Brazil’s electoral court sets deepfake criteria but finds no violation in Bolsonaro AI video

Brazil’s Superior Electoral Court cleared Flávio Bolsonaro over an AI-generated version of his father, Jair Bolsonaro, but at the same time set new limits on deepfakes in election advertising.

The court voted 4-3 that the video shown at the convention endorsing Flávio Bolsonaro’s presidential candidacy did not violate electoral law. The event took place in July, before the formal campaign began, and the majority said the image’s use at a party convention did not demonstrate an attempt to evade the rules.

“

This is when the court begins to define, in response to new forms of communication emerging at an accelerated pace, the boundaries of its application

The ruling by Justice Kássio Nunes Marques, who presides over the court, was backed by André Mendonça, Dias Toffoli and Antonio Carlos Ferreira. “This is when the court begins to define, in response to new forms of communication emerging at an accelerated pace, the boundaries of its application,” Nunes Marques said. “In this situation, I do not believe it is possible to infer from the use of artificial intelligence at the party convention an intention to circumvent electoral law.”

The court said a deepfake is illegal when three elements are present: synthetic content, a high degree of realism capable of confusing voters, and the creation or alteration of the image or voice of a living, deceased or fictional person. Punishment for using a deepfake applies only when it appears in electoral advertising.

“

In this situation, I do not believe it is possible to infer from the use of artificial intelligence at the party convention an intention to circumvent electoral law

Dissenting ministers Villas Bôas Cueva, Azevedo Marques and Estela Aranha argued that the video may have been lawful at the convention but became unlawful when it was broadcast live on social media. Cueva said realism should not be required if synthetic content aims to help or harm a candidacy. Azevedo Marques said the court’s existing resolution already prohibits deepfakes broadly.

The court excluded stylized avatars, drawings and caricatures without realistic likeness, along with image and sound improvements, visual identity elements, opening clips and logos. Rules approved in March do not ban AI in campaign advertising, but require such content to carry labels.

“

Whether we like it or not, our resolution defined deepfake for purposes of applying electoral rules