A Brazilian court is being asked to decide a question that has quickly become one of the most difficult in modern election law: when political speech is generated, edited or amplified by artificial intelligence, who actually “authorizes” it—and what does authorization mean in practice?
The dispute centers on a digital avatar of former president Jair Bolsonaro that appeared in an AI “projection” during a period of intense legal and political wrangling over how Brazil’s election rules should apply to AI-generated content. Bolsonaro’s legal team argues that he did not authorize the projection. Lawyers representing left-wing groups challenging the use of such material say the opposite: that the avatar was deployed as part of a political effort and therefore falls within the scope of electoral obligations and restrictions.
At stake is not only the fate of a single piece of AI-generated media, but the direction of travel for Brazil’s regulatory approach to synthetic media—an area where many democracies are still improvising. The case is unfolding as election authorities and courts worldwide confront the same core problem: AI can make political messaging cheaper, faster and more persuasive, while also making it harder to trace responsibility, verify consent, and determine whether content is misleading or improperly authorized.
In Brazil, the legal fight has taken on a distinctly procedural character. Rather than focusing solely on whether the avatar looked convincing or whether it could be considered deceptive, the arguments are increasingly about governance: what counts as official endorsement, what evidence is required to prove authorization, and how electoral rules should be interpreted when the “speaker” is not a human being but a synthetic representation.
For Bolsonaro’s lawyers, the central claim is straightforward: the former president did not authorize the AI projection. Their position suggests that the avatar was either created without his permission, used by third parties outside his control, or deployed in a way that does not meet the threshold for attributing responsibility to him personally. In legal terms, they are trying to separate the existence of AI-generated political imagery from the question of who can be held accountable under election regulations.
That distinction matters because election law often treats authorization as a gatekeeping concept. If a candidate or political figure authorizes a message, that message can be treated as campaign material subject to specific rules—rules about transparency, timing, labeling, distribution channels, and sometimes even content restrictions. If authorization cannot be shown, the legal framework may shift toward other categories: unauthorized third-party content, platform liability questions, or broader concerns about misinformation and electoral integrity.
But the left-wing groups pushing back argue that Bolsonaro’s team is attempting to exploit the ambiguity that AI introduces. Their filings emphasize that the avatar was not a random novelty; it was presented in a political context and used in a way that would reasonably be understood as connected to Bolsonaro’s public persona and campaign interests. They contend that the legal system cannot allow political actors to benefit from AI tools while escaping responsibility simply by claiming lack of direct consent after the fact.
This is where the case becomes more than a dispute about one man’s permission. It becomes a test of how courts will interpret intent and control in an environment where political communication can be produced by contractors, agencies, or automated systems. In traditional media, authorization is often easier to document: a candidate signs off on a script, approves a video, or appears in person. With AI avatars, however, the chain of production can be complex. A synthetic image can be assembled from publicly available footage, voice samples, and machine-learning models, then projected or distributed through channels that may not require the candidate’s immediate involvement.
The legal question, then, is not only whether Bolsonaro personally approved the projection, but whether the circumstances surrounding its creation and deployment are sufficient to infer authorization or at least responsibility. Courts may be asked to consider what “authorization” means when the candidate’s likeness is used to generate speech-like content. Is authorization limited to explicit approval? Or can it include implied authorization—such as commissioning, directing, funding, or allowing the use of a digital likeness in a political context?
Brazil’s election rules, like those in many countries, were not written with synthetic media in mind. That creates a gap between the technology and the legal language. Judges must decide whether to stretch existing concepts—like campaign endorsement and official political communication—or whether to treat AI-generated projections as a new category requiring updated standards.
One of the most consequential aspects of the case is how it frames responsibility. If the court accepts Bolsonaro’s argument that he did not authorize the projection, it could set a precedent that makes it harder to hold political figures accountable for AI content unless there is clear proof of consent. That would likely encourage a strategy of plausible deniability: political teams might use AI to create persuasive content while keeping formal documentation minimal, then contest responsibility if challenged.
If, on the other hand, the court sides with the left-wing groups and finds that the projection was effectively authorized or attributable to Bolsonaro, it could push Brazil toward a stricter standard. Such a ruling would signal that political actors cannot hide behind technicalities when their public identity is used to generate synthetic political messaging. It could also lead to more robust compliance expectations—documentation requirements, clearer labeling practices, and tighter controls over vendors and contractors.
Either outcome will influence how campaigns operate. In the short term, it will affect how political teams manage AI projects: whether they seek explicit approvals, how they contract with AI providers, and what records they keep. In the longer term, it could shape how Brazil’s election authorities draft future guidance and how courts interpret similar disputes.
The case also highlights a deeper tension in democratic governance: the difference between legal responsibility and moral responsibility. Even if a candidate did not explicitly approve a particular AI projection, the political impact may still be real. Synthetic media can alter voter perceptions quickly, and the harm—confusion, manipulation, or erosion of trust—may occur before any legal determination is reached. Election law traditionally aims to prevent harm through rules applied before or during campaigns, but AI content can be produced and disseminated at speeds that outpace enforcement.
That is why the procedural details of this case matter. The court is not only weighing facts about authorization; it is also considering how quickly and effectively legal remedies can be applied to AI-generated content. If the legal system moves too slowly, the content may already have done its work. If it moves too aggressively without clear standards, it risks chilling legitimate political expression and innovation.
There is also the question of evidentiary burden. Proving authorization can be difficult even in conventional cases, but AI adds layers of uncertainty. Who has access to the production logs? Which party controls the model training data? Can the court obtain metadata showing who commissioned the work, who approved it, and when? If the avatar was generated using publicly available images and voice samples, the defense may argue that the candidate’s consent was not required. The challengers may respond that consent is required precisely because the likeness is being used for political persuasion.
In many jurisdictions, courts are beginning to treat synthetic media as a matter of both speech and safety. The legal frameworks are evolving, but they often struggle with a common dilemma: AI content can be protected as expression, yet it can also be harmful if it misleads voters or undermines electoral integrity. The Bolsonaro avatar case sits at the intersection of these competing principles.
A unique feature of this dispute is that it is happening while Brazil is actively debating how election rules should apply to AI-generated content more broadly. That means the court’s decision may be read as a signal about the direction of policy. Even if the ruling is limited to the specific facts of the case, it will likely be cited in future challenges involving deepfakes, synthetic voices, and AI-generated campaign materials.
The court’s reasoning may also address the role of intermediaries. AI projects often involve multiple actors: software developers, marketing agencies, political consultants, and platform distributors. If a candidate’s team claims non-authorization, the challengers may argue that the political apparatus as a whole should be responsible for what it deploys. Conversely, if the court holds candidates responsible for AI content without clear proof of consent, it may increase pressure on campaigns to centralize approvals and reduce outsourcing—potentially changing the economics of political communication.
Beyond the courtroom, the case reflects a broader societal shift. Digital avatars are no longer confined to entertainment. They are becoming tools for political messaging, capable of simulating presence, voice and charisma. In a country where political communication has long been mediated through television, radio and social platforms, AI avatars represent a new layer of mediation—one that can make messages feel more personal while reducing the friction of production.
That is why the legal debate is so intense. Voters may not distinguish between a human speaker and a synthetic projection, especially when the avatar is designed to look and sound authentic. Even when the content is not technically “deepfake” in the strict sense, the effect can be similar: a synthetic representation that leverages trust in familiar faces.
The challengers’ legal strategy appears to focus on preventing that trust from being exploited without accountability. They are essentially arguing that electoral rules must adapt to the reality that political identity can be algorithmically reproduced. If a candidate’s likeness is used to deliver political messaging, the law should treat it as campaign communication unless there is a clear, verifiable basis for denying that connection.
Bolsonaro’s legal team, meanwhile, is trying to preserve a boundary between political figures and AI outputs created by others. Their argument implies that the legal system should not automatically attribute AI-generated content to a candidate merely because the candidate’s image or persona is involved. They want the court to require stronger proof of authorization, perhaps including evidence of commissioning, approval, or control.
This is where the case could become a landmark for Brazil’s AI election jurisprudence. Courts often rely on analogies to existing categories—advertising, propaganda, endorsements, and unauthorized third-party content. But AI complicates analogies because it can blur the line between creation and endorsement. A synthetic avatar can be generated from a candidate’s public materials, but the act of projecting it in a political context may still be an endorsement. The court must decide which element matters most: the source