The announcement by the US government and major AI companies to establish voluntary self-regulation and safety controls for the development of the technology represents a “free pass” for the industry, potentially posing risks to public health and democracy in various countries.

This is the assessment of sociologist Sérgio Amadeu, a professor at the Federal University of ABC (UFABC) and the author of works on digital technologies.

Notícias relacionadas:

• Electoral Court approves restrictions on AI use in 2026 elections.
• AI, fake videos overtake fake news as threats to democracy.
• Electoral court recommends deepfake identification tool for candidates.

“What Trump did was to completely and irresponsibly hand over democratic control of AI development to the very corporations that develop it. This is a gigantic risk,” Amadeu said.

The agreement, signed Tuesday (Sep. 29) at the White House with OpenAI, Meta, Anthropic, Google, Nvidia, and X (formerly Twitter), includes the implementation of “internal controls” to monitor AI models. The controls are intended to “ensure that their models do not intrude upon or access technical systems in unintended ways.”

“This is already being done. The entire AI training process requires robust oversight, and internal monitoring is already in place. None of this adds anything significant to the regulation and control of the improper use and development of machine learning and deep learning,” Amadeu added.

Regarding the proposal for independent external audits of AI companies, the UFABC professor argues that the audits would be paid for by the companies themselves.

“To what extent will this external auditor, paid by the company itself, offer criticism and suggestions that could cause discontent among the companies’ boards of shareholders?” the professor asked.

Fragile benefits

Ettore Arpini, an AI governance expert and founder of Plures - an organization that supports the training of professionals in the AI field - told Agência Brasil that the voluntary agreement among AI companies could yield real benefits.

However, he warns that the commitment could be abandoned once the safeguards become commercially detrimental.

Arpini cites the example of the Amazon Soy Moratorium, a voluntary agreement involving companies in the sector that committed to refraining from planting soybeans in the Amazon and preventing deforestation of the biome.

“This worked for 20 years on a voluntary basis, until the political and economic environment changed and the participating companies withdrew from the agreement, causing the moratorium to fail. What happens if a company fails an audit or withdraws from the pact?” he asked.

Public health

Sociologist Sérgio Amadeu points out that AI poses environmental risks due to its high energy and water consumption, privacy risks due to its capacity for mass surveillance of populations, as well as risks to democracy and public health.

He explains that various research projects conducted in the pharmaceutical industry in partnership with big tech companies using AI models may pose risks to public health and therefore require government regulation.

“We know that when the use of AI is controlled solely by investors, it poses a huge risk to experiments,” he commented.

Among the risks, the expert cites the possibility that the use of AI could lead people to self-medicate or even use the technology to diagnose themselves.

Social concerns

The Trump administration has refused to impose limits on AI development, despite requests from the industry itself and pressure from civil society in the United States.

An opinion poll conducted by Ipsos and Reuters found that 73 percent of Americans believe AI companies are not taking sufficient measures to prevent the technology from harming society, while 39 percent think that AI is having a negative impact on society.

One of the identified risks is that these systems could make cyberattacks easier to carry out by hacking into the systems of other companies and institutions. Other risks associated with AI include its impact on the labor market and the “malicious” use of the technology by criminals to conduct cyber scams.

Democracy

AI can also be used to influence public opinion by manipulating information circulating on social media, according to the International AI Safety Report 2026, prepared by 100 experts from more than 30 countries.

“Several laboratory studies have shown that interaction with AI systems can lead to measurable changes in people’s beliefs,” the report notes.

Another risk highlighted in the report relates to the development of chemical and biological weapons.

“General-purpose AI systems can provide information on the development of biological and chemical weapons, including details on pathogens and specialized laboratory instructions,” the international report states.

Fake videos created by artificial intelligence and low-quality information have replaced traditional fake news and have become the main digital threat to elections and democracy. This is the conclusion of the Internet, Democracy, and Elections Guide, published on Tuesday (Sep. 22) by the Brazilian Internet Steering Committee (CGI.br).

The guide highlights the abrupt shift between 2018 - the year of the first edition - and the current election cycle. Whereas open networks such as Facebook were the primary channels for dissemination back then, more restricted networks such as WhatsApp are now the main means of dissemination.

Notícias relacionadas:

• Electoral court recommends deepfake identification tool for candidates.
• Electoral Court approves restrictions on AI use in 2026 elections.

CGI.br experts also noted that disinformation has become increasingly widespread, circulating simultaneously across multiple platforms. “In terms of format, video has become the primary vehicle, almost always accompanied by a short text that anchors and legitimizes the false narrative,” the document states. It is no wonder that institutions such as the World Economic Forum have considered disinformation one of the greatest global risks since 2025.

Given this scenario, the central role of the internet as a medium for information, entertainment, and communication is becoming a more important factor in democratic contests.

Recent tools, such as generative artificial intelligence, emerged during the 2022 election cycle but are expected to have an even greater impact this year when combined with dissemination tools, with more tailored targeting of specific user profiles and more complex narrative structures designed to appear more authentic.

“Today, digital platforms play a central role in how Brazilians stay informed about what is happening in the world, in the country, and in the cities where they live,” says CGI.br coordinator Renata Mielli.

She highlights the responsibility of private companies driven by economic objectives rather than the public interest. “Understanding how these environments function, how content is distributed, and what impact this has on public debate is essential to strengthening the integrity of information and democracy itself. Clarifying these mechanisms is one of the guide’s objectives.”

“Available on the CGI.br website, the publication addresses topics such as how content recommendation systems operate, the impact of generative AI on the electoral landscape, and how advanced tools - such as deepfakes and conversational language models - work.

The guide also includes guidelines for institutional efforts to combat disinformation and advice for citizens on how to avoid becoming victims of or spreading false or misleading content.

In addition to analyzing the structures of media outlets and social networks, the study highlights the role of their communicators - influencers. It acknowledges their prominent role but points to a fundamental difference from journalism: they do not follow a code of ethics, nor are they institutionally accountable for the accuracy of their statements or for inaccuracies - or even deliberate errors - in what they publish.

Warning

According to the study by the CGI.br group, the most frequently cited concern regarding deepfakes is their realism or credibility. As this type of deceptive content has undergone constant refinement, it now has the potential to deceive even voters who are highly experienced in evaluating content.

This type of content can foster widespread distrust of all kinds of information, including truthful information.

“The loss of trust in institutions such as the press and the judiciary, combined with political polarization, also contributes to this scenario, causing personal convictions to take precedence over fact-checking,” the guide warns.

Another problem that arises from this climate of mistrust is that people increasingly consume content that confirms what they already believe. “When they encounter information that contradicts these beliefs, they tend to dismiss it - a phenomenon known as motivated reasoning. In this sense, it can reinforce their ideologies,” the study points out.