Brazil's TSE deepfake ban applies only to propaganda

You must now treat synthetic media bans as turning on legal qualification of content, not on detection alone. Document the message, audience, language and context of any AI-generated material separately from whether it is synthetic, and verify compliance with mandatory labelling rules as a distinct step. A detector result is not a compliance finding.

An analysis of 16 October 2026 of how Brazil's electoral court applied its deepfake ban to material lacking AI identification argues that the ban reaches synthetic content only when it qualifies as electoral propaganda, and that qualification depends on message, language, audience and context rather than on the technique alone. The Tribunal Superior Eleitoral ordered removal of three AI-generated posts, including a simulation of a political figure that lacked the mandatory AI identification required under Brazilian electoral law, despite having a caricatured appearance. In our assessment, this decision establishes a three-part framework that separates synthetic detection from legal qualification from labelling compliance—a distinction that reshapes how you should approach AI-generated material in regulated contexts.

What does the TSE ruling actually require?

Brazil's electoral court defined deepfake as synthetic or manipulated content with sufficient realism that creates, reproduces or alters the image, voice or utterance of a person. Critically, the court then ruled that the ban applies only when such material qualifies as electoral propaganda, a legal category assessed on content, language, addressees and context. The court also imposed a separate transparency duty: material created or edited with AI must carry express, clear and accessible identification. The TSE's October decision applied this framework to three posts. One was a simulation of a political figure that the court found breached the identification duty. The court noted that the material had a caricatured, fanciful appearance, yet that did not exempt it from the labelling requirement. This tells you that synthetic detection and legal qualification are not the same question.

Which separate duties must you keep distinct?

• Synthetic detection — establishing whether material is manipulated or artificially generated through technical means.
• Legal qualification — assessing whether the content, message, language and circulation context place it within the statutory category of electoral propaganda.
• Mandatory labelling — ensuring that AI-generated or edited material carries express identification as required by the applicable rule.
• Compliance verification — confirming that the labelling duty has been met, independent of whether the material is deceptive or caricatured.
• Removal grounds — material may be ordered removed where it qualifies as electoral propaganda and the mandatory AI identification has not been complied with, regardless of whether the synthetic nature is obvious to viewers.

What controls should you put in place before publishing AI material?

1. Document the content, audience and context — record the message, the language used, the addressees and the circumstances of circulation separately from any technical analysis.
2. Apply a fixed labelling rule — establish a standing requirement that all AI-generated or edited material carries clear, accessible identification before publication, without exception for caricatured or fanciful material.
3. Conduct a final compliance check — verify that the mandatory AI identification is present and meets the applicable standard before material goes live.
4. Separate detection from qualification — do not treat a detector result as a compliance finding; use technical analysis to inform your assessment, but assess legal qualification and labelling duty separately.
5. Record the basis for any decision — document which material was reviewed, what labelling was applied, and the reasoning for any decision to publish or withhold.

How does this fit with evidence rules in other jurisdictions?

The Irish Courts Service issued practice directions on 1 September 2026 requiring that material created with generative AI be independently verified under human control within the proceedings to which they apply. The directions also prohibit the use or presentation of fabricated, manipulated or deepfake material to the court, and warn of costs decisions, rejection of pleadings or disciplinary referral for non-compliance. A peer-reviewed study notes that Polish and EU rules do not establish a separate regime for AI-generated evidence; such material is admitted and weighed under free assessment of evidence, with authentication possible through expert examination. The study emphasises that labelling under the EU AI Act does not necessarily indicate that forged material has been created with compliance in mind, and that the right to contest evidence remains a central safeguard. The Commonwealth Lawyers Association has advocated layered authentication through provenance, integrity, independent corroboration, forensic examination and genuine contestability, with the judge remaining responsible for admission, weight and procedural fairness.

What can tooling do, and what remains your responsibility?

A detector can establish that material is synthetic. It cannot establish that the material qualifies as electoral propaganda under Brazilian law, or under any other legal framework. It cannot verify compliance with a labelling duty. What matters is what labelling does and does not do: an AI label discloses that AI was used, but it does not by itself establish that the underlying material is authentic. The TSE line applies within Brazilian electoral law and is not globally binding. Its value is that it shows how a supreme court translates an abstract norm into qualification, message and labelling, each tested separately. Your professional judgement remains essential: you must assess the legal category, the compliance requirement and the consequences of non-compliance in your own jurisdiction, and you must document your reasoning before you act.