Extremist propaganda is being rebuilt with the same AI apps teenagers use to make football fan edits and dance videos. A six-month investigation published September 8 by the Institute for Strategic Dialogue found 150 of these videos across 71 TikTok accounts, with more than 5.4 million views between them.

Here's what the researchers found, and why it matters for anyone raising or teaching kids online.

ISD's own definition: "low-quality digital content produced in part, or in whole, by artificial intelligence in service of the 'media jihad.'" The word "slop" is doing real work there. This isn't sophisticated. It's fast, cheap, mass-produced content — the extremist branch of the same AI-generated filler flooding every platform.

That's a change. For most of the last decade, extremist media was recognizable precisely because it was polished and consistent. Official outlets maintained house styles; unofficial support networks copied them. ISD, which maintains a running census of more than 107 unofficial support outlets, notes these "have predominately maintained similar design and narrative standards." Consistency made the content easy to identify — and easy to automate detection against.

Generative AI broke that pattern.

ISD describes the content as "mainstream AI-templates, such as fan-cam filters, dance videos, and cartoon doppelgängers of official content, adapted by supporters to feature jihadist ideologues and attackers instead."

Three formats dominate, and all three are borrowed from ordinary internet culture:

Stadium fan cams. During the World Cup, everyone used AI templates to drop themselves into crowd shots at football matches. The same templates, unmodified, are being used to place terrorist figures in the stands. One such video drew 187,000 views on its own.

Dance videos. Using PixVerse AI and Viggle AI — consumer apps, freely available — producers animate attackers dancing to trending audio, sometimes synchronized over the death tolls of the attacks they carried out.

Anime and cartoon conversions. CapCut filters turn official propaganda images, videos and even news bulletins into anime. One account posted 25 anime versions of official material between June and August 2026, collecting more than 137,000 views collectively.

The most-cited example in the report: a user who rebuilt a real execution video as a SpongeBob cartoon, matching the original scene shot for shot in cartoon form. A second video by the same account, in the same style, drew more than 34,500 views.

This is the finding with the sharpest operational edge. ISD writes: "The anime versions of the official propaganda appear to bypass moderation systems, but it remains unclear why that is the case."

Read that carefully. It's an observation, not an explanation — the researchers are reporting a pattern they can see and can't yet account for. But the plausible mechanism is uncomfortable: automated moderation is trained to recognize what extremist media looks like, and a cartoon doesn't look like it. The content survives the conversion. The signature doesn't.

ISD makes the audience effect explicit too: rendering ideology through SpongeBob "makes ideological viewpoints both easily accessible and resonant to younger audiences."

Perhaps the most useful part of the report for parents is that it complicates the picture of who produces extremist content.

ISD sorts producers into three groups. There are ideological supporters of recognized groups, making AI versions of official material. There's a subculture drawn from true-crime fandom communities, which treats attackers as figures to venerate and content as a way to troll. And there's a third group ISD calls "shitposting slopagandists" — users the report describes as "non-ideological and typically interested in virality over ideological dimensions of the content they produce," people who self-identify as trolls or LARPers.

All three groups consume and reshare each other's material regardless of intent. ISD found members of each category interacting in the same comment sections.

Europol's 2026 Terrorism Situation and Trends report, quoted by ISD, describes the same convergence: "the boundaries between different terrorist ideologies and other forms of violence were increasingly blurred," with perpetrators "occasionally displaying less ideological depth." The latest United Nations Monitoring Team report goes further, citing "some evidence to suggest that ideology can be progressively less important for recruitment the younger the potential recruit, and that the appeal to violence itself can be a driver."

That is the shift. Not ideology first, then violence. Violence as content, first.

Not by searching for it. ISD's researchers report that once their observation accounts had viewed five videos featuring this material, the accounts "began receiving recommendations for similar content in their For You Feeds without further prompting." The researchers never followed, liked or commented — they only watched.

Five videos. That's the distance between accidental exposure and an algorithmically reinforced feed.

ISD reports that 42 of the identified accounts were still active on 2 September. By 8 September, all of them had been removed.

Both halves of that sentence are worth sitting with. The removals happened. They also happened after a six-month research investigation surfaced the accounts — which means the detection that worked here was human, funded, and slow.

The instinct is to look for a list: words to block, hashtags to watch, apps to remove. This report is an argument against that approach. The content moved into formats that are indistinguishable from harmless ones, made with tools that are overwhelmingly used for harmless things.

What generalizes better:

Talk about format, not just content. The lesson isn't "cartoons are dangerous." It's that anything can be rendered as anything now, and the wrapper tells you nothing about what's inside. A kid who understands that a cartoon can be a repackaged real video has a defense that survives the next format shift.

Take the "it's just a meme" defense seriously as a signal. ISD's finding is that a large fraction of this material is made and shared by people who genuinely don't consider themselves ideological — they're chasing engagement. That's not reassuring. It means the ironic framing is the delivery mechanism, and the sincerity question is the wrong one to ask.

Know that the feed does the work. Five videos of passive watching was enough to change what the algorithm served. That reframes the conversation from "don't seek this out" to "notice what your feed starts giving you, and say something when it turns."

Don't over-index on any one platform. TikTok is where ISD found the concentration, and TikTok removed the accounts. The tools, formats and communities are portable.

Source: Institute for Strategic Dialogue: https://www.isdglobal.org/digital-dispatch/generation-slop-jihad-the-decentralized-media-jihad-gets-sloppy/

Want practical AI guidance for parents and educators every week? Subscribe: https://www.aibyage.com/?modal=signup&utm_source=beehiiv&utm_medium=newsletter