
Brazil's data protection authority, the Autoridade Nacional de Proteção de Dados (ANPD), has temporarily banned Meta from processing users' personal data to train the company's artificial intelligence (AI) algorithms.
The ANPD said it found “evidence of processing of personal data based on inappropriate legal assumptions, lack of transparency, restrictions on the rights of data subjects and risks to children and adolescents.”

The decision comes after the social media giant updated its terms of service to allow public content on Facebook, Messenger and Instagram to be used for AI training purposes.
A recent report by Human Rights Watch found that LAION-5B, one of the largest image-text datasets used to train AI models, contains links to identifiable photos of children in Brazil, putting them at risk of malicious deepfakes and further exploitation and harm.
In Brazil, one of the largest markets with around 102 million active users, the ANPD noted that the Meta update violates the General Personal Data Protection Law (LGBD) and poses “an imminent risk of causing serious, irreparable or irreparable damage to the fundamental rights of affected data subjects.”
Meta has five working days to comply with the order or risk being fined 50,000 reais (about $8,808) per day.
In a statement shared with The Associated Press, the company said the policy “complies with Brazilian privacy laws and regulations” and that the ruling is “a setback for innovation and competition in AI development, further delaying the benefits of AI to the Brazilian people.”

The social media company has faced similar backlash in the European Union, forcing it to suspend plans to train AI models on data from users in the bloc without their explicit consent.
Last week, Meta's president of international affairs, Nick Clegg, said the EU was “no longer fertile ground for innovation and world-class companies”, adding that “the era of generative AI presents an opportunity to change that.”
The rollout comes as Cloudflare released a new “one-click” tool that prevents AI bots from scraping customers' websites for content to train their large-scale language models (LLMs).
“The feature will be automatically updated over time as we find new fingerprints of bad bots that we identify as scraping the web extensively to train our models,” the web infrastructure company said.