Artigo de revisão
PROFESSIONAL LIABILITY OF THE MEDICAL EXAMINER IN THE USE OF GENERATIVE ARTIFICIAL INTELLIGENCE
Como citar: Trunckle YF, Bizutti GDSC. PROFESSIONAL LIABILITY OF THE MEDICAL EXAMINER IN THE USE OF GENERATIVE ARTIFICIAL INTELLIGENCE. Persp Med Legal Pericia Med. Vol. 11, 2026; 260541.
https://dx.doi.org/10.47005/260541
Recebido em 05/02/2026
Aceito em 13/02/2026
The authors report no conflict of interest.
PROFESSIONAL LIABILITY OF THE MEDICAL EXAMINER IN THE USE OF GENERATIVE ARTIFICIAL INTELLIGENCE
Resumo
INTRODUÇÃO: O uso da Inteligência Artificial (IA) generativa na prática pericial promete eficiência, mas impõe conflitos bioéticos e normativos, especialmente quanto à responsabilidade profissional do médico legista nas esferas cível e penal. O objetivo foi analisar os limites dessa responsabilidade considerando a natureza de "caixa-preta" dos algoritmos. MATERIAL E MÉTODO: Trata-se de revisão narrativa da literatura nas bases PubMed, Scielo, LILACS e Google Scholar, realizada em setembro de 2025, utilizando descritores sobre responsabilidade legal e sistemas inteligentes. RESULTADOS: Foram selecionados e analisados artigos que tratam da responsabilidade médica genérica, dada a escassez de literatura específica para a perícia oficial. Identificou-se que a falta de transparência algorítmica conflita com o dever de fundamentação do laudo. DISCUSSÃO: A discussão confronta a eficiência produtiva da IA com os riscos éticos, destacando o conflito com o método cartesiano e a exigência do Art. 473 do CPC. A legislação atual (PL 2.338/2023) tende a classificar a IA na saúde como alto risco. CONCLUSÃO: Conclui-se que, devido à responsabilidade subjetiva e à exigência de fundamentação científica reprodutível, o médico legista permanece integralmente responsável pelo laudo, devendo a IA atuar apenas como ferramenta auxiliar.
Palavras Chave: Responsabilidade Civil; Medicina Legal; Inteligência Artificial Generativa; Médico Legista; Ética Médica.
Abstract
INTRODUCTION: The use of generative Artificial Intelligence (AI) in forensic practice promises efficiency but imposes bioethical and normative conflicts, especially regarding the professional liability of the medical examiner in civil and criminal spheres. The objective was to analyze the limits of this liability considering the "black-box" nature of algorithms. MATERIAL AND METHOD: A narrative literature review was conducted in PubMed, Scielo, LILACS, and Google Scholar databases in September 2025, using descriptors regarding legal liability and intelligent systems. RESULTS: Articles addressing generic medical liability were selected and analyzed due to the scarcity of literature specific to official forensics. It was identified that algorithmic lack of transparency conflicts with the duty to substantiate the report. DISCUSSION: The discussion confronts the productive efficiency of AI with ethical risks, highlighting the conflict with the Cartesian method and the requirement of Art. 473 of the CPC. Current legislation (Bill 2.338/2023) tends to classify AI in health as high risk. CONCLUSION: It is concluded that, due to subjective liability and the requirement for reproducible scientific reasoning, the medical examiner remains fully responsible for the report, with AI serving only as an auxiliary tool.
Keywords (MeSH): Damage Liability; Forensic Medicine; Generative Artificial Intelligence; Coroners and Medical Examiners; Ethics, Medical.
1. INTRODUCTION
The use of Generative Artificial Intelligence (AI) has been intensifying greatly in daily life, with the prospect of increasing productivity, work efficiency, and reducing errors. These aspects are also applicable in the forensic field, and the debate regarding its use in medicine is evident. The discussion on how far AI can advance within medical practice, such as in diagnostic aid, image analysis, and even prescription (1), is of paramount importance due to the bioethical, normative, and technical conflicts it imposes.
Given the inevitable technological innovation of AI in medical practice, a normative evaluation implies its limitations for the full replacement of the medical examiner’s professional activity is necessary. When dealing with the professional liability of the physician, we must recall that it falls under civil, criminal, administrative (Regional Medical Councils), and functional spheres, which in the case of the medical examiner adheres to the Organic Law of the Civil Police of the State of São Paulo (2), for example.
It is necessary to establish the differences between each of them. In the civil sphere, the focus is on repairing damage, seeking to restore the claimant’s patrimonial, extra-patrimonial, and personal assets through compensation. Generally, medical liability in this scope is subjective, meaning it requires proof of intent or fault through malpractice, negligence, or imprudence, which is also evaluated by the Regional Medical Councils (3) when dealing with administrative liability from an ethical-professional standpoint. In the criminal sphere, in turn, the analysis relates to the defense of collective rights with punitive consequences.
The importance of AI is clear as an auxiliary tool. The limitations regarding professional liability for potential damages caused by this technology, directly or indirectly, call for a legal determination to ensure safety for the population, given the complexity of the medical act as well as the occurrence of the “black box” phenomenon, which is the inability to fully understand the decision-making process of an AI and the inability to predict AI decisions or results, as described by Chan et al. (4). Given this scenario of such complexity, questions regarding civil and criminal liability regarding the use of this technology become necessary, as it challenges established paradigms (5).
The core of this issue, from a medico-legal point of view, lies in the growing ambiguity and evolutionary nature of the physician’s liability — especially in Brazil — both in the civil and criminal spheres, because, when AI systems participate in patient care, there is naturally the risk of adverse events occurring, just as occurs in healthcare provided by humans.
There is an inherent conflict in the use of AI both by the physician and the patient themselves. This is due to the operational nature of AI systems, which reflect the so-called “black boxes” (4), meaning their operational complexity is difficult for humans, including their own developers, to understand. This generates an obscure decision-making process by the AI, making it susceptible to methodological questioning and, consequently, biases. Furthermore, such a context makes auditing difficult, as a potential failure can occur both in the data provided and training data, as well as a failure in the internal logic of this technology’s algorithm.
We also have a normative conflict regarding the physician’s professional liability with the use of AI. To date, there is no clear legal delineation on who is responsible for potential damage (the AI or the professional using it). As well as the standard of civil liability (subjective or objective), for example, in addition to the definition of culpability by the judge (negligence, malpractice, or imprudence) considering the intervention of an algorithm in this decision-making process.
Given this context, regarding the specific activity of the medical examiner, there are several possibilities for implementing this technology in their routine, especially in drafting Reports. It is undeniable that the use of AI in forensic practice tends to be very useful in orthographic and textual optimization, allowing for greater clarity in exposing the argumentative logic the expert uses in the document they draft. It can also greatly assist in the speed with which scientific evidence is found to substantiate the forensic report.
However, the misuse of this tool can generate damages to society that imply the need for professional accountability, such as the unrestricted application of generative AI for the full drafting of the document. Such questioning is important, considering that it generates a methodological flaw in forensic practice directly related to the professional liability of the physician who signed the report that was not drafted by them but by an algorithm, taking into account the “black box” concept discussed earlier.
The fact that the expert has the duty to describe the method of their work, whether explicitly in Art. 473 of the Civil Procedure Code (6) or implicitly in the Criminal scope, already determines an important limitation on the unrestricted use of AI in medico-legal practice, since this technology does not allow clarity in the description of the method used by it and, therefore, prevents the reliability of the conclusion issued and consequently the validity of the technical evidence to be used.
Nério Rojas advocated the use of the Cartesian method for the practice of the medical examiner (7). Descartes discusses in his work “Meditations on First Philosophy” that the accumulation of false opinions since childhood makes the knowledge built upon them doubtful and uncertain, so that undoing these beliefs, starting from the beginning and, therefore, their scientific foundations is of paramount importance (7). We see, therefore, that the use of the Cartesian method cited by Nério Rojas (7) is of great value in forensic practice, since we start from scientific principles and not questionable premises (7), aligning with good forensic practice.
However, the introduction of Artificial Intelligence (AI) presents immeasurable productive potential in various areas of daily life and including in medical expertise, given that AIs act through algorithms, imitating human cognitive functions allowing for problem-solving and learning (8). Such questioning also raises the need for definition for potential distribution of liability: to the AI developer, the professional using the tool, and/or the institution implementing it. A potential failure of the AI algorithm used may characterize a specific liability of the technology developer (9).
Given this context, we note a fundamental limiting factor for the full insertion of this technology in medical practice: the professional liability of the physician. Linked to this factor, we have the importance of norms regulating the practice of medicine, such as Law No. 12.842 of 2013 in Brazil – the Medical Act Law – which delimits the private activities of these professionals such as performing diagnoses and invasive procedures, for example (10).
This work aims to answer a question about the limits and implications in the civil and criminal spheres for the professional liability of the medical examiner when using Generative Artificial Intelligence (AI) in drafting forensic reports in Brazil. Therefore, we can list hypotheses for the answer to this question, such as: the absence of specific legislation on the use of AI in Legal Medicine fully attributes liability for potential damages resulting from reports drafted with the aid of this technology to the medical examiner; and the “black box” nature of generative AIs represents an insurmountable methodological obstacle to their unrestricted use, violating the expert’s duty to describe the method employed so that their work is reproducible.
2. MATERIAL AND METHOD
This is a narrative review of scientific literature and norms related to the topic. The bibliographic research covered scientific articles published in databases such as PubMed, Scielo, LILACS, and Google Scholar, using the following search strategy (((“Liability, Legal”) AND (“Coroners and Medical Examiners” OR “Physicians”) AND “Intelligent Systems”))), carried out on 09/23/2025.
Inclusion criteria included articles in Portuguese, English, or Spanish, selected through thematic compatibility of titles with the thematic scope of this review and which were also open access. Exclusion criteria were characterized by works in languages other than those established above, as well as outside the thematic scope of this review.
Eleven articles were found in the search performed through Google Scholar, and no results were found in the other databases. Of the 11 articles found in the cited search, only two were selected by the established inclusion criteria, and both did not specifically deal with the professional liability of the medical examiner but rather of the physician in their performance in the health area.
3. RESULTS
The search strategy applied in the selected databases returned a total of 11 references, all located through the Google Scholar tool. In the PubMed, Scielo, and LILACS databases, no exact corresponding results were found for the combined descriptors on the research date.
After applying the inclusion and exclusion criteria, performing a manual search in the described databases, followed by reading the titles and abstracts, 2 articles were selected for qualitative analysis, as shown in Table 1. A scarcity of specific literature on the medical examiner’s performance regarding AI is noted, with most studies focused on clinical care practice.
Table 1 – Bibliographic search results
| Database | Search terms used | Total found | Selected |
| PubMed | “Liability, Legal AND Coroners/Physicians AND Intelligent Systems” | 0 | 0 |
| Scielo | “Liability, Legal AND Coroners/Physicians AND Intelligent Systems” | 0 | 0 |
| LILACS | “Liability, Legal AND Coroners/Physicians AND Intelligent Systems” | 0 | 0 |
| Google Scholar | “Liability, Legal AND Coroners/Physicians AND Intelligent Systems” | 11 | 2 |
| Total | 11 | 2 |
Source: elaborated by the author.
4. DISCUSSION
Forensic practice, and consequently that of the forensic physician, demands a complex method of technical argumentation of medical and legal knowledge in essence. This method is part of the scientific premise, demanding rigor and clarity, in face of Cartesian precepts and other philosophers of science who explicitly address this topic (7). The capacity for factual demonstration of how a forensic conclusion is reached is elementary to the validity of a technical-scientific work (7) such as the one under discussion, considering that its legal validity is directly proportional to this.
This has legal grounds, including in current Brazilian norms, such as the Civil Procedure Code itself (6) in its Article 473, which explicates the need for the expert (whether medical or not) to clarify the method used. Although the medical examiner does not have this obligation in their daily practical legal framework because they act eminently before the Criminal Procedure Code (11), it is notorious that before forensic medical doctrine such a demand is standard in forensic activity, since such a precept implies the reproducibility of the expertise.
As previously brought up, we reinforce the doctrine of Nério Rojas and numerous other authors who ratify this fundamental element, so that we evidence an important limitation of the misuse of generative AI in medico-legal practice. This occurs because it presents algorithms in its structuring that are not transparent – the “black boxes” – in the way they formulate answers, diametrically opposed to the basic precept of the forensic method.
In practice, in a hypothetical situation where the expert fully uses the content produced by generative AI, they would be validating a document made by “third parties,” in this case, the technology under analysis. This impasse generates a significant normative conflict regarding the professional liability of the forensic physician using it, since reparation for potential damage caused is not peaceful in our legal system at the moment.
Currently in Brazil, we have as a regulatory panorama for AI in the health context a legal framework and Bills of Law (PL) that specifically propose to establish limits on the use of this technology. They are: Civil Code, Consumer Defense Code (Law No. 8.078/90), General Data Protection Law (12), and Medical Act Law (Law No. 12.842/13). Regarding legislative proposals on AI, the PL aiming to establish a legal framework for technology in the country stands out.
The most recent draft of Bill 2.338/2023 (13) seeks to guarantee the protection of fundamental rights based on the centrality of the human person. The Bill presents a risk classification for AI systems and expressly records “applications in the health area to assist diagnoses and medical procedures” as being high risk (13). The text addresses Civil Liability in a specific chapter, which establishes following the Consumer Defense Code (14) in its Art. 35 and generally the Civil Code.
Regarding the establishment of the liability regime (subjective or objective) in a concrete case, the level of autonomy of the AI system and its degree of risk must be considered, in addition to authorizing the judge to reverse the burden of proof when the claimant is hypofunctional or when the characteristics of the AI system make proof excessively difficult for them. Such a situation makes the defense of the supplier or operator of the technology more complex, corresponding to the level of inherent risk attributed to medical activity.
Within the scope of Medical Councils, the recent positioning makes it clear that AI should be used as a tool for the medical act and not as a substitute for it (15). It is worth noting that there is no direct mention of forensic activity, although applicable in its essence, but which diverges greatly from clinical care medical practice.
In the Brazilian legal system, considering that medical liability is subjective, implies that for this professional to be obliged to indemnify an individual for damage suffered in the provision of this service, it is necessary to prove that the physician acted with fault, as established in Art. 951 of the Civil Code (16) and specifically in consumer relations, more precisely in Art. 14 §4 of the Consumer Defense Code (14). It is also worth highlighting that in the case of clinics and hospitals, liability is objective, therefore demanding an obligation of result (17). This is the Brazilian jurisprudential understanding.
One must also question the difference between obligation of means and obligation of result in the context of AI use. It is known that a priori the obligation of means is what characterizes medical activity, that is, the physician commits to employing technical knowledge, diligence, and available resources in the best way aiming at the patient’s health, but does not guarantee obtaining a specific result. It differs, therefore, from the obligation of result, where there is a reversal of the burden of proof presuming the professional’s fault. Such situations, although controversial from a medico-legal perspective, are characterized in medical activities with aesthetic purposes or even in laboratory tests (18).
The understanding of the Federal Council of Medicine regarding these concepts tends to maintain, therefore, medical activity as an obligation of means (3) even in the context of AI use. The fact that it uses algorithms with a high accuracy rate with potential scientific validation can certainly generate questions regarding the appropriate obligation of the professional in these cases.
However, it is worth noting that understanding the use of AI as a tool does not change the obligation of means established for the physician, but makes the situation more complex, as the professional now has the requirement not only of the vast technical-scientific knowledge incumbent upon them, but also of the careful choice and appropriate use of an AI tool. A potential failure of these aspects in handling this technology can trigger the breach of the obligation of means and consequently medical culpability in these cases.
In this context, the great legal challenge introduced by Bill 2.338/2023 is the potential conflict between the traditional subjective liability of the physician – based on the obligation of means – and the new mechanism of reversal of the burden of proof. Although the draft does not establish objective liability explicitly, by transferring to the professional the duty to prove that they acted correctly in the face of a high-complexity tool, the practical effect may approach an obligation of result, generating legal insecurity in medical activity.
Still without a concrete normative definition regarding these situations in Brazil, we have as a fact the need for specific and balanced regulation in these cases, avoiding unfair accountability in a complex profession like medicine and, on the other hand, ensuring protection for society regarding the use of this technology in the health area, which in turn has great beneficial potential when applied appropriately.
Regarding the criminal liability of the physician in the use of AI, we have even greater challenges than in civil liability, given the need for proof of fault in the strict sense and the principle of legality governing Criminal Law. There is no criminal typification for these situations in the Brazilian legal system to date, which leads us to analyze based on existing provisions in the Penal Code (CP) that are pertinent to cases involving damages associated with health care with harmful results, whether bodily injury in its Art. 129, or homicide under Art. 121 (19).
Although currently we have legal obstacles to the use of this technology in medico-legal practice, it is important to highlight its positive potential. Specifically in the forensic area, generative AI tools allow for optimization and greater agility in the delivery of reports, whether due to their corrective possibilities for spelling, cohesion, and textual coherence, as well as critical analysis of the expert’s argumentative logic. This is fundamental for them to fulfill their role of clarification to authorities regarding their area of knowledge.
Likewise, with the advent of AIs that enable reliable and accurate search for scientific evidence, it allows for appropriate deepening and direction to the technical discussion of the matter of the concrete case. As previously explained about the forensic method allied with good technical-scientific practice, we understand that analogously to regulations on the use of AI for producing scientific articles (20), the use of these tools must be made clear in the report.
5. CONCLUSION
Given this context, we note up to the time of drafting this work that if there is damage due to the inappropriate use of AI by the medical examiner, they will be considered fully responsible, also considering current jurisprudence. The signature of the one who subscribes the report is what validates the forensic work, which aligns with our legal system and consequently, the judiciary system will rely on the only human agent responsible for the final act.
From a methodological and scientific point of view, the obstacle regarding the characterization of objective and subjective liability conflicts with the aforementioned “black boxes” of AIs, invalidating their transparency and consequently the analysis of potential fault for their use in forensic practice.
Regarding the criminal sphere, the challenges are even greater, given the difficulty of proving intent or fault in the strict sense, making the difficulty of criminal typification of an expert who is potentially induced to error due to an AI algorithm patent.
References
1. Alowais SA, Alghamdi SS, Althewaibi S, et al. Revolutionizing healthcare: the role of artificial intelligence in clinical practice. BMC Med Educ [Internet]. 2023;23(1):689.
2. São Paulo (Estado). Lei Complementar nº 207, de 05 de janeiro de 1979. Dispõe sobre a Lei Orgânica da Polícia do Estado de São Paulo. Diário Oficial do Estado [Internet]. 1979 Jan 06.
3. Conselho Federal de Medicina (BR). Código de Ética Médica: Resolução CFM nº 2.217, de 27 de setembro de 2018. Brasília: Conselho Federal de Medicina; 2019.
4. Chan B. Applying a common enterprise theory of liability to clinical AI systems. Am J Law Med [Internet]. 2021;47:351-85.
5. Januário TFX. Inteligência Artificial e responsabilidade penal no setor da medicina. Lex Medicinae. 2020;17(34):37-63.
6. Brasil. Lei nº 13.105, de 16 de março de 2015. Institui o Código de Processo Civil. Diário Oficial da União [Internet]. 2015 Mar 17.
7. Freire JJB. Da doutrina e do método em Medicina Legal: ensaio epistemológico sobre uma ciência bio-psico-social [tese]. São Paulo: Universidade de São Paulo, Instituto de Psicologia; 2009.
8. McCarthy J. O que é inteligência artificial? [Internet]. 1998.
9. Wimmer M. Inteligência artificial, algoritmos e o direito: um panorama dos principais desafios. In: Lima APMC, Hissa CB, Saldanha PM, coordenadores. Direito digital: debates contemporâneos. São Paulo: Thomson Reuters Brasil; 2019.
10. Brasil. Lei nº 12.842, de 10 de julho de 2013. Dispõe sobre o exercício da Medicina. Diário Oficial da União [Internet]. 2013 Jul 11.
11. Brasil. Decreto-Lei nº 3.689, de 3 de outubro de 1941. Código de Processo Penal. Diário Oficial da União. 1941 Out 13.
12. Brasil. Lei nº 13.709, de 14 de agosto de 2018. Lei Geral de Proteção de Dados Pessoais (LGPD). Diário Oficial da União [Internet]. 2018 Ago 15.
13. Brasil. Senado Federal. Projeto de Lei nº 2.338, de 2023. Dispõe sobre o uso da inteligência artificial. Minuta de relatório apresentado pelo Senador Eduardo Gomes. Brasília: Senado Federal; 2024 Dez 19.
14. Brasil. Lei nº 8.078, de 11 de setembro de 1990. Dispõe sobre a proteção do consumidor e dá outras providências. Diário Oficial da União [Internet]. 1990 Set 12.
15. Conselho Regional de Medicina do Estado do Rio Grande do Sul (BR). Resolução nº SEI-6, de 12 de março de 2025. Diário Oficial da União. 2025 Maio 28.
16. Brasil. Lei nº 10.406, de 10 de janeiro de 2002. Institui o Código Civil. Diário Oficial da União. 2002 Jan 11.
17. Melo ND. Da culpa e do risco: como fundamentos da responsabilidade civil. 2a ed. São Paulo: Atlas; 2012.
18. Brasil. Superior Tribunal de Justiça (3. Turma). Recurso Especial nº 1.097.955/MG. Relatora: Min. Nancy Andrighi. Diário da Justiça Eletrônico. 2011 Out 03.
19. Brasil. Decreto-Lei nº 2.848, de 7 de dezembro de 1940. Institui o Código Penal. Diário Oficial da União [Internet]. 1940 Dez 31.
20. Sampaio RC, Sabbatini M, Limongi R. Diretrizes para o uso ético e responsável da Inteligência Artificial Generativa: um guia prático para pesquisadores. São Paulo: Editora Intercom; 2024.
