Nearly one-third of Brazilian workers face exposure to AI

These are the results of a study by the Brazilian Institute of Economic Research (FGV Ibre) and the institute's monthly report, which were shared with Valor. “Occupations that involve repetitive analytical tasks or information-intensive tasks show high potential for automation with AI,” they write. Luis Guilherme Simra, Director In the letter of FGV Ivre.

Researchers estimate that in the third quarter of 2025, 29.8 million workers, or 30% of the employed workforce, were exposed to generative AI. This is the latest data available from IBGE's ongoing quarterly PNAD survey. In 2012, that number was 23.2 million, representing 27% of the employed population. Exposure to AI can be positive if it complements traditional operations, or negative if it leads to complete replacement of employees.

“The glass is less than half full. It is full for workers. The labor market is very strong. We won't see the impact of this AI on an aggregate level, but if technology adoption continues at this pace and the substitution effect strengthens, it will happen,” he said. Fernando de Horanda Barbosa Filhothe authors of this study Paulo Perchetti, Janaina Feijo, Daniel Duque. “The pace of adoption of new technology around the world has never been faster,” he added.

Based on a 2025 study by the International Labor Organization (ILO), FGV Ivre researchers have replicated the methodology for mapping the extent of occupational exposure to AI for Brazil, updating the Global Index of Occupational Exposure to Generative AI.

First, we converted the list of occupations from the international ISCO-08 system into the occupational classifications used in Continuous PNAD. We then classified Brazilian occupations based on AI exposure across six levels proposed by the ILO: none, low, and gradient 1 to 4 (lowest to highest).

From the first quarter of 2012 to the third quarter of 2025, the proportion of workers exposed at slope 1 decreased (from 35.8% to 30.5%). All other gradients increased exposure. The percentage of workers exposed to the highest level 4 AI rose from 16.1% in 2012 to 17.6% in 2025.

The researchers cross-referenced this analysis with other socio-economic indicators. By gender, the exposure rate for women (35.4%) was higher than for men (25.2%). The region with the highest risk is the South East (32%), the lowest in the North (25.3%) and then the North East (26.7%). Exposure to AI decreases with age, being highest among those aged 14–29 years (35.9%) and lowest among those aged 45–59 years (24.5%) and 60 years and older (25.7%).

Regarding education, the impact increases sharply from 10.2% for those with no schooling or incomplete primary education to 42.7% for those with a university degree.

According to the researchers, international studies have already shown that exposure to AI tends to increase the higher the level of qualification required for an activity, which typically equates to higher education. Therefore, one reason for the higher exposure of women (compared to men) and young workers (compared to older people) in Brazil may be their higher education level. Similarly, they say the contrast between the southeast and the north and northeast may reflect differences in regional development.

Among the sectors analyzed, agriculture has the lowest exposure, with only 1.5% of occupations exhibiting some degree of AI exposure. The sector most at risk is financial intermediary services, with 90.6% at risk. “Some jobs will disappear, no doubt about that. There will also be new jobs created. Those who know how to use AI will protect their jobs. Those who do not know how to use AI will probably not be able to change their position,” Barbosa Filho said.

FGV Ivre researchers cite research from the International Monetary Fund (IMF) that shows that in developed countries such as the United States and the United Kingdom, around 60% of workers are exposed to AI, compared to around 40% in emerging countries such as Brazil, Colombia, and South Africa.

However, the challenge is that while around half of AI exposure in developed countries is complementary in nature, in Brazil around 20% of the employed population has high exposure to AI and low complementarity with AI, making these workers highly vulnerable to job loss, the researchers said. Meanwhile, just over 20% of Brazilian workers have high exposure to highly complementary AI, which could lead to productivity and wage gains, according to the IMF.

“The train of history is moving forward. The question is whether we can ride it well or not. And even if we do everything right, there will be losers. We need to think about how to protect these people,” Barbosa Filho said.

FGV Ivre researchers advocate a public policy agenda to address this scenario. “This agenda is premature, because once the wave starts, it doesn't stop, even if we don't yet see the full impact. When technology becomes affordable, people start using it en masse. Then there's no point chasing the problem. We need to anticipate it,” Barbosa Filho said.

Researchers say education and reskilling requires investment in high-quality basic education, statistics, digital skills, programming and, above all, cross-cutting competencies such as problem-solving, communication and teamwork. “I think in some ways people are realizing the importance of upskilling. The market is demanding this more and more. If people don't adapt, this is going to be a problem,” Perchetti said.

From a social protection and active labor policy perspective, AI-enabled sectoral and occupational shifts have increased the importance of active policies such as unemployment insurance systems, income transfer programs, job matching, training, and reemployment subsidies, which could reduce unemployment duration and support transitions between occupations, the researchers said.

“From a public policy perspective, I am not so worried about this social safety net, because during the pandemic we have already shown that we have the capacity for it. What I am more concerned about is the preparation of people entering the labor market and our reskilling model,” Barbosa Filho said.

He said the introduction of agile and flexible work mediation and training systems is an urgent measure. “Today's professional training is no longer a long course lasting many months as it used to be. It consists of small skill modules needed to adapt to the workforce. We need an agile system that can quickly identify problems and flexibly adopt solutions.”

Researchers say measures are also needed to regulate the use of AI in the workplace. They argue that there is a need to control the use of technology for monitoring, continuous performance evaluation, and automation of HR decisions. The study also warns of risks related to algorithmic bias, discrimination, labor intensification, and power asymmetries between companies and workers.

All of this requires a regulatory framework that guarantees transparency, minimal accountability, and mechanisms to challenge automated decisions, as well as the participation of trade unions and worker representatives in workplace technology governance, the researchers say. “The more widespread the laws that allow the use of AI, the greater the challenge. The first brake is regulation. That's why we're talking about responsible AI. In my opinion, this is an attempt to regulate its use so that the impact occurs more slowly,” Barbosa Filho said.

Finally, there are issues regarding competition policy and technology diffusion. They say the concentration of AI capabilities in a few large technology companies raises questions about the allocation of market power, data capture, and productivity gains.

“Two forces are in play at the same time. Within the same company, the use of AI can reduce the disparity between employees. Some studies have shown that the workers who benefit most within the same job are the least productive. However, when compared to other sectors where AI cannot be implemented, the disparity may widen further. What will determine the outcome is how much technology can be deployed horizontally across the economy,” Barbosa Filho said.

Researchers say competition policies, open standards, promoting open AI ecosystems and supporting small and medium-sized enterprises to adopt technology will be key factors in preventing profits from being concentrated in the hands of a few companies.