Loading today’s stories
Product1 distinct publisher3 min readPublished
The Tribunal Superior Eleitoral is weighing a labelling mandate, a blackout on new synthetic political content and a ban on candidate-recommending systems, with about five weeks left for campaigns and platforms to comply.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
Of the three proposed obligations, labelling is the one that gets built. It is a field on an asset template and a badge in a render pipeline, and the TSE already required disclosure of synthetic content before the 2024 municipal elections [9], so Brazilian campaign shops and platform policy teams have the workflow sitting there already. The July avatar carried its disclosure partly because that habit existed [9].
Notice how little the disclosure bought. Lula's coalition petitioned the court anyway [5], and Flavio Bolsonaro's campaign answered that a clearly marked simulation is not a deepfake in the first place [6]. One file, two readings, both of them pointing at the same visible label.
Lay it out on two axes and the awkward quadrant is obvious. Axis one: does the content say what it is. Axis two: does it still move voters. Disclosure-based regimes, including the rules being developed under the EU AI Act, are built on the assumption that those two are joined, that removing the deception removes the effect [10]. The Next Web's reading of the July video is that the persuasion survived the label [11]. The same clip raises something the existing rules were not drafted to answer, since a man barred from public communication never physically addressed the crowd while his message reached voters close to as well as if he had [12].
The calendar is the part nobody can negotiate. Five weeks remain [13], which is 35 days [16], and the blackout window sits at the end of that count rather than adding to it [3]. Whatever definition the court lands on, campaigns, platforms and regulators get that runway to adjust [13].
The obligation to staff is the third one. Prohibiting systems from recommending candidates moves the question off the individual asset and into ordinary platform mechanics, which per The Next Web is considerably harder to monitor than checking whether one video carries a disclosure [14]. A label is inspectable by a person with a browser. A ranking system is inspectable by whoever can read the config and reproduce the output, and that is a different team with a different backlog.
The blackout has the same shape of problem. It only does work if someone spots synthetic material fast enough to stop it spreading, and that gets hardest when the material travels through private messaging groups instead of public feeds [15]. Enforcement, on that account, is the harder half of this [15].
The useful exercise, for anyone who has to answer for compliance in October, is to name the artefact for each obligation. For labelling it is a screenshot of the badge. For the blackout it is a date-gated publishing switch plus a log of what was promoted after it flipped. For the recommendation ban it is a ranking configuration, which most teams do not keep in a form a court can read. Where there is no artefact, that is where the exposure is, and the July case suggests it is not in the label.
Ranked by verification strength, evidence, and original report placement.
Prohibiting algorithms from recommending particular candidates would require regulators to examine ordinary platform mechanics, which is considerably more difficult to monitor than whether an individual video carries an AI disclosure.
Enforcement is likely to be more difficult than defining the rules: a blackout on synthetic political content around election day is useful only if someone can identify synthetic material quickly enough to stop it spreading, which is particularly difficult when content moves through private messaging groups rather than public social networks.
Brazil's Tribunal Superior Eleitoral spent Tuesday working through the question of what exactly counts as a deepfake, and was expected to rule on broader restrictions for AI-generated campaign material, Reuters reported.
The proposed TSE rules go beyond banning deceptive deepfakes: they would require AI-generated campaign material to be labelled, prohibit systems that recommend candidates, and introduce a blackout period during which new synthetic political content could not be published or promoted immediately before and after voting.
In July, Jair Bolsonaro, who is under house arrest and barred from public communication, appeared at his son's campaign launch as an AI-generated avatar that explicitly identified itself as a simulation.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · September 1, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
product
Microsoft never announced a China exit. Five years of filings did it instead1 distinct publisher
product
Anthropic's IPO case rests on a 2028 number, and your roadmap sits inside it1 distinct publisher
product
Reuters found 39 more banking-malware ads still running on Meta after India's advisory1 distinct publisher
product
Three EU employment laws now make the employer produce the evidence1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single outlet, and the ruling itself is missing
The Next Web is the only newsroom here, and the fact everything hangs on — that the electoral court was set to rule on the wider AI package — is credited to Reuters rather than reported first-hand. No proposal text, no petition, no order. The piece describes a draft and a hearing, then stops before the outcome, so a reader learns what Brazil might do and not what it did.
One avatar and an older rulebook
Everything that has actually happened in the world amounts to two things: the July avatar appearance, and disclosure requirements the court put in place before the 2024 municipal elections. There is no count of labelled campaign posts, no platform implementing the blackout, no enforcement action, no named service preparing for a recommendation ban. The rules in question have not begun operating on anyone.
The argument outruns the anecdote
The reporting is admirably honest about the parts it finds hard — enforcement, private messaging, the untested blackout. The overreach sits in its thesis: that disclosure regimes worldwide, the EU's included, are undermined because a labelled avatar can still persuade. One appearance at one campaign launch supports the possibility, not the conclusion, and no one here measures whether voters were moved. Calling this a case that forces a definition is a step ahead of what is on the record.
Both legal positions come from interested parties
Two of the three pillars of the dispute are supplied by people with something to win: Lula's coalition arguing the labelled video breached the rules, Flávio Bolsonaro's campaign arguing a declared simulation cannot be a deepfake. The Next Web says as much itself — the actors bounded by these rules have a direct interest in where the lines are drawn — which is candid, but does not make the filings neutral. Layered on top is a technology outlet's own interest in casting Brazil as the case Europe must watch.
Confident on the facts, thin on the outcome
What the avatar did, who objected, and what the proposal contains are all reported consistently and plausibly. But a single outlet, a Reuters-sourced procedural core, an unresolved ruling and a persuasion argument with nothing measured behind it cap how much weight this can bear. Treat the direction as reliable and any specific rule as provisional until the court's own text appears.