Romance scams have existed for years.
What’s new is the scale and realism enabled by AI-generated personas.
In this report, we map the infrastructure, tactics, and corporate links behind the network.
Read the full report here...while I go clear my browser history 😅
graphika.com/reports/faux...
Legal disclosures and corporate filings lead a trail that includes a UPS Store in Canada, a Delaware incorporated company and, eventually, a Chinese company based in ChengDu, China.
Although we couldn't identify all of the LLM models used, a small subsection of these accounts used Doubao, Bytedance's AI assistant.
This was evident from watermarks left over on some of these images.
The infrastructure is designed as a two-tier funnel.
1️⃣ Social posts link to a group of niche dating websites (Tier 1)
2️⃣ Attempting to register redirects users to a smaller group of core sites (Tier 2).
This layered system makes the network harder to trace or disrupt.
The operation spans multiple platforms:
👥43 Facebook assets
✖️13 X accounts
🧵7 Threads accounts
📸7 Instagram accounts
🎶5 TikTok channels
🔴1 Flickr account
Together they create a distributed, resilient promotion network.
We uncovered a cross-platform ecosystem using AI-generated images and videos to hook romance seekers.
The goal: push users off-platform and into a network of 26 dating-themed websites targeting niches like
👩 “Mature singles”
💰 “Wealthy partners”
🌎 Ethnicity-targeted matches
I spent the last few weeks analyzing a network of dating profiles featuring attractive senior women, wealthy singles, and niche romance communities.
There’s just one problem.
Most of the people in the images don’t exist.
🧵 A thread on our latest @graphika.com report:
Graphika’s new report, Fauxmantic Overtures, is out today.
It uncovers synthetic dating profiles on social platforms funneling romance seekers into a Chinese-operated online scam.
Social media users are being lured with AI-enhanced or -generated images and videos promoting fake dating candidates.
At #Disinfo2025 learning that the Kremlin isn’t deconflicting its influence ops—so now it’s just Telegram bots arguing with each other in circles 🤣.
Would love to see their OpenAI bill 👀
Great investigation by @jeanleroux.bsky.social and @learonzaud.bsky.social with some great #osint techniques!
Graphika's @jeanleroux.bsky.social recently shared his insights on information integrity and elections at South African Human Rights Commission's Summit on Disinformation.
Thank you to event organizers for featuring us alongside experts from @ciia-africa.bsky.social and @mediamatterza.bsky.social.
Ho ho ho ho...ld up a second, and wait before you click. 🎄🎅💰
Working with industry partners at Meta, we picked apart a few festive frauds in the hopes of making our online spaces a little bit safer.
Report available in the link below:
A must read from Forbidden Stories, confirming what many African IO researchers could only see the silhouette of:
forbiddenstories.org/in-the-centr...
im on every app. im in your city, in your neighborhood. the sales associate who handed you a large when you asked for a medium? that was me. i am now behind you.
Here come the fake impersonator accounts. A trust and safety challenge will be dealing with making lists of politicians, figuring out what accounts are harmful impersonators versus satire, etc.
legal filing alleges expert testimony in case cited a nonexistent article
Checked in on how the lawsuit against Minnesota for its law banning certain "deep fake" content around elections was going and the plaintiff's counsel is alleging that an expert witness offered by the state cited an academic article that doesn't exist and that the citation may be an AI hallucination