Industry Voices in AI Regulation: A Structural Mismatch

Executive Summary

This study investigated how the industrial sector perceives and evaluates Bill No. 2,338/2023 (PL 2,338/2023), which proposes a regulatory framework for Artificial Intelligence (AI) in the country. Through a focus group with legal executives from large companies in the manufacturing industry and a documentary analysis of the public hearings held at the National Congress, the studied combined qualified listening

and empirical analysis to identify representation gaps, regulatory challenges and possible paths towards more balanced regulation.

Among the main findings, we highlight:

• The industry not only feels underrepresented, but is in fact absent from the regulatory debate – despite its economic weight. Just 2.9% of participants in public hearings of Congress represented the industrial sector, in contrast to their contribution of 24.7% to national GDP;
• Civil liability is the main source of uncertainty. The study identified widespread concern about the joint liability regime provided for in the PL, which reflects the Brazilian tradition of consumer protection, and which does not distinguish AI applications at intermediate stages in production chains – many

of them internal and without contact with the public. According to executives, this can lead to predatory litigation and increased compliance costs.

• Executives from the sector criticized the transposition of the European risk classification model, pointing out that it ignores the reality of an innovation ecosystem marked by scarcity of resources, creativity and productive diversity. For study participants, “copied and pasted” regulation tends to discourage experimentation and increase legal uncertainty.
• The project language is technical, but not very accessible. Focus group participants highlighted that the PL text is difficult to understand even for specialized professionals. The perception is that the law is born obsolete, without instruments adaptable to the speed of technological innovation.
• Regulation can impact industrial competitiveness by imposing potentially redundant obligations between different sectoral authorities, which can generate regulatory fragmentation and excessive encumbrance of production chains.

Introduction

Artificial Intelligence and Industry

In the early 2010s, German engineers and scientists began studying new approaches to industrial policy, identifying the first stirrings of a new industrial revolution. These ideas left the factory floor in 2015 when Klaus Schwab, one of the founders of the World Economic Forum (WEF), introduced the term “Fourth Industrial Revolution” in an article for the magazineForeign Affairs. Critics questioned the legitimacy of the concept, classifying the Industry 4.0 as a mere marketing strategy – the main argument was that, although there were significant transformations in specific sectors, there was still no

systemic change in the global industrial panorama, which was comparable to the impact that digitalization in the late 1990s had on the sector.

These questions have practically ceased to exist in recent years, as advances in machine learning and AI technologies have not only demonstrated transformative power, but have become the central gear of Industry 4.0.

The adoption of AI in industry, especially in the Brazilian manufacturing industry1, is experiencing a moment of acceleration. From 2022 to 2024, the use of AI jumped from 17% to 42% in the sector, according to the Semiannual Innovation Survey of the Brazilian Institute of Geography and Statistics (IBGE). In 2024, 41.9% of industrial companies with more than 100

employees implemented AI solutions representing a growth of 163% compared to 2022, when only 16.9% used this technology. This advance places Brazil above the global average for the use of AI in the manufacturing industry, which is 35% (Artsmart AI, 2024).

Despite growth, Brazilian industry still faces challenges in fully adopting advanced digital technologies. According to PINTEC Semestral 2024, the high costs of technological solutions continue to be the main obstacle, cited by 78.6% of

companies. The lack of qualified personnel comes in second place (54.2%), followed by security and privacy risks (47.2%). These challenges mean that the topic of AI is not only an issue in industrial parks, but also in the Central Plateau.

It is in this context that this study proposes to listen to the industry’s perspectives on PL 2,338/2023, which is being processed in the National Congress, and intends to regulate AI applications in Brazil. We want to understand and offer inputs for a regulatory design that can understand industrial policies and regulation of the use of new technologies as inseparable and inseparable themes for socioeconomic development.

• For the purposes of this study, the definition of transformation industryand sectoral classifications used by National Confederation of Industry (CNI), which comprises the set of activities aimed at the physical, chemical or biological transformation of raw materials into new products, ranging from basic sectors — such as metallurgy and chemistry — to high-technology segments, such as electronic and pharmaceutical equipment. This definition guides the empirical outline of the work and reflects the strategic importance of the manufacturing industry in the national productive structure.

Bill 2,338/23

For the purposes of this study, we understand AI technologies as software systems capable of processing and analyzing data through algorithms and mathematical models, employing statistical and computational techniques to identify patterns, make predictions or generate new content. In other studies published by Reglab, we distinguish Analytical AI systems, aimed at solving delimited problems, such as fraud detection or demand forecasting, and Generative AI systems, capable of creating texts, images or codes from large volumes of data. Unless there is a differentiation expressed in the text, this research understands both systems together, within the comprehensive concept of AI technologies.

In December 2024, the Plenary of the Federal Senate approved Bill 2,338/2023, which establishes a regulatory framework for AI in Brazil (Agência Senado, 2024). The approved text was the replacement for Senator Eduardo Eomes (PL-TO), drawn up based on the project presented in 2023 by the then President of the Senate, Rodrigo Pacheco (PSD-MG). The PL originates from the draft developed by the Commission of Jurists on Artificial Intelligence (CJUSBIA), in 2022, and was subsequently analyzed by the Internal Temporary Commission on AI in Brazil (CTIA), before being voted on in the Plenary (Federal Senate, 2024).

PL 2,338/2023 establishes an extensive set of measures to regulate the development, promotion and ethical and responsible use of AI, presenting itself as a regulatory framework for the entire AI ecosystem in Brazil. With the aim of protecting fundamental rights and guaranteeing the centrality of the human person in the face of the rise of AI in Brazil, PL 2,338/2023 has an extensive list of rights for affected users

by these technologies, including the right to information regarding interactions with AIs, explainability and contestation of decisions, and human review.

The text, inspired by the EU AI Act of the European Union (ITS, 2025), proposes a risk-based approach, assigning obligations depending on the level of criticality of the system and distinguishing between excessive-risk and high-risk AIs (Sagawa; Gonçalves, 2025). Excessive risk technologies, such as autonomous weapons or biometric identification remotely, in real time and in public spaces, are prohibited. On the other hand, high-risk systems, due to their potential for adverse impact on fundamental rights, such as autonomous vehicles in public spaces, diagnostic applications and medical procedures in healthcare and immigration management and border control mechanisms, are subject to specific obligations.

The legislative proposal also creates the National Artificial Intelligence Regulation and Governance System (SIA), coordinated by the National Data Protection Agency (ANPD) and integrated by sectoral authorities, the Permanent Council for AI Regulatory Cooperation (CRIA) and the Committee of AI Experts and Scientists (CECIA). This regulatory ecosystem aims to promote and guarantee cooperation in the implementation and compliance with the rules proposed by PL 2,338/2023, through ANPD supervision and the valorization of the regulatory, sanctioning and normative competencies of sectoral authorities.

While Brazil is still discussing the creation of its own regulatory framework for AI systems, other countries have already advanced in implementing specific standards on the topic. In the European Union, the EU AI Act, proposed in 2021 and approved in 2024, came into force in August of the same year (European Commission, 2024). The regulation establishes a risk-based regulation model, with gradual obligations depending on the potential impact of each AI system. In February 2025, the first provisions came into force, including the prohibition of systems classified as

of “unacceptable risk,” such as cognitive manipulation technologies, social scoring by governments, and real-time remote biometric identification in public spaces (Harvard University Information Technology, 2025). The remaining obligations will be implemented

in a staggered manner, according to the risk category of the activity (Haikal; Becker; Sotomayer, 2024).

In Peru, Law 31,814 had been approved in 2023, to promote the use of AI. However, only in September 2025 was its regulation approved, establishing a risk classification model and principles that should guide the collective and beneficial use of AI.

for society (Panez, 2025). Other countries, such as the United Kingdom and the United States, have to date adopted flexible regulatory approaches based on principles, strategies and guidelines, rather than formulating specific national laws for AI (White C Case, 2025). This option seeks to ensure greater adaptability in the face of the accelerated pace of innovation in AI by guiding the actions of public and private agents involved in its development and use chain.

Currently, PL 2,338/2023 is being processed in the Chamber of Deputies. As it is a transversal topic, involving matters within the competence of more than four permanent committees, the analysis takes place in a Special Committee installed in May 2025, chaired by deputy Luisa Canziani (PSD-PR) and reported by deputy Aguinaldo Ribeiro (PP-PB) (Câmara dos Deputados, 2025; Haje, 2025). With the closure of

discussions in the Special Committee, and the publication of the rapporteur’s opinion, PL 2,338/2023 must be voted on in the Plenary of the Chamber of Deputies (Kaufman, 2025), before returning to the Federal Senate, which must discuss the proposed changes in the review body. PL 2,338/2023, therefore, still must go through some stages before its potential approval in the National Congress.

Research Methodology: the Focus Group

Why listen to those affected by regulation? The question seems obvious, but it is rarely exercised in a meaningful way when dealing with the regulation of new technologies (Ribeiro; Ramos, 2025). Understanding how the industry views AI regulation requires more than mapping institutional positions or reviewing technical documents, but also create a space where perspectives not only express themselves, but

dialogue with each other. It is in this meeting that nuances, tensions and convergences emerge that would hardly emerge in isolated conversations.

O focus group is a qualitative methodology aimed at structured listening to perceptions and experiences based on interaction between participants with similar experiences around a common theme. The technique seeks to understand how meanings are constructed collectively, allowing the identification of practices, perceptions and behaviors that emerge from dialogue and interaction between participants (Oliveira et al., 2020).

In practice, the focus group works like a collective interview, in which the moderator presents the topic, encourages debate and ensures balanced participation, without interfering with personal opinions. The statements are analyzed qualitatively, taking into account both the content and the dynamics of interactions, which makes it possible to capture nuances and meanings that would be difficult to emerge in individual interviews.

In the context of this study, we chose to listen to representatives from legal and regulatory areas of companies in the manufacturing industry sector. Our objective was not to endorse these positions, but to understand their concerns, convergences and divergences about the impacts of PL 2,338/23 on the national regulatory and productive environment, in addition to perceptions about the participation of the productive sector in legislative debates on the topic.

The statements were treated as qualitative evidence of sectoral perception, analyzed critically and contextualized. We sought to identify how the productive sector interprets its role and influence in legislative debates on AI regulation, recognizing the limits and biases inherent to this approach.

Participant	Genre	Industry Sector
A B C D And F G H I	Man	Metallurgy
A B C D And F G H I	Woman	Consumer Goods
A B C D And F G H I	Woman	Metallurgy
A B C D And F G H I	Woman	Pharmachemicals and pharmaceuticals
A B C D And F G H I	Man	Pharmachemicals and pharmaceuticals
A B C D And F G H I	Man	Metallurgy
A B C D And F G H I	Woman	Consumer Goods
A B C D And F G H I	Woman	Metallurgy
A B C D And F G H I	Woman	Motor Vehicles

The relevance of the Industry perspective in the Regulatory Debate

The industry is a stakeholder central to the debate on AI regulation in Brazil. The National Confederation of Industry (CNI) recognized this importance by including PL 2,338/2023 among the 14 Bills defined as legislative priorities for 2025. Although the public debate on this bill already includes perspectives from technology companies (Conselho Digital, 2024), the health industry (Bezerra, 2025), the financial sector (FEBRABAN, 2024), agribusiness (ABAG, 2024), government authorities (ANPD, 2023), academia (IAEDU; NEES/UFAL,

2025) and civil society (Direitos na Rede, 2024), the transformation industry had not yet been heard in the same depth even by the legislative process itself (as we will see in section 3.1 below).

In other words, the segment was chosen intentionally: it is a sector that represents 14.4% of total GDP (58% of industrial GDP – CNI, 2025), directly affected by the regulation proposal and whose participation is essential to understand its broad regulatory and economic effects.

By bringing voices from the industry through a focus group, the study does not seek to overlay them with other legitimate perspectives, nor to generalize its findings, but rather to broaden the spectrum of the regulatory debate, recognizing that

the formulation of balanced policies depends on listening to all impacted agents. The objective is to understand a specific section of the regulatory debate, expanding the spectrum of evidence available for decisions based on evidence and plurality. Listening does not mean endorsing.

Results

Assessment of PL 2,338/23

Focus group participants expressed critical perceptions regarding the legislative process and the content of PL 2,338/2023, which proposes the regulation of AI in Brazil. In general, they expressed skepticism about the material quality and breadth of the public debate conducted to date. The following observations detail the main perceptions about the PL, its gaps and the potential impacts for the industry:

Legislative Overrun and Obsolescence: Participants shared the perception that PL 2,338/2023 is being conducted in a hurry and without the proper convening of the committees necessary for the formation of an in-depth understanding on the part of deputies and senators, characterizing a “legislative overrun”.

“Here, for me, legislative overrun. They didn’t create the commissions that should have been created, they did, but they didn’t call on society to understand if this made sense. For me, it didn’t.” [Participant E]2

Although not yet approved, participants pointed out that the regulation would already be “obsolete”. Participant I highlighted that the positivist model of law

in Brazil, combined with the speed of evolution of AI, tends to produce a permanent normative lag in which regulation will always be reacting to a changing reality, without being able to fully reflect the state of technology or its social and economic impacts.

“Basically, we started dealing with artificial intelligence about a year and a half ago in a more open way… and look where we have already arrived. So, like this… Unfortunately, we, within Brazilian law, because it is positivist, we will be forever chasing a situation that will never reflect reality. So, especially the way the regulation is being proposed […] how are we going to do a business that is at the same

safe and ample time so that we don’t fall behind?”

[Participant I]

Lack of National Contextualization: A central criticism from participants is that the text of PL 2,338/2023, especially when adopting a structure based on risk classification, replicates the EU AI Act model without considering the specificities of the Brazilian context.

• In order to preserve the anonymity and confidentiality of research participants, specific changes were made to the quotes presented in this study. In certain circumstances, specific linguistic adaptations were made to ensure the original intention of the interviewees in the textual transcription. The preservation of the discursive record was maintained whenever possible, respecting the established methodological principles.

According to the group, the proposal is based on a European regulatory logic, more rigid and guided by abstract norms, while Brazil has an innovation ecosystem marked by creative informality and scarcity of resources, environment in

that experimentation and improvisation are part of the dynamics of technological development. By not recognizing these structural differences, the project would tend to impose a regulatory framework that is foreign to the daily practice of the national economy, limiting the sector’s capacity for innovation and response.

“I think the law focuses very much on the technical aspect of classification, what is high risk, what is low risk, again, a European copy paste, without worrying not only about the industries that are behind

us represented, but also with the end users, we were talking a little, you know, about it not being a clear project, easy access, that people can read, or with other legislation”. [Participant E]

Inaccessibility and Language: The difficulty in understanding and interpreting the text of PL 2,338/2023 was highlighted as one of the main concerns among industry representatives. Participants highlighted that the proposed wording is excessively complex and, at the same time, not very technical – becoming

inaccessible to both the general public and IT and development teams. According to them, the project attempts to standardize technical aspects of the functioning of AI systems without using appropriate terminology, which compromises regulatory clarity and increases the risk of divergent interpretations in the application of the standard.

“We have the Brazil challenge, right? Because it’s also something for us to talk to the corporate population, lawyers and so on, we even have to think about the use of AI for the Brazilian population as a whole, from North to South. I think this care is also lacking in legislation.” [Participant C]

“And for me, one of the flaws that this law has is that, again, it is incomprehensible. If you give it to anyone who doesn’t have a degree in Law or who is curious about these things, the person reads this here and doesn’t understand anything.” [Participant I]

Legal uncertainty: Given the complexity and subjectivity of the proposed text, the participants assessed that the difficulty in translating its concepts and technical terms into business practice represents a significant obstacle to the business environment. This lack of clarity regarding regulatory operability tends to generate legal uncertainty and make it difficult for companies in the sector to implement compliance measures.

“If you read [PL 2,338/2023], it seems like they pasted a part of a legal text with another that doesn’t fit. So, I think that has a lot of impact. I think that having things clearer and more… and being talked about, helps a lot, thus, to have a conducive environment to encourage innovation. Without that, I think it’s very difficult.” [Participant D]

Priority Themes

In the focus group dynamics, participants analyzed the most worrying topics raised during the discussion and recognized that PL 2,338/2023 covers, to some extent, the main points of attention. However, they assessed that these topics were treated inadequately, without offering effective responses or even worsening the sector’s practical concerns. The four priority themes for participants were:

Civil Liability:The topic emerged as the area of ​​greatest concern and priority for improvement in PL 2,338/2023. The topic was highlighted during the prioritization exercise, indicating the perception that there is an urgency to promote greater clarity and balance in the distribution of responsibilities provided for in the text.

Participants linked to consumer goods industries noted that the bill focuses excessively on holding developers and suppliers of AI systems accountable, leaving the user conduct. It was mentioned, for example, the misuse of generative AI tools by consumers to create false content that could cause damage to the reputation of

products or brands, a situation that highlights gaps in the shared responsibility approach foreseen in the proposal.

“I have a base of consumers who are already using AI to file lawsuits against the brand saying that, for example, I don’t know, I’ll give an example here, there are insects in the product. But the images were generated by AI.” [Participant B]

In this context, participants expressed concern about the unbalanced nature of the liability regime provided for in PL 2,338/2023. They highlighted that, as AI systems go through continuous cycles of improvement and evolution, it becomes difficult to identify the specific origin of each failure or behavior. This scenario would favor

the excessive use of the right of return, directing liability mainly towards large companies in the production chain, even though they are not always directly responsible for the damages observed.

“When the problem appears, no one knows exactly where it came from. And someone will have to answer for it. Normally, it falls to those at the top of the chain — the larger companies. From there, the right of return downwards applies. The problem is that, often, we are unable to identify the real source of the failure, because AI was developed to improve and evolve. If this improvement occurs out of control, it becomes practically impossible to map

where the error arose — and even so, it continues to replicate itself at different points in the system.” [Participant I]

The lack of clarity regarding the operability of the proposed rules was also highlighted as a concern, mainly due to the difficulty in understanding how the provisions of PL 2,338/2023 could be applied in practice.

“So, I understand that the concerns must be with civil liability. There are several, several issues in the law that deal with civil liability. There is no exactly how we are going to achieve this liability.” [Participant E]

Risk Classification: The second most discussed topic was the risk classification system. Participants interpret the PL 2,338/2023 model as a transplantation of the European model,disregarding the country’s structural particularities, which could produce counterproductive effects. They highlighted that, unlike economies

more regulated, Brazil has an innovation ecosystem characterized by creativity in the face of scarcity of resources and the ability to adapt to less predictable contexts. In this sense, excessively rigid regulation could precisely limit the inventive and experimental dynamics that drive national innovation.

“Brazil […] has a society that is highly resourceful, full of innovation, people manage and innovate daily with very scarce and small resources. If the legislation is foreseeing this society, or it is thinking about a European society that is more rigid, in which people already have a culture of being bound by laws and norms in an abstract way, you can or cannot and that’s it. For me, this is the main problem and within the industry we are suffering from it, because it is what everyone brought here.” [Participant I]

In this sense, participants expressed strong concern about the range of risk categories provided for in PL 2,338/2023, warning that its generic application could compromise the operation and economic activity of entire sectors. They also highlighted the importance of distinguishing between the levels of risk associated with different contexts of AI use, reinforcing the consensus that the classification should be more granular and corresponding to the type of application.

“If it’s an AI made just for company use, something very specific. It’s going to be a type of regulation, there’s going to be some kind of… not a need, right, but some different perspective than an AI that’s practiced by a company, maybe it’s B2C. That’s more on a large external scale and not just internally.” [Participant C]

In this context, the Participant F drew attention to an operational aspect directly linked to the breadth of risk classification: the risk of regulatory overlap resulting from excessive autonomy of sectoral authorities. According to him, the possibility of each authority providing technical and specific aspects of AI applications in their respective regulated market can create a scenario of

regulatory fragmentation, in which companies that develop or launch AI solutions would face multiple and cumulative compliance procedures to enable the launch of a single product, increasing costs and inhibiting innovation.

“A large company will have to control each stage of its process: ‘Where’s the Anatel approval? And the Inmetro document? Anvisa authorization is still missing.’ There are so many requirements that monitoring becomes almost unfeasible.” [Participant F]

Responsible Innovation:Participants’ perceptions of responsible innovation reflect the need to balance encouraging innovation in industries with promoting legal certainty and ethics. For the group, technological development must incorporate principles of sustainability and responsibility from its conception. The topic received six votes, positioning itself as the third highest priority to be addressed by future regulation.

In this context, the Participant G highlighted the importance of incorporating structured steps of documentation and recording of tests carried out throughout the life cycle of AI systems, in order to ensure that technological development is safe and traceable from the initial phases of the project.

The importance of incorporating controls from the initial phases of the development of AI systems was reinforced, especially in companies that already have consolidated governance structures. Among the mechanisms mentioned are practices aimed at mitigating biases and reducing hallucinations in models, ensuring greater technical reliability and adherence to ethical standards. For the Participant H, this preventive approach must be accompanied by a long-term strategic vision, in which the value of innovation is balanced with legal and operational security, preventing failures in initial development from resulting in regulatory, reputational or financial losses in the future.

Explainability and Transparency: Although they recognized the importance of explainability and transparency mechanisms, participants noted that current technical and operational limits make it difficult to make these requirements compatible with the regulatory proposal. They highlighted that, although PL 2,338/2023 guarantees the user the right to an explanation about the decision, recommendation or prediction made by the application of AI, in practice, it may not always be possible to fully understand or review the technology’s decisions.

The topic of explainability is not new in the Brazilian legal debate. It had already been introduced by art. 20 of the LGPD, which recognizes the holder’s right to request review automated decisions and obtain clear information about the criteria used. However, even more than five years after its promulgation, this provision still lacks, according to the Participant E, of in-depth debate on its practical application. This gap highlights the complexity of the topic and reinforces the challenge of

operationalize the right to explainability also in the broader context of AI regulation proposed by PL 2,338/2023.

Although explainability and transparency are desired and PL 2,338/2023 addresses them, the lack of detail on how to achieve them in an environment where applications

AIs are autonomous and constantly improved, generating great insecurity and practical concerns for participants.

Intellectual Property (IP): The IP topic also received significant attention (4 votes), especially due to legal uncertainty regarding copyright and the use of content by AI applications. A Participant H drew attention to

addressing IP and copyright issues in PL 2,338/2023, understanding that

the text is “very dissonant”. How do you summarize Participant H, “the PL encourages lots of innovation, but at the same time, when the agent needs the legal security to follow, unable to proceed”.

Inclusion and Absence of Voices

There is a unanimous perception among focus group participants that the PL 2,338/2023 legislative process is not adequately listening to either society or the industrial sector. This was one of the few issues where there was not only agreement, but also comments from practically everyone and all participants.

It was interesting to note that some participants compared the PL 2,338/2023 with the LGPD legislative process, approved in 2018, stating that the perception was that the level of listening and engagement with the industry was substantially higher in the LGPD. The political component was also mentioned as a factor that interferes with this perception, in which participants have the view that “decisions end up being strongly influenced by political interests”.

It was also highlighted, at different times, the legislator’s difficulty in understanding the importance of the topic for the industrial sector. According to her, there is a prioritization of participation and hearing from technology companies in discussions on AI regulation and governance, to the detriment of traditional industry.

“It’s just that here in this case we are talking about a point that there is also a bias on the part of the legislator in relation to other interests. So, for example, you make cars, what is your interest in this? Does this affect you? And I think this is a problem that industries have in general because if you are not a technology company why am I going to listen to you.” [Participant I]

Some participants confirmed that they follow the legislative process of the PL 2,338/2023 through its government relations teams and associations and sector entities. The Institute for Retail Development – ​​IDV; the Brazilian Association of Credit Card and Services Companies – ABECs; the Brazilian Association of Supermarkets – ABRAS and the National Association of Motor Vehicle Manufacturers – ANFAVEA.

For participants, the absence of effective participation by the industrial sector in the legislative process can generate significant practical and macroeconomic consequences. They assessed that, if the PL 2,338/2023 is approved without due industry involvement, there is a risk of losing the transformative potential of AI for the sector.

“The biggest transformation of the industry will come from AI and digitalization — AI is the most powerful tool for this change. If the country does not guarantee freedom and security for its use, instead of attracting investments, it will end up crowding out innovation and competitiveness.” [Participant F]

Furthermore, participants warned that the lack of listening to the industry could result in the approval of legislation that is barely understandable, opening space for legal disputes and increasing legal uncertainty and the possibility of applying strict sanctions, without due legal clarity.

“If that’s the case, even based on the content of the norm, well, if I have a norm that I don’t understand the content of what is described here, okay, I won’t apply it and then we will deal with future problems: judicialization, legal uncertainty”. [Participant E]

“Imagine the supervisory body that will apply this and impose million-dollar fines on industries, right? So, we could also have a problem at the front of applying million-dollar fines without a clear rule, without the necessary legal certainty, that no one knows what and for what.” [Participant B]

Finally, participants also warned about the possible macroeconomic and operational effects that the approval of the PL 2,338/2023 can generate on business dynamics. They pointed out that, by imposing requirements that can slow down processes and restrict companies’ agility, regulation can directly affect corporate goals, production flows and delivery models currently consolidated in the industry.

In the group’s view, the impact goes beyond regulatory compliance: it is a systemic effect, with potential repercussions on the competitiveness and efficiency of companies’ day-to-day operations.

“And if we raise a question of speed, a question of level of delivery, if we introduce legislation that will also slow down a lot, we will have a problem with the company’s goals, with the way the company operates today. So it will have a gigantic macro effect, which I think when you look at the law by itself, you say, wow, that’s it. But when you look at the macro in day-to-day practice, we will have very… relevant, competitive effects, but in the day-to-day running of the company same.” [Participant C]

Suggestions and Referrals

During the focus group dynamics, participants shared their perspectives on possible solutions and directions to improve the text of the PL 2,338/2023, as well as to strengthen the effective and qualified participation of the industry in the legislative process.

Participants highlighted the importance of future AI regulation being guided by structuring principles capable of guiding the interpretation and application of the norm in a manner consistent with constitutional values. They highlighted that, faced with a rapidly evolving technological scenario, an excessively detailed and prescriptive law, as they perceive in the current text of the PL 2,338/2023, tends to quickly become obsolete and restrictive. The incorporation of broad and guiding principles, on the other hand, would allow regulation to remain relevant and balanced, accompanying transformations in the technological and industrial ecosystem.

“So, when you are going to develop a law, I think, as my colleagues said, it is a more principled issue. You have to have principles very much based on what is, for example, in the Federal Constitution, of protection, privacy, etc., etc. Because if you don’t have a guide in the construction and use of it, I think it becomes a very powerful tool for misuse.” [Participant A]

Given the complexity of the proposed text, participants also highlighted the importance of simplifying and making the language of the standard more accessible, without renouncing technical rigor. They argue that the use of terminology harmonized with already consolidated standards is essential to avoid anachronisms and reduce the risk of regulatory rigidity over time.

“The standard is very little explored, the technical standard. And it is not a regulation, it is a consensus of society. It should be used much more. In the area of information security, it is very common to follow a technical standard. And the regulation, it cannot enter this area, it should not enter, otherwise it will get stuck, this problem will happen. So, there has to be this standardization, which is that everyone can speak the same language, everyone can walk in the same way.” [Participant F]

With regard to ensuring effective and qualified participation of the industry in the legislative process of PL 2,338/2023, participants highlighted the importance of expanding listening and including stakeholders who do not traditionally belong to the technology sector. They highlighted that the active involvement of users, developers and representatives of other economic segments is essential to promote a more plural debate, capable of reflecting the diversity of realities and challenges associated with the application of AI in the country.

“So, call for conversations and debates, especially on this that will affect everyone, and for technologies, yes, but the industry too, just as you need to call, companies that often don’t make any sense initially, but they are living there because they use AI within their process. So let’s involve developers, users, when we think about industries and businesses and the population as a whole.” [Participant I]

Participants also highlighted the importance of define clear premises for the actions of industry associations with sectoral regulators, in order to strengthen dialogue with regulators and promote more cohesive and aligned regulation between different sectors. They assessed that the current text of PL 2,338/2023 grants excessive autonomy to sectoral authorities, which can result in disparate interpretations and requirements, creating a fragmented environment.

“I think one thing that the legislation is putting in place is leaving it very open for sector regulators to do what they want. Therefore, we could put the entities’ voice in the premises, make it clearer what they want…” [Participant F]

Furthermore, participants mentioned the importance of carrying out Regulatory Impact Analysis (AIR) as a prior step to implementing AI standards. This instrument, commonly used by regulatory agencies, allows evaluating the economic and social relevance of proposals, comparing alternatives and measuring their possible effects (CNI, 2021). In the group’s view, the systematic adoption of AIRs would contribute to supporting regulatory decisions that are more balanced, efficient and in line with the sector’s reality.

“There must also be a regulatory impact analysis when you are going to make a change, legislation of this type. There must be a culture of carrying out a regulatory impact analysis, including from an economic perspective, to see if it makes sense for society.” [Participant A]

Analysis and Comments

Faced with widely shared perceptions, especially in qualitative research, it is necessary to adopt a critical look and compare reports with other data and methods to assess their consistency and formulate exploratory hypotheses. This is because apparent consensus often arises from imbalances in visibility – and not from a faithful representation of the facts.

In other words: Are the perceptions of the focus group disconnected from the facts, or are they supported by objective data? We decided to analyze two specific points

– exactly those on which there was most consensus: the perceptions of listening and participation in the legislative process of PL 2.338/23 and the emphasis on concern about the topic of civil liability. In both cases, Our research showed that this convergence is directly anchored in evidence from the legislative process itself and the text of the PL 2,338/2023.

Listening Perceptions and Public Hearings

As we showed in item 2.3, focus group participants realize that the public debate on AI regulation has privileged the voice of technology companies and sectors already consolidated on the digital agenda, leaving the industry in a secondary position, despite being directly impacted by the adoption and regulation of these technologies.

During the processing of PL 2,338/2023 in the National Congress, several public hearings were held to gather different perspectives on the regulation of artificial intelligence in Brazil. In the Federal Senate, the Temporary Committee on Artificial Intelligence (CTIA) held 14 sessions between October 2023 and September 2024 (Federal Senate, 2024). In the Chamber of Deputies, the Special Committee responsible for analyzing the project conducted 12 public hearings between June and September 2025 (Commission Special on AI, 2025).

Analysis of participants in public hearings promoted by the National Congress indicates that only 1.1% in the Chamber of Deputies and 4.7% of guests in the Federal Senate directly represented the interests of the industry. When analyzing the total distribution of participants in the two houses of the National Congress, we found that sOnly 2.9% of guests at public hearings represented industry interests.

This data contrasts with the economic weight of the industrial sector, responsible for 24.7% of national GDP and 21% of formal jobs (CNI, 2025). The contrast remains when looking at the manufacturing industry, which represents 14.4% of GDP and 14.3% of the country’s formal jobs (CNI, 2025).

The analysis of the full transcripts of the Senate and Chamber public hearings showed that the references to the industry were punctual and, in most cases,

sometimes anecdotal or generic. Even when mentioned by experts who did not directly represent the sector, the term was often used in a

abstract or to refer to service sectors (e.g. “software industry” or “cultural industry”), without reference to the manufacturing industry or the industrial sector itself.

Among the substantive mentions of the industry, there is a small but coherent set of interventions, which recognize both the sector’s dependence on AI and the gap between the regulatory debate and the real conditions of Brazilian industry. They also drew attention to the need for requalification and adaptation of industrial chains in the face of technological transition, emphasizing that

regulation must “focus on high-risk uses and avoid obstacles to innovation” (CNI), ensuring legal certainty and proportionality.

These statements also revealed a recurring argumentative pattern – also present in the focus group: the attempt to reposition the industry as an active and legitimate subject of the AI debate, as opposed to the dominant narrative centered on digital platforms

and protection of personal data. The industrial discourse mobilized through categories such as promotion, diffusion and regulatory proportionality, structuring itself in the search for symmetry and regulatory realism.

So, thinking about a predictive manufacturing system, within an industrial plant, the burden of regulatory impact analysis, the burden of self-explanation that this model has to have, will not be able to be the same burden, obviously, of a public system to say whether it is that or another citizen who is eligible to receive a benefit from the Brazilian Government. (MDIC)

What I want to bring with this is that technologies, or systems, are transversal. I can apply the same fatigue identification technology to other situations, which may be lower or higher risk. (…) These technologies permeate any type of sector and application (…) This is essential for neo-industrialization. We want to modernize the industry (CNI)

Civil Liability and the Brazilian Reality

As already highlighted in this study, focus group participants perceived civil liability as a topic of greatest concern in PL 2,338/2023. At different times, this and other themes were contrasted with the AI Act of the European Union,

which was the explicit and implicit reference to foreign regulation practically present throughout the focus group.

However, we believe that there is a difference between the two that can explain, albeit partially, why civil liability stands out as a source of concern for different industry representatives. These clues emerge from the study prepared by the Institute of Technology and Society of Rio de Janeiro, (ITS/RJ), entitled “Report Comparative matrix of obligations: PL 2,338/2023 vs. I AI act”and updated in June 2025.

In this preliminary research, ITS/RJ concludes that PL 2,338/2023 is more extensive than the AI Act in terms of the number of obligations: there are 68 in the Brazilian proposal compared to 43 in the European Union.

Furthermore, PL 2,338/2023 implements a more horizontal approach, establishing more shared obligations: 34 of the 68 total obligations with joint and several liability between the three main agents (supplier, applicator and distributor). The European regulatory framework follows a more specific strategy, assigning responsibilities

to specific actors. Even so, of the 43 obligations, only 9 are shared among three or more agents in the supply chain.

Obligations in PL 2,338/23

Obligations under the EU AI Act

A joint liability reflects a structural characteristic of Brazilian law: the historical emphasis on consumer protection as the axis of regulation and

civil liability. By adopting a model that favors the protection of the final recipient, it appears that PL 2,338/2023 aligns with the tradition of the Consumer Protection Code, which helps to explain the concern of industry representatives, which operates complex chains and shares technological inputs with multiple agents.

At the same time, it is important to recognize that this approach can potentiallyignore the diversity of uses of AI in industry, many of them completely different from a traditional consumer relationship. AI systems applied to manufacturing processes, quality control, energy monitoring or logistics management do not involve direct interaction with the end consumer, but could be covered by liability regimes equivalent to those of public-facing applications. This lack of

distinction increases the risk of predatory litigation, in which broad interpretations of the law are explored in opportunistic legal demands.

Both hypotheses, expressed in the perceptions of the focus group, show that this concern presents a legitimate warning that deserves specific studies to understand how the wording of the PL can, unintentionally, create uncertainty in compliance processes and incentives for excessive judicialization.

Conclusion

The results of this research reveal a gap between the advancement of the regulatory agenda in AI and the effective participation of the industry sector in this process. Although PL 2,338/2023 represents an important milestone for the consolidation of a legal environment focused on responsible innovation, its current formulation reflects the predominance of models

and foreign references, which are not very sensitive to productive diversity and national technological reality.

The perceptions collected in the focus group indicate that the manufacturing industry recognizes the value of ethical and rights-oriented regulation, but expresses legitimate concerns regarding the clarity of obligations, the operationalization of rules and the compatibility between new requirements and the dynamics of production chains.

Issues such as civil liability, comprehensive risk classification system, overlapping regulatory powers and complexity of the legislative text appear as central challenges for the implementation of the PL.

More than resistance to regulation, the voices heard in this study point to the need for more dialogical, predictable and proportional regulatory governance to the real risk of industrial AI applications. The absence of structured channels for listening to the productive sector not only weakens the legislative process and normative construction, but also reduces the capacity of regulation to generate gains in competitiveness and legal security.

In this sense, the study reinforces that AI regulation and industrial policy should not be treated as separate agendas. The convergence between the two is an essential condition for Brazil to build a regulatory model that combines protection of rights, stimulation of innovation and strengthening of the production base. Ensuring the industry’s qualified presence in decision-making forums, incorporating regulatory impact analysis practices and adopting accessible regulatory language are decisive steps towards this alignment.

Listening to representatives of the transformation industry, Reglab seeks to contribute to a more plural, technical and evidence-based debate. The regulation of AI will be more effective the more it reflects the complexity and diversity of uses of these technologies in the national economic fabric. This policy brief It is, therefore, an invitation to continue this dialogue.

Direction for Future Studies

This study used a focus group methodology, bringing together senior legal representatives from companies in the manufacturing industry to understand their perceptions about Bill 2,338/2023 and the legislative process that accompanies it. The central objective was to give protagonism to a sector with low effective representation in discussions about AI regulation, promoting a space for qualified listening and collective reflection on the potential impacts of the standard.

Although the focus group made it possible to identify trends, consensus and strategic concerns from a legal-regulatory point of view, the results do not exhaust the

debate. There is significant scope for further study for future studies, which can contribute to a more comprehensive understanding of the relationship between innovation, governance and regulation of AI in the Brazilian industrial context.

Among the possible directions for future studies, the following stand out:

• Perceptions beyond legal areas: The present study focused on professionals in the areas of control, especially law and regulation. Future research could include innovation, marketing, operations and technology managers, to understand how AI regulation is perceived and incorporated into business routine from different organizational perspectives.
• Normative analysis and legal effects: This study analyzed participants’ perceptions, not having looked extensively at the legal content of PL 2,338/2023. Future work can deepen the analysis of the normative provisions of the proposed regulation, investigating how each obligation can legally impact the industry on topics such as civil liability, risk classification and governance of AI systems.
• Regulatory impacts and sectoral competitiveness: Future research with a multidisciplinary approach can conduct specific regulatory impact analyzes for the Manufacturing Industry, estimating compliance costs, entry barriers and effects on competitiveness and innovation resulting from the implementation of the future law.
• Institutional capabilities and regulatory coordination: Considering the design of the National AI Regulation and Governance System (SIA) provided for in the PL, future research can examine the regulatory capacity of sectoral authorities to act in this new ecosystem, evaluating gaps in coordination, technical expertise and institutional resources.
• Multisectoral expansion of perspectives: The present study focused on the perceptions of professionals in the legal and regulatory areas of the manufacturing industry. Future research may include government representatives,

civil society, business associations, startups and research entities, to capture convergences and divergences between different sectors and actors in the AI regulatory ecosystem, expanding the analytical and comparative plurality of results.

Together, these directions reinforce the need to deepen dialogue between industry, regulators and academia, consolidating an empirical and legal basis that supports the development of ethical, adaptable and competitiveness-promoting AI regulation.

References

ABAG. Copyright rules and Brazil’s competitiveness in AI. 2024. Available at:https://abag.com.br/regras-balanced-copyright-and-competitiveness-of-brazil-in-artificial-intelligence/?utm_source=chatgpt.com.

Accessed on: 08 Oct. 2025.

SENATE AGENCY. Senate approves regulation of artificial intelligence; text goes to the Chamber. Available at: https:// www12.senado.leg.br/noticias/materias/2024/12/10/senado-aprova-regulamentacao-da-inteligencia-artificial-texto-vai-a-chamber. Accessed on: 02 Oct. 2025.

ARTSMART AI. AI in the Manufacturing Statistics: Comprehensive Insights into Adoption, Impact, and Future Trends. 2024. Available at: https://artsmart.ai/blog/ai-in-the-manufacturing-statistics/. Accessed on: 07 Oct. 2025.

National Data Protection Authority. Preliminary analysis of Bill No. 2338/2023, which provides for the use of Artificial Intelligence. Available at: https://www.gov.br/anpd/pt-br/assuntos/noticias/analise-preliminar-do-pl-2338_2023-formatado-ascom.pdf?utm_source=chatgpt.com. Accessed on: 07 Oct. 2025.

BEZERRA, Bruno Boldrin. Regulation of AI in Healthcare, long awaited by the sector, must be balanced so as not to compromise the implementation of new technologies that improve diagnoses and treatments. 2025. Available at:https://abiis.org.br/regulamentacao-da-ia-na-saude-tao-aguardada-pelo-setor-deve-ser-equilibrada-to-not-compromise-the-implementation-of-new-technologies-that-improve-diagnosis-and-treatments/?utm_ source=chatgpt.com. Accessed on: 08 Oct. 2025.

CHAMBER OF DEPUTIES. Chamber installs commission to debate project on the use of artificial intelligence in Brazil. 2025. Available at: https://www.camara.leg.br/noticias/1159111-camara-instala-comissao-para-debater-projeto-sobre. Accessed on: 03 Oct. 2025.

CHAMBER OF DEPUTIES. Special Commission on Artificial Intelligence (PL 2338/23). 2025. Available at: https:// www2.camara.leg.br/atividade-legislativa/comissoes/comissoes-temporarias/especiais/57a-legislatura/comissao-special-about-artificial-intelligence-pl-2338-23/other-documents/summaries-of-public-hearings/summary-of-public hearings. Accessed on: 16 Oct. 2025.

RIGHTS COALITION ON THE NETWORK. Regulate to promote responsible and rights-protective AI: warnings about setbacks, threats and guarantees of rights in PL nº 2,338/23. 2024. Available at: https://direitosnarede.org. br/2024/11/29/regular-to-promote-a-responsible-and-protective-of-rights-alerts-about-setbacks-threats-and-guarantees-of-rights-no-pl-no-2-338-23/. Accessed on: 07 Oct. 2025.

EUROPEAN COMMISSION. AI Act enters into force. 2024. Available at: https://commission.europa.eu/news-and-media/ news/ai-act-enters-force-2024-08-01_en. Accessed on: 03 Oct. 2025.

NATIONAL INDUSTRY CONFEDERATION. The importance of the Manufacturing Industry for Brazil. Available at: https://static.portaldaindustria.com.br/media/filer_public/da/40/da400886-16b5-4d4e-9681-00e85d555137/flyer_a_ importancia_da_industria_no_brasil_transformacao_marco2025.pdf. Accessed on: 07 Oct. 2025.

NATIONAL INDUSTRY CONFEDERATION. The importance of Industry for Brazil. 2025. Available at: https://static. portaldaindustria.com.br/media/filer_public/b2/70/b27009a8-5332-4203-9c73-b1a523d1f56b/flyer_a_importancia_da_ industria_no_brasil_geral_marco2025.pdf?utm_source=chatgpt.com. Accessed on: 07 Oct. 2025.

NATIONAL INDUSTRY CONFEDERATION. Regulatory Impact Analysis: Greater Efficiency for Regulatory Agencies’ Decisions. 2021. Available at: https://static.portaldaindustria.com.br/media/filer_public/c8/7d/c87daae4-405b-482b-803f-5c320f3a6f80/cni_-_analise_de_impacto_regulatorio_web_-_2021.pdf. Accessed on: 16 Oct. 2025.

DIGITAL ADVICE. 36 entities, including the Digital Council, sign a letter with guidelines for the AI Regulatory Framework in Brazil. 2024. Available at: https://conselhodigital.org.br/2024/04/36-entidades-incluindo-o-council-digital-sign-letter-with-guidelines-for-the-regulatory-landmark-of-ia-in-brazil/?utm_source=chatgpt.com. Accessed on: 13 Oct. 2025.

BRAZILIAN FEDERATION OF BANKS. Legal Framework for Artificial Intelligence. 2024. Available at: https://portal.febraban. org.br/noticia/4229/en-us?utm_source=chatgpt.com. Accessed on: 13 Oct. 2025.

HAIKAL, Beatriz; BECKER, Daniel; SOTOMAYOR, Gabriela. Some notes on the European Union’s AI Act. 2024. Available at: https://www.jota.info/opiniao-e-analise/artigos/alguns-apontamentos-sobre-o-ai-act-da-uniao-europeia. Accessed on: 03 Oct. 2025.

HAJE, Lara. Commission that will analyze project on artificial intelligence will be chaired by Luisa Canziani; Aguinaldo Ribeiro will be rapporteur. 2025. Available at: https://www.camara.leg.br/noticias/1160383-comissao-que-vai-analisar-project-about. Accessed on: 03 Oct. 2025.

HARVARD UNIVERSITY INFORMATION TECHNOLOGY. EU AI Act Prohibited Use Cases. 2025. Available at: https://www. huit.harvard.edu/eu-ai-act#:~:text=As%20of%20February%202%2C%202025,materially%20distort%20a%20person’s%20 behavior. Accessed on: 03 Oct. 2025.

Artificial Intelligence and Education Initiative (Iaedu); Center for Excellence in Social Technologies (Nees/Ufal). Technical Note nº 2: What impact on Basic Education if PL 2,338/2023 were approved today. Maceió: Federal University of Alagoas, April. 2025. Available at: https://iaedu.nees.ufal.br/wp-content/uploads/2025/04/NT-2-Qual-impacto-na-

Educacao-Basica-se-o-PL-2338_2023-fosse-aprovado-hoje.pdf. Accessed on: 7 Oct. 2025.

INSTITUTE OF TECHNOLOGY C SOCIEDADE DO RIO (ITS). Comparative matrix of obligations: PL 2338/2023 vs. I AI act. Available at: https://itsrio.org/wp-content/uploads/2017/01/2025.06.23-Matriz-comparada-de-obrigacoes_-PL-2338_2023-vs.-EU-AI-act_Relatorio_AI-16-jun-1.pdf. Accessed on: 06 Oct. 2025.

KAUFMAN, Dora. Chamber concludes phase of public hearings on AI. 2025. Available at: https://valor.globo.com/ opiniao/coluna/camara-conclui-fase-de-audiencias-publicas-sobre-ia.ghtml. Accessed on: 06 Oct. 2025.

LOPES, Magno; BELLO, Luiz.From 2022 to 2024, the percentage of industrial companies using Artificial Intelligence rose from 16.e% to 41.e%. 2025. Available at: https://agenciadenoticias.ibge.gov.br/agencia-noticias/2012-agencia-de-noticias/noticias/44551-from-2022-to-2024-percentage-of-industrial-companies-using-artificial-intelligence-rose-from-16-9-para-41-9?utm_source=chatgpt.com. Accessed on: 07 Oct. 2025.

MARKET RESEARCH FUTURE. Brazil Industrial AI Market. 2025. Available at: https://www.marketresearchfuture.com/ reports/brazil-industrial-ai-market-46544. Accessed on: 13 Oct. 2025.

OLIVEIRA, Guilherme Saramago de et al. Focus Group: A Data Collection Technique in Qualitative Research? Fucamp notebooks, [S.I.], v. 19, no. 41, p. 1-13, Oct. 2020. Available at: https://revistas.fucamp.edu.br/index.php/ notebooks/article/view/2208. Accessed on: 16 Oct. 2025.

PANEZ, Óscar Montezuma. Peru takes its first steps toward AI regulation. 2025. Available at:https://chambers.com/ articles/peru-takes-its-first-steps-toward-ai-regulation. Accessed on: 03 Oct. 2025.

RIBEIRO, Natália Góis; RAMOS, Pedro Henrique. It is necessary to consider the perspective of those impacted by digital regulation. Available at: https://www.jota.info/opiniao-e-analise/artigos/e-preciso-considerar-a-perspectiva-dos-impactados-na-digital-regulation. Accessed on: 03 Oct. 2025.

SAGAWA, Karen Yui; GONÇALVES, Gabriella Almeida de Oliveira. AI, platforms and elections: a comparative perspective between PL 2338 and the AI Act. 2025. Available at: https://www.jota.info/opiniao-e-analise/artigos/ia-plataformas-e-eleicoes-a-perspective-compared-between-pl-2338-and-ai-act. Accessed on: 03 Oct. 2025.

FEDERAL SENATE. CTIA Public Hearings. 2024. Available at: https://legis.senado.leg.br/atividade/comissoes/ commission/2629/audiencias?p=1. Accessed on: 03 Oct. 2025.

WHITE C CASE. AI Watch: Global regulatory tracker. 2025. Available at:https://www.whitecase.com/insight-our-thinking/ ai-watch-global-regulatory-tracker-united-states. Accessed on: 03 Oct. 2025.

YEUNG, Luciana. “The Greatest Justice in the World”: would Coase be impressed? 2024. Available at: https://veja.abril.com. br/coluna/direito-e-economia/a-maior-justica-do-mundo-estaria-coase-impressado/?utm_source=chatgpt.com#google_ vignette. Accessed on: 08 Oct. 2025.

Reglab Methodology Annex

Title	Industry Voices in AI Regulation: A Sectoral Analysis of PL 2,338/2023
Question search	How do legal industry executives evaluate PL 2338/2023 and the inclusion (or absence) of the sector’s voices in the legislative process?
Summary of methodology	This research adopts a qualitative and exploratory approach, based on a focus group with senior legal representatives from the Manufacturing Industry. Data collection aimed to capture perceptions and interpretations about Bill 2,338/2023 and its legislative process, based on participants’ practical experience with topics of artificial intelligence regulation and governance. Data analysis was conducted using the reflective thematic analysis technique, with initial manual coding carried out by the research team and subsequent cross-checking with the support of NotebookLM software, used to check consistency of the categories and identify possible interpretative gaps. The emerging categories were subsequently consolidated and validated based on the full empirical corpus, ensuring analytical coherence and methodological transparency.
Data collection	The research used the focus group methodology (Oliveira et al., 2020), through a qualitative session, conducted based on a previously structured script. The choice of this method aimed to create a collective environment for dialogue, favoring exchanges between peers and the shared construction of perceptions on a common topic: the participants’ perception of PL 2,338/2023 and its legislative process. This format enabled the emergence and interaction between different perspectives and more in-depth analyses, which would be difficult to emerge in individual interviews with legal executives in the sector. The sample was composed following criteria of diversity and representation, including: participation of women and representatives from different sectors of the manufacturing industry. The selection of participants combined active search on LinkedIn as the main strategy, complemented by convenience sampling. Of the 34 people contacted, 9 agreed to participate in the research, while 8 reported unavailability and 16 did not respond to the invitation. The focus group was held on September 23, 2025, in a face-to-face format, lasting approximately two hours. The dynamics was conducted by the Reglab research team and recorded in audio, with the express authorization of the participants. The recordings were fully transcribed and accompanied by analytical memos prepared by the researchers present, in order to capture contextual impressions and non-verbal elements of the interaction. All material was subsequently stored and encoded in the Atlas.ti software, with the participants’ names being anonymized to ensure the confidentiality of the information collected. The focus group dynamics were conducted based on a semi-structured script, combining moments of open debate and structured collective prioritization exercises. The session It began with broad engagement questions, aimed at capturing the participants’ general perception of the topic, followed by visual systematizations using sticky notes to map the main impacts of PL 2,338/2023 on the Manufacturing Industry. Afterwards, moderation promoted rounds of discussion and collective reflection, encouraging dialogue between participants and clarification of doubts. As part of the methodology, the Nominal eroup Technique (NGT) was incorporated — a structured method of collecting and prioritizing ideas, which combines individual generation of contributions with collective discussion and voting (Delbecq C Van de Ven, 1971). This technique, widely used in applied research, aims to reduce biases such as speech dominance or hierarchical influence, ensuring balance in participation and structured consensus around the topics debated. Using this technique, it was possible to identify and prioritize the points of PL 2,338/2023 considered most critical by the industry, at the same time as capturing divergent perceptions and complementarities between participants, strengthening the qualitative robustness of the analysis. After the NGT dynamic, a voting simulation was conducted in a board format, inspired by corporate decision-making processes. At this stage, each participant received a fixed number of votes to distribute among the topics listed in the previous dynamics, according to the degree of relevance attributed to each one. The objective was to prioritize the topics perceived as most critical for the sector, allowing us to observe the relative weight of concerns.

Data analysis

Data analysis followed the reflective thematic analysis technique (Braun; Clarke, 2006), recognized for its suitability for exploratory qualitative studies in highly complex contexts. This approach favors contextualized interpretation rather than exhaustive coding, allowing the use of open and iterative analytical strategies. The material from the focus group was fully transcribed and processed in the Atlas.ti software. The initial coding was carried out manually by a team researcher, and was later reviewed by two other researchers to ensure consistency and interpretative reliability. In the next stage, visual software tools were used — such as concept clouds, thematic maps and correlation graphs — to identify competition patterns and relationships between codes. This process resulted in the emergence of central themes, which were tested, discussed and validated collectively by the research team, ensuring their adherence to the original empirical corpus. The analysis stage was conducted between September 23 and October 3, 2025.

Classification of participants in public hearings of the National Congress

To understand the diversity of voices present in the legislative debate on the regulation of AI, the 26 public hearing sessions held in the Federal Senate and the Chamber of Deputies were analyzed. In the Federal Senate, the CTIA held 14 sessions between October 2023 and September 2024. In the Chamber of Deputies, the Special Committee responsible for analyzing Bill No. 2,338/2023 conducted 12 public hearings between June and September 2025. Participants in these hearings were classified into six analytical categories, based on the institutional nature of the entity represented and the interests formally associated with their participation. Civil Society – Includes representatives of non-governmental organizations, associations and collectives that represent the interests of the community in general. Academy – Covers researchers, teachers and representatives of teaching and research institutions. Sector – Industry – Brings together entities representing the transformation industry and strategic industrial segments, which directly express the interests of the production sector. Sector – Artistic – Includes artists and creators. Business Sector – Includes companies, business associations and sectoral federations focused on technology and services. Government Sector – Includes public authorities, representatives of government bodies and entities. Each participant was counted only once, according to the predominant category of institutional representation, considering the entity indicated in the official call for participation in the public hearing. Thus, the same individual was not simultaneously classified into more than one category, even if they had multiple professional affiliations or hybrid trajectories. This methodological choice sought to guarantee classificatory coherence and avoid overlapping records. To ensure consistency and minimize bias in the classification, an independent double-check procedure was adopted, in which a researcher’s work was fully reviewed by another team member. Any divergences identified were discussed and consolidated with the mediation of a third researcher, ensuring greater reliability to the categories assigned and reinforcing the study’s methodological commitment to replicability and analytical credibility. As an additional procedure to minimize bias, shorthand notes from the Chamber of Deputies and Senate hearings were analyzed using the Atlas.ti software, using the tools (i) text search (term “industry” and its morphological variations), (ii) concepts and (iii) conversational AI. The objective was to identify any mentions of the industrial sector made by other actors that were relevant to the object of the research.

Bias reduction procedures

Consolidated theoretical-methodological references: the data collection and analysis techniques adopted in this study followed practices recognized in academic literature. The methodological approach was discussed internally before and after the focus group session, allowing the incorporation of criticisms and suggestions into the final research design, before the analysis process began. Complementary checking tool: although the initial coding of the data was conducted and triangulated by the research team, a second analytical support software (NotebookLM) was used as a cross-checking tool. The feature it was used by researchers who participated directly in the focus group, with the aim of validating the consistency of the identified categories and detecting any interpretative gaps that could have gone unnoticed in the initial coding. Triangulation of methods: in the analysis and comments section, the empirical findings were contrasted with documentary analysis of secondary sources, with the aim of comparing, validating and reinforcing the consistency of the interpretations constructed from the interviews. These references, when used, were expressly cited throughout the text. Independent double analysis: two researchers reviewed the set of codes and themes cross-referenced, reducing individual biases. The final definition of the themes was carried out in a collective discussion between the two authors, ensuring multiple perspectives and control of individual biases in the interpretation of the data. Methodological registration and transparency: all stages of the analytical process were documented, including successive versions of the writing files. This practice allows traceability of the methodological path, in accordance with Reglab guidelines for transparency and replicability.

Other Methodological Limitations

Dependence on external tools: part of the analytical process depended on the use and performance of proprietary software, which may limit replicability in different contexts. Qualitative scope and exploratory nature: the findings of this study derive from a single focus group conducted with legal representatives from the Manufacturing Industry. The discussions offer analytical depth and interpretative richness, but do not claim statistical representation. Sampling strategy and scope of participants: the selection of participants was carried out for convenience, which may have reflected biases in availability and professional network, even though sectoral diversity and business size criteria were adopted. Part of the sample was also made up of participants invited through a direct approach on professional networks, such as LinkedIn, which expanded the reach of the invitation, but maintained dependence on voluntary response and availability to participate. These conditions limited the expansion of the number of guests, making it possible to hold a focus group session. Time frame and legislative update: the conclusions of this study reflect the state of the legislative and regulatory debate in force at the time of the focus group, taking as a reference the text of PL 2,338/2023 sent to the Chamber of Deputies. Any subsequent changes to the wording of the project or the progress of the legislative process were not considered in the analyzes and interpretations presented here and may, therefore, affect the applicability or relevance of some analyzes and comments.

	Software	Description
	MS Office Suite	Editing text, spreadsheets and graphics
	Adobe C Suite	Layout and finalization of graphics and illustrations
	Atlas.ti	Organization, coding and analysis of qualitative data
	Cockatoo	Audio transcription of focus group dynamics into text
Software use	Chat9PT 5th	Brainstorm, information systematization, grammar review (spelling, synonym search grammar), language adaptation, adaptation to the Reglab Writing Manual
	Notion AI	Organization of research and structuring of schedule
	Lex.page	Text review (brevity, clichés, readability, passive voice, statements without evidence, repetitions)
	More UFSC	Generation of bibliographic references in the ABNT model
Ethical Guidelines	Research funding: this publication is part of a series of research sponsored by Google, Meta and b/luz. Although it is a contracted study, Reglab maintained full editorial and methodological control over the project, with autonomous definition of the methodology, analysis of results and writing of this research report. The authors have preserved total professional independence and assume full responsibility for the content and conclusions presented. Processing of personal data: the research involved the processing of personal data only in the collection and analysis stages, in a limited manner and proportional to the objectives of the study, in accordance with Law 13,709/2018 (LGPD). Legal basis: all participants formally authorized their participation by signing a consent form, with knowledge of the research objectives and the use of data. Purpose and suitability: the data were used exclusively for the purposes of this research, in accordance with the consent obtained, and were not used for other purposes. Minimization and anonymization: personally identifiable information that was not relevant to the study objectives was anonymized in the transcripts and excluded from the active database. Secrecy and confidentiality: when presenting the results, the data were kept confidential and citations were adjusted, when necessary, to preserve the confidentiality of the sources. Only a limited number of researchers directly involved in the project had access to personal data and original documents. Registration and information security: the files were stored using password access control and in accordance with Reglab’s internal information security policies. Retention and disposal: data will be stored for up to 12 months, exclusively for the purposes of methodological auditing and eventual replication, and will subsequently be eliminated. Responsible use of public data: although some analyzed data is public, its use was carried out in a responsible and ethical manner, for the sole purpose of independent research. Methodological transparency: the research methodology was described in detail to ensure transparency and replicability, contributing to scientific integrity and enabling independent validation of results. Non-Discrimination and Respect for Diversity: the research was conducted in order to respect diversity and avoid any form of discrimination.	Research funding: this publication is part of a series of research sponsored by Google, Meta and b/luz. Although it is a contracted study, Reglab maintained full editorial and methodological control over the project, with autonomous definition of the methodology, analysis of results and writing of this research report. The authors have preserved total professional independence and assume full responsibility for the content and conclusions presented. Processing of personal data: the research involved the processing of personal data only in the collection and analysis stages, in a limited manner and proportional to the objectives of the study, in accordance with Law 13,709/2018 (LGPD). Legal basis: all participants formally authorized their participation by signing a consent form, with knowledge of the research objectives and the use of data. Purpose and suitability: the data were used exclusively for the purposes of this research, in accordance with the consent obtained, and were not used for other purposes. Minimization and anonymization: personally identifiable information that was not relevant to the study objectives was anonymized in the transcripts and excluded from the active database. Secrecy and confidentiality: when presenting the results, the data were kept confidential and citations were adjusted, when necessary, to preserve the confidentiality of the sources. Only a limited number of researchers directly involved in the project had access to personal data and original documents. Registration and information security: the files were stored using password access control and in accordance with Reglab’s internal information security policies. Retention and disposal: data will be stored for up to 12 months, exclusively for the purposes of methodological auditing and eventual replication, and will subsequently be eliminated. Responsible use of public data: although some analyzed data is public, its use was carried out in a responsible and ethical manner, for the sole purpose of independent research. Methodological transparency: the research methodology was described in detail to ensure transparency and replicability, contributing to scientific integrity and enabling independent validation of results. Non-Discrimination and Respect for Diversity: the research was conducted in order to respect diversity and avoid any form of discrimination.