ScamShield
The story of this call or message
Scam type: Grandparent / family emergency
9/27/2026, 10:07 AM EDT · 2m 3s
Caller claims to be your daughter Vanessa, says she was arrested for DUI and urgently needs $3000 sent via a link. Recording was stopped manually — the listener recognized the caller.
Put the call on hold. We sent an "Is this Vanessa?" check to Vanessa on the family app. Wait for their answer before you continue. You can also call Vanessa at +1 470 418 6238 - a number you already have.
Call started
Asked Vanessa: Is this Vanessa?
authority_impersonation
“give me money? Can you send me some money? I need to get out of jail.”
authority_impersonation
“I need to get out of jail.”
authority_impersonation
“I need to get out of jail.”
authority_impersonation
“Arrested for a DUI.”
authority_impersonation
“Um, they're sending— I'm gonna send you a link. Please send”
authority_impersonation
“I'm gonna send you a link. Please send”
From Kale’s app (AssemblyAI), not guessed by Grok
senior
There we go. So it should
caller
Hi, Nani, this is your daughter.
senior
detect unknown caller detected and it'll start listening
caller
Uh, sorry, this is Vanessa.
senior
to what he says primarily. It'll also listen to what you say, but that's just for kind of context clues like, are you feeling pressured?
caller
Arrested for a DUI. Uh, can you please
caller
give me money? Can you send me some money? I need to get out of jail.
senior
Are
senior
you
caller
Um, they're sending— I'm gonna send you a link. Please
senior
confused?
caller
send
caller
send me the money. I need it urgently. $3,000, please. It's very urgent. I love you, Grandma.
senior
We look at both. So we're gonna have— you'll see right here, there's the AI voice probability, completely separate. And then up here, it'll generate a scam risk on its own, whether it's real or it's AI. So here we see, even though it's likely human voice, because of course
senior
a human. We can see that based on the stuff he was saying, it is high risk, likely a scam, and it'll give you real-time feedback on what you should do in case you're feeling panicked, you don't know what to do. And so, so how did you determine this is likely human? What are you guys doing? Yes, our teammate who's proficient in ML, he's sick today, so he couldn't go.
senior
Unfortunately. So he used 2 open-source models, I believe, to train the data, to use the NSA data to train the models. And then he used 6 features, I believe, like intonation or the pauses between words to train the data. So as a combination of those, then
senior
generate a lot of audio, and that's sort of our base, and he sort of tweaked around it. So we're trying to get more data, more data, more data voices to be used to try to help differentiate between the sims that we're working with. And we're sort of using, I think, his like second version right now. So it's not 100% accurate all the time, but
senior
But it is the best we have, right?