Bill proposed by Senate president moves to Lower House for review.
The Senate approved Bill 2338/2023 on Thursday (5) in a symbolic vote, setting the regulatory framework for artificial intelligence in Brazil. The proposal will now be reviewed by the Lower House.
The bill, proposed by Senate President Rodrigo Pacheco, was prioritized for a vote within this year while he remains in office. Rapporteur Eduardo Gomes introduced several amendments to the text, presenting a total of seven reports on the matter. The progress was achieved by the rapporteur through negotiations with leaders and a government-opposition agreement to withdraw amendments aimed at altering the proposal.
Copyright Rules and AI Risk Classification
The bill removed a provision to change the rules on copyright. Consequently, it maintains the requirement for developers to compensate authors for content used to supply AI system databases.
Under this understanding, the Workers’ Party (PT) also withdrew its amendment, preserving one of the rapporteur’s most significant changes. Mr. Gomes removed the section classifying AI systems used by big techs—such as algorithms for content promotion—as high risk. This regulation is to be addressed through specific legislation.
According to the bill, activities classified as high risk will be subject to stricter regulatory oversight. Companies developing and operating these systems must assess and test the technology’s safety and implement measures to prevent discriminatory behavior.
Unlike Mr. Pacheco’s original proposal, the rapporteur made preliminary assessments—conducted before the launch—of new AI systems optional. This analysis is to be carried out by companies to determine the risk level of new technologies. Mandatory assessments are maintained only for generative AI systems, which can create new content like text, images, or music from existing data patterns.
Who is the regulator?
The responsibility to regulate high-risk AIs will lie with the National System for Regulation and Governance of Artificial Intelligence (SIA). However, the section that classified risk to information integrity, free speech, democratic process, and political pluralism—which could have included big techs—was removed.
The legislation establishes the creation of the National System for Regulation and Governance of Artificial Intelligence (SIA), responsible for overseeing AI usage in the country. The National Data Protection Authority (ANPD) will coordinate the regulatory agencies’ activities in each sector.
The bill prohibits developing tools that could, for example, “instigate or induce natural persons or groups to behave in ways that harm health, safety, or other fundamental rights,” and “exploit any vulnerabilities” of individuals or groups with the same intent.
Biometric data usage is permitted only for capturing fugitives, executing arrest warrants, and instances of crimes carrying sentences of over two years. Judicial approval allows its use for gathering evidence in police investigations.
What about the fines?
In case of non-compliance by companies, the bill provides for partial or total suspension of activities, fines of up to R$50 million or 2% of revenue, a ban on processing specific data sets, and prohibition or restriction from participating in regulatory “sandbox” environments for up to five years. A “sandbox” is a controlled environment that allows for experimentation without real-world risks.
Goals and strategies for AI regulation
Celso Camilo, an AI professor and researcher at the Federal University of Goiás (UFG), emphasized the need for Brazil to accelerate AI regulation to secure a spot in the global race for innovation investment. “More than being perfect, the regulation needs to be enacted,” Mr. Camilo told Valor.
The expert noted that Brazil has the chance to gain an edge among developing countries in attracting AI investments if it acts swiftly.
“Despite the challenges, I see Brazil has a unique opportunity to effectively execute AI strategies. It’s just a matter of speeding up.”
Regarding the pressure from major technology companies to ease AI regulations in Brazil and globally, the professor advocates for a balance between “pro-market, more liberal regulation with less oversight, pro-human rights, more restrictive approaches, and a hybrid between the two streams.”
“I believe we need to balance this; AI shouldn’t be regulated in an amateurish way, but we need to move faster than anticipated and understand that the regulation won’t be perfect,”
he stated.
Among AI’s biggest challenges, Mr. Camilo highlighted algorithm biases, transparency about how AI training data is handled, and the necessary knowledge for people to access AI.
Besides expediting AI regulation, the expert said Brazil should hasten the implementation of initiatives like the Brazilian Artificial Intelligence Plan (PBIA), announced in June. In his view, the PBIA is a sound plan, but there are execution timing challenges. “That is not an issue isolated to Brazil but affects developing countries where there’s a significant challenge in executing plans and prioritizing science and technology, which are not traditional focus areas in these countries,” he said. At UFG, Mr. Camilo is working to accelerate AI projects connecting the academic, public, and private sectors.
Source: Valor International
