I used AI to generate a fake persona to talk to an SMS Scammer
I was the type of kid who would turn over big rocks just to see what horrors lived underneath. I suppose that hasn’t changed much for me as an adult. That’s why when I started getting a higher volume of SMS Spam / Scam messages, I decided that I needed to know more about the motivations and workings of these types of modern attacks. You may have seen something similar:
Good hook here, seems like a harmless, genuine case of mistaken identity. I’ve had several of these types of messages in the last few months, and so I was ready for this one. I found that something about the way I was responding made me either seem like a bad “mark” or otherwise revealed me to be disingenuous / not worth scamming. This time, I decided to try something different, and rather than typing a response myself, I took the text message, and put the content into OpenAI’s text generation playground. A text generator that takes any text you provide and using the latest models comes up with what’s next. In this case I filled it with a simple prompt:
Mark just got a text message:
> Hi Gina, when are you going to travel to Japan with your sisters
I hit generate, and every time the other person responded, had more conversation “generated” for Mark. The hope being that a friendly AI would seem more gullible than my earlier attempts at seeming simple.
At this point I felt a little bad, as it seemed this was a genuine case of mistaken identity (This doubt happens to me often, even once I know that the other side is false) and so I came clean:
In the meantime, I decided to gather what I could about my latest “friend”. The number came up as Sacramento, and owned by “Astound Broadband” which provides an API based SMS hosting service — Perfect for bots. That, and the odd response…