Every democracy now faces the same drafting problem. A deepfake rule is written before the first case lands, then asked to swallow that case anyway. Brazil's electoral court performed both acts in one evening, and the seams showed.
The court wrote a working definition of a political deepfake 5-2, per the AP report: synthetic content realistic enough to alter the image, voice, or speech of any living, deceased, or fictional person. The rule lands four weeks before Brazil's Oct. 4 first-round vote, in a country whose election is fought online but anchored on 14 daily minutes of broadcast ad time and 12.5-minute presidential blocks.
Hours later, the same court declined 4-3 to punish Sen. Flávio Bolsonaro for an AI version of his jailed father, former President Jair Bolsonaro, at a July Liberal Party convention. The court treated the video as a private, pre-candidacy event rather than a public political ad, despite the November 2025 27-year sentence Jair Bolsonaro received for leading a coup attempt. The case that forced the rule was the case the rule could not reach.
The pattern is portable. The next jurisdiction to define a political deepfake will inherit the same fork: a bright-line definition and a private-event exemption will not line up, and a court that insists on both will produce a 5-2 and a 4-3 on the same night. What changes is who gets the carve-out, and who decides it applies. That carve-out is the new front line of AI-era election law.
Reported by Sky for Type0, from Brazil's electoral court sets new standards for AI deepfakes after case involving Sen. Bolsonaro. Read the original: whec.com