Brazil runoff tests rules meant to slow AI election misinformation

In short: Brazil’s October 25 presidential runoff is putting the country’s election rules to the test as AI-made fake political content spreads quickly online.

What's going on

Brazil is heading to a presidential runoff on October 25. President Luiz Inácio Lula da Silva will face Flávio Bolsonaro after a close first round, with Bolsonaro at 47% and Lula at 45%.

A monitoring project from the Lupa Observatory counted 920 political posts that were AI-generated or manipulated between August 16 and September 28. That is nearly one per hour. More than half were deepfakes, meaning fake videos or audio made to look real (like a convincing impersonation). The project also found that about 60% of the AI content did not include a label saying it was made with AI.

The posts included made-up statements and endorsements, fake journalists and news reports, and false claims about Brazil’s electronic voting machines. The problem is not limited to videos. Researchers have also found that AI chatbots sometimes gave wrong election information or suggested and ranked candidates, which Brazil’s election court rules do not allow.

Brazil’s Superior Electoral Court, known as the TSE, requires AI-made campaign material to be labeled. It bans deepfakes meant to help or hurt candidates. It also limits new AI-made political content in the 72 hours before voting and the 24 hours after.

What to watch

Enforcing these rules is hard when the content comes from regular users, anonymous accounts, and private messaging groups rather than official campaigns. In a close race, the key risk is speed. Fake content can travel faster than courts and fact-checkers can check it, like trying to stop rumors once they have already spread through a crowd.

Source: NYTimes