Neighbor News
This Is The Face That Will Break AI
Fraudsters are using AI to scam money out of family members. But AI can't fake every picture -yet.

Why it matters:
AI is being used to create fake videos of people that fraudsters use to scam money out of family members. But AI can’t fake every picture - at least not yet.
The Big Picture:
The FTC is getting a lot of consumer complaints about AI harms, including romance scams and phishing emails. Analyzing reports from the public from the Consumer Sentinel, the FTC has announced it “is keeping a close watch on the marketplace and company conduct as more AI products emerge.”
Find out what's happening in Washington DCfor free with the latest updates from Patch.
AI is being used to create images of loved ones to trick people into sending money to fraudsters. But there’s a catch - AI has serious limitations in how it creates images of people. If you know what they are, you can protect yourself against fraud.
Catch Up Fast:
CEOs of major companies are being impersonated by deepfake AI clones. A murder case defense relies on a voice message which may be a deepfake audio message to provide an alibi. Older parents are getting fake videos imitating loved ones and claiming that money is needed to save them.
Find out what's happening in Washington DCfor free with the latest updates from Patch.
How It Works:
Using an AI system available to anyone, called OpenAI, it is possible to “train” the AI to create pictures and videos of a target person that can fool most people. Then, a relative will be contacted and asked for money on behalf of what they think is their loved one - but is really a fraudster.
One Neat Trick:
Right now, AI can’t digest a very high resolution photo of a person without makeup. (Exception for Multi-Instance Learning, but that requires honing focus down to the parts of a high-resolution photo that are critical, and won't apply to a whole face.) All the blemishes on a highly detailed, up-close photo creates too much calculation for current AI to manipulate. By toning down the number of data points to create a manageable amount of data to compute, distortions are created in the end product. Et voila! The AI-generated video will look weird enough that you will know it is not human, let alone your long-lost Aunt Nellie. Like this:

To Catch a Thief:
- Insist on a close-up and high-resolution video conversation and that they don’t wear makeup.
- Ask them to hold up their hand and wave at you. This is a common failure point of AI videos: accurately showing four fingers on a hand. Honestly, it is also a common failure point for most artists.
- Ask them if they were adopted.
- Ask them if you were adopted.
The Problem:
This strategy will work for the time being. Eventually AI will be advance enough that realistic videos will be created from photographs. But not today.