Skip to content

Brazil suspends Meta’s policy to train AI on user data

  • by
  • 3 min read

Photo: Murilo Fonseca

The Brazilian National Data Protection Authority (ANPD) ordered Meta to immediately halt the implementation of its new terms within five working days. These terms would allow Meta to use personal information from its social media platforms for AI training purposes. If Meta fails to comply, it must pay daily fines of 50,000 Brazilian Reais.

The ANPD’s preliminary analysis indicated that users generally share information on Meta’s platforms within the context of personal relationships or specific communities. It highlighted that users likely did not anticipate that their data, possibly shared many years ago, would be used to train AI systems—a technology that may not have even existed when the data was initially shared.

The investigation also revealed that personal data of children and adolescents, including photos, videos, and posts, could be harvested for AI training. The ANPD mandates that the processing of minors’ data must be prioritised in their best interests and include robust safeguards and risk mitigation measures, which were found lacking in Meta’s policy.

The decision to ban Meta stems from a previous Human Rights Watch (HRW) investigation revealing that about 170 images of Brazilian kids from the LAION-5B data set were used to train AI models, reports The Verge.

Meta’s updated privacy policy, effective June 26, applies to its major platforms, including Facebook, Messenger, and Instagram. The policy allows Meta to use publicly available user data to enhance and train its generative AI systems. With Facebook alone boasting around 102 million active users in Brazil, the potential impact on personal data is substantial.

The ANPD investigated Meta’s new privacy policy independently, without any external provocation. The authority cited potential violations of Brazil’s General Data Protection Law (LGPD) as the basis of its scrutiny. Preliminary analysis revealed significant risks of serious and hard-to-repair harm to users, prompting the ANPD to issue the suspension as a cautionary measure.

ANPD highlighted several critical concerns, including an inadequate legal basis, a lack of transparency, excessive restrictions on users’ rights to access information, and the potential use of personal data of children and adults.

The agency stressed that Meta failed to provide adequate and necessary information for users to understand the implications of their data being used to train AI models. The investigation found excessive and unjustified obstacles to users’ access to information and exercise their rights to oppose data processing.

The authority also questioned the legitimacy of Meta’s legal justification for processing personal data under the ‘legitimate interest’ clause. The ANPD noted that this hypothesis is not applicable when sensitive personal data is involved due to its higher risk of discrimination. The assessment emphasised the need to consider legitimate user expectations and adhere to the principles of purpose and necessity.

A Meta spokesperson told Associated Press that the company is disappointed and it will comply with privacy laws and regulations in Brazil. The company also said there are “excessive and unjustified obstacles to accessing the information and exercising the rights” to opt out.

In May 2024, EU launched a formal probe into Meta’s impact on minors’ addictions. The same month, Spain halted Meta’s election data processing before EU voting.

In the News: Asus introduces first routers with NordVPN integration

Kumar Hemant

Kumar Hemant

Deputy Editor at Candid.Technology. Hemant writes at the intersection of tech and culture and has a keen interest in science, social issues and international relations. You can contact him here: