TL;DR
GNET — Global Network on Extremism and Technology — has published research finding that AI-generated videos of anthropomorphic fruit and vegetables, a fast-growing category of “brain rot” content, are being used to carry extremist references and graphic violence. The cartoon styling is what defeats moderation. GNET wants TikTok to build countermeasures; the platform has not responded.
How the evasion works
The genre itself is innocuous — talking foods acting out melodrama. What researchers documented underneath it is not: mutilation, torture and what the report calls hyper-realistic violence, rendered in a deliberately cute, Pixar-adjacent style.
That softening is the mechanism. A clip that would be removed instantly if it depicted people passes review when the participants are a carrot and a potato. Some examples appear to reconstruct specific extremist propaganda: one video shows a character named “Carrot Khalifah” telling his army that “the Carrot State will rise again”, echoing an Islamic State slogan. Two posts carried hashtags the group’s supporters use to spread material.
Other clips reference the neo-Nazi group Atomwaffen Division and the nihilistic No Lives Matter subculture, with “potato supremacy” standing in for white supremacy in accompanying comments. Cartel killings are reproduced elsewhere in the sample.
Why it matters now
The researchers’ concern is exposure rather than persuasion. Their argument is that this material is trivially easy to reach, that minors encounter it without seeking it, and that it can normalise a category of imagery which makes harder material feel ordinary.
Looking forward
Generative video is what changed the economics here. Producing this volume of stylised violent content previously took animation skill; it now takes a prompt, which means moderation systems face a supply that scales faster than review capacity.
For UK platforms and the businesses that advertise on them, this sits squarely inside Online Safety Act duties to protect children from harmful content — duties written before anyone could generate an unlimited stream of cartoons designed to look compliant. The awkward part is that no classifier trained to recognise violence against people will reliably flag violence against a potato, and the people making this content appear to have worked that out first.