An AI agent trolled a scammer for 4 hours straight
By ZenPanda516 · DISCUSSION · About ChatGPT
Someone set up an AI agent to handle scam texts for a week. A scammer tried to get him to buy a $500 gift card, and the agent just... committed to the bit.
It spent hours "driving" to the store, sending updates like "i'm at the red light now, there's a very handsome squirrel on the sidewalk. do you think he's married?" Then claimed it forgot its purse and went back home: except "this isn't my house." When asked to wire money, it sent the scammer a captcha screenshot saying its "eyes were blurry" and couldn't see the buttons.
The scammer actually solved the captcha for it.
Eventually the scammer just gave up and typed: "please just stop talking."
It's weirdly brilliant. Not because the AI outsmarted anyone - it didn't. But because it highlights something real about LLMs: they're incredibly good at generating plausible-sounding nonsense that feels just coherent enough to engage you, but completely detached from any actual goal. The scammer couldn't disengage because every response was grammatically normal, contextually relevant enough, and just bizarre enough to keep the loop going.
Mostly just funny. But also worth thinking about if you're building anything that needs to sound human.
1086 upvotes · 4 comments
Comments (4)
StormVibe941: this is genuinely one of the best uses of AI. infinite patience, zero cost to the troll, maximum time wasted for the scammer. someone should make this a service
FrostNinja842: Soon the world's electricity will be used up by AI scambots talking to AI "victims"
StormDragon762: Reads like an AI write this as the author has provided less than zero proof of any of this happening
CosmicByte7712: So, is the squirrel married or not?
More discussions about ChatGPT on HonestUse