Brazil suspends Meta from using Instagram posts to train AI

Brazil has blocked Meta from using Brazilians’ Instagram and Facebook posts to train its artificial intelligence (AI) models.

It comes weeks after the company abandoned similar plans to use UK and European users’ posts for the same purpose.

On Tuesday, Brazil’s national data protection agency (ANPD) said it would immediately suspend Meta’s latest privacy policy, which allows it to train generative AI models such as chatbots based on posts from its users.

A Meta spokesperson told the BBC the company was “disappointed by the decision”, adding that their approach complied with local privacy laws.

“This is a step backwards for innovation, competition in AI development and further delays bringing the benefits of AI to people in Brazil,” the company added.

Meta has a significant market in Brazil. There are 102 million Facebook users and more than 113 million Instagram users in the country.

The ANPD said it had acted over the “imminent risk of serious and irreparable damage, or difficulty repairing fundamental rights of the affected [account] holders”.

Meta was given five working days from ANPD’s decision to show it has amended its privacy policy to exclude the use of personal information found in public posts to train generative AI. If it fails to comply it will face a daily fine of R$50,000 (£6,935).

The company’s updated policy was also the focus of scrutiny in the UK and the European Union (EU).

Under its privacy policy changes, which were due to take effect in the region on 26 June, Meta users’ information would be used to “develop and improve” its AI products.

In Europe, the policy change would include posts, images, image captions, comments and Stories that users over the age of 18 had shared with a public audience on Facebook and Instagram, but not private messages.

But that was put on hold after Meta said it had received a request from the Irish Data Protection Commission (DPC) on behalf of other European stakeholders to delay its training of large language models (LLMs).

LLMs are a type of artificial intelligence that powers chatbots, such as OpenAI’s ChatGPT and Google’s Gemini.

On 14 June, when it announced the delay, Meta said this was a “step backwards” for AI in Europe.

However Meta decided to press ahead with the policy change in Brazil.

Pedro Martins, from Data Privacy Brasil, welcomed the ANPD’s decision. He told the BBC there was a discrepancy between Meta’s data protection measures for its Brazilian and European users.

Meta had planned to use posts from Brazilian children and teenagers to train its AI models, he said, while in Europe nobody under 18 would have their posts used.

Brazil’s data protection regulator also found that personal data found in children and teenagers’ posts could be collected and used to train Meta’s AI systems, which could be in breach of the country’s data protection law.

In addition, Mr Martins said, in Europe the steps users can take to prevent Meta from using personal information are more straightforward than in Brazil, where he said it can take as many as eight steps for users to block the company from using their posts.

The BBC has asked Meta to respond to the claim that it had planned to use posts from Brazilian children and teenagers to train its AI models, and whether it imposed more onerous steps for opting out on users in Brazil.