On Saturday, July 25, the PL’s national convention officially nominated Senator Flávio Bolsonaro as its presidential candidate. At the same event, a 46-second video was shown featuring an AI-generated avatar that replicated the image and voice of former President Jair Bolsonaro, asking for support for his son. The incident immediately sparked two parallel legal actions. At the Federal Supreme Court, Justice Alexandre de Moraes gave Jair Bolsonaro’s defense team 48 hours to report whether prior authorization had been obtained for the screening, within the context of the proceedings to enforce the sentence imposed on the former president for attempting a coup d’état. At the Superior Electoral Court, a complaint against the use of the video is being reviewed by the Court’s president, Justice Kassio Nunes Marques.
Public debate surrounding the case tends to focus on a question that has already been answered. Brazilian election law regarding artificial intelligence is clear on this point. The TSE resolution prohibits the use of synthetic content to benefit or harm any candidate, leaving no significant room for ambiguity. The central issue is whether this rule—enforced by a court headed, for the first time in history, by appointees of the very former president whose family is running in the election—will continue to produce the same result it did before, in a similar case, against a different political opponent. The letter of the law has settled the question of what is permitted. The question that remains open is who decides what it means in practice.
The ruler that already existed
In 2024, Evandro Leitão of the PT, then a candidate for mayor of Fortaleza and now in office, posted a campaign video on social media featuring deepfakes of Barack Obama, Taylor Swift, Tom Cruise, and Cristiano Ronaldo. The four appeared in the video declaring their support for him. The TSE fined the candidate 15,000 reais. In May 2026, the Court unanimously upheld the penalty, rejecting the appeal filed by the defense. The justice presiding over the case, Minister Villas Bôas Cueva, established a legal principle that should serve as a benchmark for any similar incident. The tampering with digital content for electoral purposes is, in and of itself, sufficient to constitute an irregularity. This criterion does not require proof that the voter was actually misled, nor does it depend on the stated intent of the content’s creator.
While the world debates the urgency of curbing the impact of artificial intelligence on democracies, Brazil imposes paltry fines for fraud. In the Fortaleza case, the Superior Electoral Court (TSE) imposed a fine of 15,000 reais. By way of comparison, in the United States, the communications regulatory agency fined the political consultant responsible for cloning Joe Biden’s voice in phone calls during the New Hampshire primaries $6 million. The vast difference in figures reveals the Brazilian dilemma: digital violations are not treated as an existential threat to democracy, but rather as an operational cost built into the budgets of million-dollar campaigns.
The distinction invoked by the defense and the paradox of the law
Flávio Bolsonaro’s defense team rejects the comparison and points out a specific difference. Jair Bolsonaro’s avatar declares itself to be synthetic within the first few seconds of the video, when it identifies the content itself as a simulation of his image and voice produced by artificial intelligence. Evandro Leitão’s video used the images of real celebrities without any equivalent disclosure. Flávio’s campaign went a step further and attempted to flip the narrative, presenting the video itself as an educational warning about the risks of artificial intelligence, in an ironic tone directed at his political opponents.
The distinction exists, but it does not resolve the underlying legal issue. The ruling in the Fortaleza case expressly precludes an analysis of the potential for misleading the public. If the criterion is objective and the tampering alone is sufficient, the disclaimer included in the video is not enough to remedy the violation.
Moreover, the defense’s argument directly contradicts the Electoral Court’s own wording. The TSE Resolution is unequivocal in prohibiting the use of deepfakes to favor candidates “even with authorization.” The legislature had already considered the consent argument and rejected it. If the reporting judge accepts the distinction between prior notice and consent, he will not be engaging in legal technicalities; he will, in practice, be revoking the express text of the rule that his own Court drafted just a few months earlier.
The defense also invoked another technical argument: the TSE’s case law on early campaign advertising, which requires an explicit and unequivocal call for votes to constitute a violation, in addition to the rule authorizing internal party campaigning during the fifteen days prior to the convention. These are arguments that are consistent with the case, but they leave open the central question that the Fortaleza case had already answered.
The third regulatory approach: a new rule in the works?
Faced with the dilemma of whether to enforce a strict ban or accept the “consent” defense, the possibility that the Bolsonaro case could serve as a testing ground for creating a third regulatory approach is gaining traction behind the scenes at the Superior Electoral Court. Sources at the Court indicate that the Electoral Court is already considering ways to close loopholes in the current resolution governing presidential elections.
The proposed solution would be to relax the absolute ban on deepfakes when there is not only the express consent of the person depicted, but also explicit labeling and an internal party context, such as that of a convention. This maneuver would allow the TSE to avoid penalizing Flávio Bolsonaro’s campaign without having to formally overturn the Fortaleza precedent, creating a precise distinction based on the context of dissemination and the transparency of the avatar. The risk of this third approach, however, is that it could legitimize the use of synthetic simulations of political leaders as a permanent campaign tool, thereby undermining the rigor of the original rule under the pretext of technological advancement.
The rigged lottery and who decides
The answer to that question will, at first, rest with Kassio Nunes Marques. The justice did not reach this position merely by chance through an electronic lottery. In May 2026, in one of his first acts as president of the Court, Nunes Marques amended the TSE’s internal rules to include himself and Vice President André Mendonça in the select group of judges assigned to hear cases involving election advertising. Previously, this role had been concentrated in the hands of a justice appointed by Lula. The change was not accidental; it was a calculated move in institutional architecture to ensure control over the most sensitive narratives of the election.
Even before formally ruling on the case for which he was appointed rapporteur, Nunes Marques had already given a public interview on the subject. In an interview with *Estadão*, he stated that using artificial intelligence for campaigning is lawful, and that what is not allowed is using the technology to harm someone. The statement, while reproducing part of the text of the resolution, omits the passage that prohibits its use to “favor a candidacy.” It publicly anticipates the interpretation that the case’s rapporteur is likely to adopt before any formal ruling by the Court.
As the American jurist Oliver Wendell Holmes pointed out as early as the 19th century, the life of the law has always been more about experience than logic. This statement accurately describes what is at stake in Brazil in 2026. The legal text applied in Fortaleza is the same one that will—or will not—be applied to the Bolsonaro case. The difference lies in who sits in the judge’s chair, how that person rose to that position, and the political experience they bring to the table.
The institutional friction is already public knowledge
The tension between the STF and the TSE over this case is a well-documented fact. TSE justices believe that Moraes’s demand for an explanation constitutes undue interference in a matter they consider to be within the exclusive jurisdiction of the Electoral Court. Behind the scenes at the Court, discussions are underway regarding a lenient resolution to the incident, involving a fine against the party, the removal of the video from the internet, and a formal warning—without revoking the party’s registration or declaring Flávio Bolsonaro ineligible to run for office.
The friction predates the avatar incident. Weeks earlier, Nunes Marques had ordered the suspension of the release of an election poll by the AtlasIntel institute, a decision that was criticized both within the TSE and within the STF itself. The institutional dispute between the two courts was already underway before Bolsonaro’s video reignited the debate, and it is likely to remain one of the judiciary’s top priorities through October.
The ruler and the person holding it
By October 2026, Brazil will have an answer to the question regarding artificial intelligence and elections. The answer already exists; it was provided in Fortaleza, in a case involving a PT candidate, and unanimously confirmed in May 2026. The tampering with digital content for electoral purposes constitutes an irregularity, regardless of whether there is evidence of intent to mislead.
The Bolsonaro case tests—more than the mere existence of this standard—its endurance and the resilience of the Brazilian democratic system in the face of Juvenal’s age-old question: Quis custodiet ipsos custodes? (Who watches the watchmen?). An Electoral Court led, for the first time, by appointees of the former president—whose family is running in the election—now faces the challenge of applying the same standard to an ally that it applied, without hesitation, to an opponent.
If the criterion remains objective, the Fortaleza precedent will be established as solid case law, capable of withstanding changes in court composition and shifts in political power. If, on the contrary, the Court chooses to create a third regulatory path or yield to the distinction regarding prior notice—ignoring the very law that prohibits manipulation “even with authorization”—Brazil will learn, in practice, the most cynical lesson of electoral law: the rule is worth less for what is written and much more for who designed the rules of the game to uphold it when the time comes to apply it.
More about
Politics
