Detecting Romance Scams & Catfishes in 2025

The Evolution of the Scam

Romance scams (sometimes called 'Pig Butchering' when crossing over into crypto fraud) have evolved. In 2025, scammers no longer steal single images from Google; they use Generative AI (like Midjourney or Stable Diffusion) to create entirely synthesized human faces, coupled with LLMs to manage the dialogue.

Standard reverse image searches will fail against an AI-generated face because the image has never existed on the internet before.

1. Detecting AI Synthetics

OSINT analysts train to look for the artifacts that AI models still struggle with.

• Background Topology: Does the background melt together? Are the lines of a bookcase physically impossible?
• Lighting Inconsistencies: Pay close attention to the reflections in the subject's pupils.
• Voice Analysis: If a target uses voice notes, deploying deepfake detection models can quickly authenticate (or invalidate) the audio stream. (See our Verify Someone's Identity workflow.)

2. Infrastructure and Telecom OSINT

When visual analysis is inconclusive, pivot to the infrastructure.

• VOIP vs. Sim: Is the provided phone number a physical SIM registered to a major carrier, or is it an untraceable Voice-over-IP (VOIP) number like Google Voice or TextNow? Our Deep Search workflow handles this rapidly inside the broader evidence review.
• IP Grabbing: While engaging with the subject, sending a benign, trackable link (a 'canary token') can reveal their true IP address. If the target claims to be a soldier in Ohio, but their IP pings a data center in Lagos, the investigation is essentially complete.

3. The Scripted Engagement

Scammers operate in coordinated networks using scripts. Taking a specific, idiosyncratic phrase the target uses and searching for it in quotes ("exact match") on Google can often lead to anti-scam forums where victims have posted the exact same script.

Protecting vulnerable individuals and high-net-worth targets requires systematic verification. Trust nothing without OSINT validation.