AI Disinformation in the 2026 Elections: Brazil, Serbia, Israel
In three weeks of October 2026, Brazil, Serbia and Israel went to the polls. In all three, AI-generated images, videos and deepfakes flooded social platforms. We ran the content through our detection tools. Here is the short version.
By the UncovAI research team · Published 29 October 2026
Brazil
Growth of AI-created fake content, 2024 to 2025 (Agência Brasil).
Serbia
Share of tracked disinformation cases targeting the Student List (EDMO).
Israel
Likud posts with AI-generated content by mid-October (Deutsche Welle).
Three elections, one playbook
Lula won in Brazil with nearly 52%. The Student List ousted President Vučić in Serbia with over 53%. Israel's vote left no side with a majority. The outcomes differ, but the tools look alike: fake news anchors, humiliating images of leaders and invented crowds.
Brazil: deepfakes are the main threat
Before the first round, Agência Lupa found 920 AI-assisted political posts, and 60% of them were deepfakes. The surprise is who made them: 47% came from ordinary users and about 24.5% from candidates themselves.
Serbia: AI was only part of the story
The Student List was the main target. Two AI cases were documented, including a viral image of Novak Djokovic confirmed through a SynthID watermark. Distorted quotes and false voting rules did the rest.
Israel: AI in official party campaigns
Parties used deepfakes, automated calls and bot networks in their own official channels. Germany's Foreign Ministry called one party video "completely unacceptable".
Can AI election fakes be detected?
We took six widely shared items from the Brazilian campaign and ran them through UncovAI. All six were flagged as AI-generated, five at 100% confidence and one at 91%. We also tested two clips outside that sample, and in one of them the video was flagged as AI while the audio track came back as human.
Which items? Where exactly did the heatmaps light up? Why did one clip split between video and audio? That is in the report.
Elections under AI-Driven Disinformation
Full data tables with every source, the six detection heatmaps, side-by-side comparison of all three campaigns and our methodology notes.
- The six Brazilian campaign items and their detection heatmaps
- Full Brazil, Serbia and Israel data tables with sources
- Documented campaign content, case by case
- A five-step method to check a suspicious election post
- Limits of the data and methodology notes
Get the full report
Three elections, one side-by-side analysis, with the sources behind every figure.
Download the full report →Want to test a suspicious image, video or audio clip yourself? Create a free UncovAI account.