A giant screen lit up at Brazil’s Liberal Party convention on July 25, and Jair Bolsonaro addressed the crowd. Sort of. The figure on screen opened with a confession: “What you see here is not me. This is a simulation of my image and my voice using artificial intelligence.” Then, for the next 45 seconds, the avatar did exactly what the real Bolsonaro is legally forbidden to do — it campaigned. Hard. The AI clone urged supporters to back his son Flávio as Brazil’s “future president.” Now two of Brazil’s highest courts must untangle a question their rulebooks never anticipated about AI age laws and governance.
The Ghost in the Machine Had a Script
A labeled AI clone delivered explicit vote appeals on behalf of a legally silenced politician — and Brazil’s courts must now decide whether the label changes anything.
The synthetic Bolsonaro declared he was “imprisoned and silenced by an unjust and arbitrary decision based on something I never did,” according to Reuters. He called Flávio “flesh of my flesh” and “the person I have chosen to replace me.” He asked for votes by name.
The real Bolsonaro is serving a 27-year sentence after Brazil’s Supreme Court convicted him of plotting a coup following his 2022 election loss to President Lula. He’s under house arrest. Justice Alexandre de Moraes set conditions that explicitly bar him from disseminating political messages “including through third parties, regardless of the means used,” as reported by Valor Internacional.
That phrase is doing extraordinary legal work right now. Three details define the legal exposure:
- The video opened with a disclosure: “What you see here is not me. This is a simulation of my image and voice using artificial intelligence.”
- Bolsonaro’s lawyer Paulo Cunha Bueno publicly stated his client “did not authorize” the video and “could not have done so” given his restrictions.
- Brazil’s Superior Electoral Court (TSE) bans political deepfakes used to favor a candidacy — even with consent, even when labeled.
- The Workers’ Party and Brazil of Hope federation filed complaints with both the Supreme Court and the TSE on July 26.
“A very clear violation.” — Law professor Ivar Hartmann, on the avatar’s use under Brazil’s electoral rules, as reported by Reuters.
Flávio’s campaign argues this isn’t a “classic deepfake” because nobody was deceived. Their legal bet: the TSE’s rules target deception, not disclosed avatars. TSE president Cássio Nunes Marques has said AI in campaigns is permissible when it doesn’t cause harm — but he hasn’t ruled on whether this promotional use crosses that line.
The Precedent Nobody Wanted to Set
This case stress-tests a regulatory gap that no democracy has cleanly closed — what happens when the deepfake is honest about being a deepfake.
The EU AI Act requires deepfake disclosure. The US lacks equivalent federal rules entirely. Brazil went further — its TSE resolution explicitly bans synthetic political content “even with authorization.” The avatar is less Terminator, more legal ventriloquist’s dummy: speaking on behalf of someone the courts have silenced, then shrugging and insisting it wasn’t really him.
The real danger here isn’t deception. It’s the “liar’s dividend” — the emerging era where politicians deploy AI clones for curated messaging, then dismiss any inconvenient footage as synthetic. It’s the kind of scenario that sounds like a Black Mirror episode until it lands in actual Brazilian election law.
Penalties could include campaign sanctions or even Flávio’s disqualification, according to Reuters and Rio Times Online. If Brazil’s courts rule that a labeled clone still counts as a surrogate, every legally cornered politician on earth just found a new campaign strategy. The rulebook needs updating before the technology rewrites it for them.
