPOSE — e.g. grammar and style revision / translation of the abstract / preliminary organisation of the bibliography]. After using the tool, the content was reviewed and edited in full by the author(s), who assume(s) full responsibility for the content of the publication."
Where no AI tool was used, a negative declaration is recommended: "The author(s) declare(s) that no artificial intelligence tools were used in preparing this work."
4.1 Citation, notes and references for AI-generated content
ABNT has not yet issued a specific standard for citing AI-generated content (NBR 6023:2018 does not address it). Until such a standard exists, the journal adopts the format below, adapted from NBR 6023. Since AI systems cannot be authors (item 3, "a"), the entry is made under the developer of the tool (company or institution). Because AI responses are not reproducible, authors must retain a full transcript of the exchange (prompt and response), which the Editorial Board may request.
a) Table or chart prepared with AI assistance — state in the source note beneath the element:
Table 1 – Limitation periods in damages claims (fictitious example)
Source: prepared by the authors (2026), with the assistance of ChatGPT (GPT-5, OpenAI) in organising and formatting the data. Content checked in full by the authors.
b) Graph prepared from real data with AI assistance (the only form of image permitted — item 3, "d"):
Graph 2 – Trends in AI-related claims before the Paraná Court of Justice (2020–2025) (fictitious example)
Source: prepared by the authors from data in the CNJ Statistics Panel (2025), with the assistance of Claude (Anthropic) in tabulation. Checked in full by the authors.
c) Transcript of an AI response as the object of analysis — where the tool's response is itself the material examined in the article, direct quotation applies (4 cm indent if longer than three lines), with a footnote:
In the body of the text (fictitious example):
Asked about the nature of the developer's civil liability, the tool replied that "liability would be strict, founded on the theory of activity risk, under the sole paragraph of article 927 of the Civil Code".¹Corresponding footnote:
¹ OPENAI. ChatGPT (GPT-5). Response generated on 12 March 2026 to the prompt: "What is the nature of the civil liability of AI system developers under Brazilian law?". Full transcript held by the authors.
Under the author-date system, the in-text citation would read: (OPENAI, 2026).
RDF note: the author-date system is given here as general guidance, for authors and editors of journals that use it. At Revista de Direito FAE, citation is accepted exclusively in footnotes (item 8 of the Guidelines for Authors).
d) Translation of the abstract with AI assistance — footnote attached to the abstract heading:
² English version prepared with the assistance of DeepL, revised by and under the responsibility of the authors.
e) Reference at the end of the article — required only in the case of paragraph "c" (AI as a cited source); in all other cases, the mention in the element's note and in the declaration under item 4 is sufficient:
OPENAI. ChatGPT (GPT-5) [large language model]. San Francisco: OpenAI, 2026. Available at: https://chat.openai.com. Accessed: 12 March 2026.
In every case, the Declaration on the use of artificial intelligence (item 4) remains mandatory; it neither replaces nor is replaced by the notes and references in this section.
4.2 AI used in empirical research methodology
Where an AI tool forms part of the methodological design of the work — for example, to classify judicial decisions, extract information from large volumes of documents, code interviews or identify patterns in data — it must be described in the methodology section of the article, and not merely in a footnote. This is a requirement of scientific replicability: under CNPq Ordinance No. 2,664/2026, the use of generative AI must be declared at any stage of the development of the research, specifying the tool and the purpose.
The methodological description must state, as a minimum: (i) the tool and version used; (ii) the task assigned to it; (iii) the prompts or parameters used, in full or in an appendix; (iv) the human validation procedure applied to the results; and (v) the known limitations of the technique (possible misclassification, model bias, non-reproducibility of responses).
Fictitious example of wording in the methodology section:
A total of 1,240 judgments of the Paraná Court of Justice handed down between 2020 and 2025 were collected by searching the case-law system for the terms "artificial intelligence" and "civil liability". The preliminary classification of the decisions by outcome (upheld, dismissed or partly upheld) was carried out with the assistance of Claude (version Opus 4.5, Anthropic), using the prompt reproduced in Appendix A. For validation, a random sample of 20% of the judgments (n = 248) was reclassified manually by the authors, with an agreement rate of 96.4%; divergences were reviewed individually and, in every case, the human classification prevailed. The authors take full responsibility for the results. The headnotes processed are public and contain no sensitive personal data, in compliance with Law No. 13,709/2018 (LGPD).
Further observations:
a) it is prohibited to enter confidential documents or personal data without a legal basis into the tool, which in legal research covers case files under seal, non-anonymised interviews and non-public pleadings (item 1.3; Law No. 13,709/2018);
b) human validation of the results is indispensable: AI cannot be the final instance of analysis, on pain of breaching item 3, "c" (substitution for the intellectual activities proper to scholarly work);
c) the methodological use of AI must also appear, in summary form, in the Declaration on the use of artificial intelligence (item 4) — the methodology section details the procedure; the declaration consolidates all uses of the tool in the work;
d) where the research involves human subjects (interviews, questionnaires) with data processed by AI, the applicable ethical approvals (Research Ethics Committee) remain required, in accordance with the CNPq Integrity Policy.
5. Reviewers and editors
Consistent with the IBFD policy, reviewers and editors are prohibited from: (i) entering submitted manuscripts, in whole or in part, into AI tools, as this breaches the authors' copyright and the confidentiality of review; (ii) preparing reports or confidential communications by means of generative AI; and (iii) delegating the review or the editorial decision to AI systems. Responsibility for review and for the editorial decision is always human.
6. Verification and consequences
The journal may use similarity-detection and AI-content-detection tools when screening submissions. A breach of this policy may result in rejection of the submission and retraction of an article already published, without prejudice to the sanctions provided for in the CNPq Integrity Policy for researchers and scholarship holders (ranging from a warning to suspension of grants and awards and repayment of funds).
7. Note on the Brazilian legal framework
To date, Brazil has no enacted general statute on artificial intelligence. Bill No. 2,338/2023 (the AI Legal Framework) was approved by the Federal Senate on 10 December 2024 and is under consideration by a Special Committee of the Chamber of Deputies. The General Personal Data Protection Law (Law No. 13,709/2018), the Copyright Law (Law No. 9,610/1998) and, within the Judiciary, CNJ Resolution No. 615 of 11 March 2025 (as amended by CNJ Resolution No. 674 of 25 March 2026), which governs the development and use of AI in the Judiciary, apply so far as relevant. This policy will be updated as the legislation develops.
8. Sources used in this policy
- CNPq — Ordinance No. 2,664 of 6 March 2026 (Scientific Integrity Policy). Available at: gov.br.
- IBFD Policy on the Use of Artificial Intelligence in Content Creation (updated May 2026). Available at: ibfd.org.
- ICMJE Recommendations — use of AI by authors. Available at: icmje.org.
- Bill No. 2,338/2023 — AI Legal Framework. Special Committee of the Chamber of Deputies. Available at: camara.leg.br.
- CNJ Resolution No. 615 of 11 March 2025, as amended by CNJ Resolution No. 674 of 25 March 2026. Available at: atos.cnj.jus.br.
- Law No. 13,709/2018 (LGPD) and Law No. 9,610/1998 (Copyright) — legislation applicable on a subsidiary basis.
- Guidelines for Authors of Revista de Direito FAE. Available at: revistadedireito.fae.edu.
Document prepared during the first half of 2026 and finalised on 30 July 2026 for consideration by the Editorial Board.
Declaration on the use of artificial intelligence by the author of these guidelines
These guidelines were prepared by the Editor of Revista de Direito FAE with the assistance of the tool Claude (model claude-fable-5, Anthropic), used through the Claude application, in July 2026, for the following purposes: identifying and systematising the regulatory and editorial sources on the use of artificial intelligence in scholarly production (CNPq Ordinance No. 2,664/2026; IBFD AI policy; Bill No. 2,338/2023; CNJ Resolution No. 615/2025; related legislation) and preparing a first draft of the text, including the declaration models, notes and references. After the tool was used, the content was reviewed, edited and adapted in full by the Editor, who assumes full responsibility for the substance of these guidelines.
Curitiba, 30 July 2026.
Thaïs Savedra de Andrade
Editorial Committee of Revista de Direito FAE – RDF
