Loading today’s stories
Product1 publisher3 min readPublished
ISD and Tech Against Terrorism describe cartoons and fan edits that carry no movement branding for a classifier to grab, which moves the moderation problem from known files to novel ones. TikTok disputes the methodology.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The pathway Adam Hadley of Tech Against Terrorism describes has two steps, and the second one belongs to the product team. A viewer meets glorification of an attacker inside something shaped like entertainment, and then the watching itself becomes the signal that pulls more overt propaganda into the next set of recommendations [6]. To a ranker, curiosity and commitment are the same event. Moustafa Ayad of ISD puts the concentration on TikTok down to a For You Page tuned tightly to user interests, a large existing volume of AI slop and a jihadist support ecosystem already settled on the platform [11]. He found only a limited number of these videos on Instagram and Facebook [9].
Divide the sample and it thins out fast: about two videos per account, and an average near 36,000 views each [19][18]. WIRED's account of the ISD and Tech Against Terrorism reports gives no removal rate, no time to detection, no view distribution and no platform-wide denominator [21], which are the four numbers a trust and safety lead would want before deciding whether this is a sprint or a memo.
The tooling is not the new part. The ISIS-affiliated news program News Harvest was already using AI-generated anchors in 2024 to publish quickly and cheaply [14], and one TikTok account has been turning the weekly al-Naba newsletter into anime [16]. Borrowed aesthetics have precedent too, in the early 2020s Alt Jihad co-option of alt-right visual style [15]. What changed is the look, and the look is what a lot of detection reads.
Hadley's sentence about branding is the one to reread if you own a queue: highly stylised AI-generated material, he says, makes violent Islamist extremist content more accessible to younger audiences while reducing the visibility of conventional terrorist branding [5]. Two axes sort the work. Whether an item carries recognisable movement branding, and whether it depicts or glorifies a specific attributable act. Three of those four cells are tractable. The awkward one is unbranded and act-specific, where a cage of cartoon fish in orange jumpsuits sits [3], and where the call rests on a reviewer recognising a video ISIS released more than a decade ago [22]. Staffing that cell takes people with case knowledge rather than more policy text.
Growth is the contested part. Ayad says the trend is new, still growing and has not peaked [10], while also reporting animosity from traditional supporters who doubt the producers are true believers, alongside accounts that share both official propaganda and the cartoons [17]. In the portion WIRED published, TikTok's response addressed methodology and its own prohibition on violent and hateful organisations rather than what its systems caught [12]. So the honest test runs on your own data, not theirs: whether time to detection on unbranded, act-specific items looks anything like time to detection on branded ones. The distance between those two numbers is the problem these reports are actually describing.
Ranked by verification strength, evidence, and original report placement.
Adam Hadley, executive director of Tech Against Terrorism, a UN-backed counter terrorism organisation, says the principal concern is the growing volume and diversity of material.
Hadley says highly stylised AI-generated content can make violent Islamist extremist material more accessible to younger audiences while reducing the visibility of conventional terrorist branding.
Hadley says users may encounter perpetrator glorification inadvertently through fan edits, cartoons, memes and other entertainment-oriented formats, after which engagement can drive algorithmic recommendations of related, more overt propaganda.
A new report from the Institute of Strategic Dialogue (ISD) calls "Slop Jihad" a trend where young supporters of Islamic terror groups like ISIS and al-Qaida appear to use generative AI tools to create fan edits, cartoons and memes designed to radicalise a new generation of online backing.
An Islamic State video showing men locked in a cage being lowered into a swimming pool and drowned has been reproduced as a SpongeBob SquarePants cartoon, with fish in orange jumpsuits inside the cage, and shared on TikTok.
Follow any of these and your For You feed starts watching them — no settings page required.
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.
A single researcher's count, run by one outlet
Every number in this story traces to Moustafa Ayad's sample, in a report ISD handed to WIRED and nobody else, with no method described. It rises above bare assertion on a few counts: outside specialists WIRED consulted agreed with the reading and Aaron Zelin put his name to it, the OECD had already placed TikTok in its top ten for terrorist and extremist content in 2024, and TikTok's own answer removes the referenced material rather than denying it was there. The SpongeBob remake is described in enough detail to be checkable; the 5.4 million views are not.
Real videos, small totals, no denominator
What is measured is the propaganda's reach, not a product's: 150 clips, 71 accounts, 5.4 million views, which works out to about 36,000 views a clip and roughly two clips per account. On TikTok's scale those are modest totals, and the story offers no share of anything to compare them with. The sturdier signal is persistence rather than volume — accounts with pre-announced backups, content Ayad says lasted months, and takedowns that arrived only after a reporter forwarded links.
Conclusion outruns the sample
ISD's phrase is a fundamental transformation in the aesthetics of jihadist propaganda online, and the SpongeBob image does a great deal of the persuading. Underneath sits a sample of 150 videos on one platform, an unmeasured growth claim, and Ayad's own concession that some of the wilder material is hard to identify as terror-related at all. The mechanism Hadley describes, unbranded content that classifiers built for known files and known logos will not catch, is the substantive finding here and gets less room than the cartoon.
Naming a trend is the product
ISD and the UN-backed Tech Against Terrorism exist to find, label and publicise this material, and a coined term released to a single large outlet is how such work reaches funders and policymakers. Exclusivity gave ISD the first frame before anyone could check the sampling. On the other side, the party calling the methodology vague is the one that would carry the cost of catching unbranded cartoons at scale, and it says the disputed material is gone. WIRED's own outreach became the trigger for takedowns, which makes the outlet a participant in the enforcement event it reports.
Direction firmer than the magnitude
That unbranded AI remixes of terror group material are circulating on TikTok is about as solid as a single-source story gets: two counterterrorism organisations reached it separately, outside researchers endorsed it, and the platform deleted the examples. How much of it there is, whether it is growing, and whether recommendation systems push viewers from cartoons toward overt propaganda are all carried by one researcher's judgement without measurement.
product
One suspension, four students, and an AI image nobody in the chain could delete1 publisher
product
Senators want every TikTok test that disabled a safety feature, not just the one that hit 15 million3 publishers
product
How a short GNET essay on AI slop became a settled fact about children's brains1 publisher
product
A $500 fake cleared Rolex's own London workbench, and staff called it beautiful1 publisher
Publishers with included, body-backed reporting in this cluster.
1 article · September 8, 2026