TL;DR
An organised crime group based in Albania ran more than 200 profiles and some 900 videos advertising routes into Britain. What unravelled it was a branding decision: the accounts kept using AI-generated lions as a recognisable marker for would-be customers. An investigator at the National Crime Agency spotted one profile, and TikTok used the motif to trace the rest. Every account has now been taken down.
A calling card that worked too well
Researchers labelled the cluster “The Golden Lion”. The images functioned as a signal to migrants that an otherwise blank profile belonged to someone able to move people from Albania towards the Dover crossing. Its usefulness lay in being oblique — a lion carries no obvious relationship to trafficking for automated classifiers, beyond a loose echo of the royal coat of arms.
That obliqueness is exactly what made it a fingerprint once a human noticed it. Investigators built out the network by mapping which accounts interacted with which, then examined where those accounts originated, work complicated by VPNs, proxy addresses and eSIMs that let a phone number from anywhere in Europe front an account run from elsewhere.
The scale of the problem is not marginal. Home Office figures put the share of small boat arrivals who interacted with such posts en route at 80%, and Mike Hulett, who leads the NCA’s online organised immigration crime team, notes the figure is 18 months old and probably understates it.
Evasion has professionalised
Five years ago gangs advertised routes openly. Now they misspell deliberately, substituting numerals and punctuation so “Europe” becomes “Eur0p”, push text into the video frame rather than the caption, lean on emoji and imagery, and move any real conversation onto encrypted messaging within a few exchanges. Hulett’s phrase for it is “hiding in plain sight”.
Looking forward
Across January to June, TikTok pulled 10,000 UK videos that breached its human exploitation rules — 0.17% of everything it blocked in that period, which indicates how much rests on automated moderation. The lesson cuts against the prevailing assumption that AI content is the threat: here, a generated image became the identifying feature. As Hulett puts it, “We can’t arrest our way out of this problem.”