Scam detecting chatbot for you and your vulnerable loved ones

[hearty flume]

Let me start by asking a moderator to delete this if it's not appropriate, as it is not really HA-related. (Sorry!)

For a while now, I've been working together with my colleagues on perfecting an AI-powered chatbot which can receive text, images, URLs and QR codes to see if they may be related to a scam or fraud. It's called Bitdefender Scamio and it may help you, but mostly your vulnerable loved ones determine if that delivery SMS is legit, if that job offer that sounded too good to be true is true or if the website from where they are about to purchase smart light switches with 90% discount might end up draining their wallets.

It's free (you do need to set up an account, though) and it's available 24/7 on already popular chat platforms like WhatsApp, Facebook messenger, Discord and also in a dedicated web app, so basically you'd just be chatting with it, asking whether something you've seen is shady or not. When it does detect anything fishy, it also tries to explain what red flags to look out for and it can even deep-dive in learning more about the type of scam/fraud you're experiencing, all in context, without leaving your favorite chat app.

Here is a short demo of how it works https://www.youtube.com/shorts/iAMIqJ7EJ1g and a bit more info on the how's and why's https://www.techradar.com/pro/this-new-free-ai-chatbot-from-bitdefender-protects-discord-users-with-ai-technology-to-identify-potential-scams

Hopefully it will protect you from ever growing online fraud 🤞

Here are a few examples of how it works

[celest folio]

I love the idea but I’m worried that it’s still too complicated to use for the people that would need it the most.
I will have to try it out, but it looks like it might be a solution for some people I know that aren’t afraid of tech but aren’t good at identifying scams.
What is the likelihood of hallucinations?

[hearty flume]

There are multiple layers of detection, LLMs are basically just used to wrap stuff in words. The risk of hallucinations is never 0, but it should not mark stuff "safe" as a result of hallucinations.

Re: complicated - give it a try. It's a semi-guided process and it's really just a chat with someone on a platform you're already confortable with.

[neat bridge]

is there a api i can use so i can integrate it in my asterisk system?

[hearty flume]

No. It's not that kind of open product 🙂

[celest folio]

The person I'm thinking of is elderly and doesn't handle anything new well at all. If it requires multiple steps than she will not remember them. But that is a high bar to cater to someone like that.
I gave it a go and it works really well with text messages. I know a bunch of people that this will help a tone. So, thank you for sharing.
Side note, I don't think it mentioned working with emails, but I tested a few and it didn't do as good on those. but, it's easier to hide malicious content in an email. Just thought I'd mention the results