O2's AI granny knits tall tales to waste scam callers' time Watch out, scammers. O2 has created a new weapon in the fight against fraud: an AI granny that will keep you talking until you get bored and give up. O2, the mobile operator arm of Brit telecoms giant Virgin Media (VMO2), says it has built the human-like AI to answer calls from fraudsters in real time, keeping them busy on the …
this is something i've envisioned a while ago, the only issue i foresee is with those telemarketers (still scammers, just not formally) since they can start a new contract with a phone recording/verbal agreement; most of the time it should be fine as long as the bot gives out wrong name and details but not sure how it works when they change your phone provider or ISP (for landline) since they can probably do it even if the personal details are wrong, not sure how strong is the cross-check (I'd like to be proven wrong)
"Give out a name and address?"
If they have your phone number, they have your name and address already. All they really need to do is verify who they have on the phone. If you left your phone lying about and gran picked up the call, and she isn't all there these days, it's definitely not you that agreed to sign up.
"this is something i've envisioned a while ago, the only issue i foresee is with those telemarketers (still scammers, just not formally) since they can start a new contract with a phone recording/verbal agreement;"
The dodgy ones try to get you to say "yes" to something so they have that to edit in. One of the ploys is to ask if you can hear them ok straight off before you know that it's a telemarketing call.
Yeah, I'd love to have it on my phone - set to answer anything coming in from a blacklist.
The other thing that struck me is that scammers will soon, if they are not already, will be using this kind of technology to generate scam calls.
So AI scam sources connecting directly to AI scam sinks - it could lead to a meltdown of the interweb as we know it!
ETA - Damn! Beaten to it by Wade Burchette.
> Yeah, I'd love to have it on my phone - set to answer anything coming in from a blacklist.
Aha! What you want is a CPU with a built-in NPU so that your PBX software can run a Daisy[1] model for you.
And to think, we've been mocking Intel and MS for pushing this stuff on us, but they knew there was a use case we could all get behind.
[1] ok, maybe just a Daisy-lite model: it only has the one grandchild and can witter on about about knitting but just never got the hang of crochet.
"Yeah, I'd love to have it on my phone - set to answer anything coming in from a blacklist."
I let my phone go to voicemail if the number isn't in my contact list. I'll look up the number or retrieve the message right away if one is left, but I do most of my business with established customers and, of course, family and friends are in my contact list and have other ringtones. I can always return the call of somebody legitimate quickly and make a polite excuse for not being able to pick up right away if I need to. The usual problem is not being able to get to the phone before it does go to VM. Running in my house is quite dangerous.
I have Lenny set up on my home phone server. I set up an auto-attendant to say "To speak to ***** or *****, please press 1", and then time-out a few seconds later to Lenny. The scammer's auto-dialers usually miss the prompt, and they end up speaking to Lenny.
Also, the built-in block list directs calls directly to Lenny. I have an allow list for friends or family that blocks the auto-attendant so that they don't have to catch the prompt.
With this set up, scam calls and robocalls are gone.
At my previous workplace, I had Lenny set up on the companies phone server. The ladies that answered the phone would transfer cold callers to Lenny. Lenny was "in charge" of a lot of things there. The funniest thing was when a sales-weasel showed up in our lobby. They said that they had an appointment with Lenny. I set the system to record any "Lenny" calls, and sometimes we would listen to them as a group on slow days. The record was someone that stayed on the line with Lenny for 38 minutes!!
A way of forwarding calls (without the caller knowing) to the AI number would be useful. Is this something that the mobile provider could offer?
It would presumably need a button in the phone app and a way for the provider to configure the forwarding number, but Android or iOS might be willing to add that, since it sounds like a pretty trivial mod. The real questions are: can it be done undetectably? is it legal to do so? and who pays for the call and for the AI?
I suspect it would be fraud to charge the scammer for such a call, so the mobile provider wouldn't be interested, but IANAL and it might vary with jurisdiction even if I were.
By coincidence, yesterday I went through the last week's worth of missed calls on my landline (I switched off the answerphone years ago) and had to add another EDF smart meter number to the blocklist... and, in honour of the international fraud week, there was also the number for an old-school 'broadband' scammer!
Most of the rest were from 'Mark, your local energy advisor' (formerly Luke) using one of thousands of numbers rented from Twilio Ireland, an SMS and VOIP provider for scammers and spammers. Strangely the have a TrustPilot rating of 1.1... I'm guessing the 28% of 5 star reviews were from scammers
More theatre, while not taking action to block scam calls. There's a conflict of interest here, as the phone companies get paid for all those scam calls. They know exactly where the scam calls come from, otherwise it would imply they don't know how to bill precisely, and their business model would collapse. They could just block the origin of the calls altogether until the phone company originating the calls gets its act together. We do this for certificate authorities on the internet.
Yes, that's exactly right. We know the complete number of the desk extension that dialed as well as the spoofed number. And yes, they could easily be blocked but we're paid to connect them. The call records show all. We do not have a recording of any calls though, so no worries about that. A judge has to sign a wiretap warrant for that to happen.
Now does the NSA record them all? I couldn't answer that as I do not know, I can only speak as a phone company tech. But yes, we know Habib at 011 91 0475 8368 76 is spoofing 800-432-1000 on his phone. He calls about 50 to 300 people a day.
> Because the way LLMs are trained is using real data
Sure, but firstly not your data, and secondly in this specific case I don't see why they would need to train the fake granny bot with any real bank account data. Its task is to prattle on, not to handle financial transactions.
Back when I had a website, I had an entire section in the code (invisible to the end-user) that was bogus email addresses. Some were just random email-like addresses, but some were specifically designed to crash any poorly-programmed scraper.
Send any replies to .@.com
You just know the scammers will begin to deploy AI voices for themselves. It is only a matter of time before "Daisy" AI talks to "Larry" AI from the local bank. What a delight that would be. Neither would get tired of talking to each other. That will be a conversation that never ends.
Then it will only be a matter of time before "Larry" AI proposes to "Daisy" AI and they get married over the phone.
And then there's telemarketers who're basically recordings played back when you answer. And these aren't new, they've been around for decades. Will the granny AI be able to detect that it's engaging a recording and terminate the call accordingly, or would we just end up with noise going both directions?
"Granny-bot keeps the scammers on the line while a weapon equipped Boston Dynamics Robo Dog infiltrates their location to say hello with extreme prejudice"
Those robodogs must have one helluva battery to get all the way to the sub-continent fast enough. Do they make them for Mr Lee's Greater Hong Kong as well?
Is the call termination fee greater or less than the cost of electricity for Daisy to answer those calls? If they're making a profit, they can deploy at scale. I somehow doubt it though.
It does beg the question though: at what point would such tactics be classed as fraud? They're encouraging a third party (albeit scammers) to call a number, which they've set to answer, and they get money each time that call connects. No court would convict O2 in this regard, but the lawyers would stand to make some money.
If automated i.e.
Caller: This is your credit card company Please confirm the following payment to Amazon.... or other scam. Press one to speak to an agent...
Me: Redirect call to Daisy
If a human i.e.
If automated i.e.
Caller: I'm calling from your credit card company Please confirm the following payment to Amazon.... or other scam.
Me: My wife does all our finances please hold the line and I'll get her
I then Redirect call to Daisy
Yeah, but theirs are tied to a much simpler chatbot model (Eliza) tied to voice recognition with specific word sequences triggering an offensive mode where the chatbot will then make you sorry you said what you did to it. And uses a TTS modeled on the voice of the PFY.
Edit: Story here => https://www.theregister.com/2000/05/19/bofh_discovers_voice_recognition/
24 years. Where has the time gone?
An alternative would be to patch a couple of incoming calls together and let the scammers try to scam each other.
I sometimes used that approach with email scammers:
"That sounds most interesting. I'm too busy to deal with it personally at the moment. Could you please liaise with my colleague at $AnotherScammersAddress"
After all, they're in the same line of business so it's only polite to introduce people with common interests.
Who has time to play around with spammers? Is your phone tied up while "granny" natters on about knitting?
My mobile phone uses Do Not Disturb priority to route all callers not in Contacts to voicemail. I.e. Whitelisting.
My home phone uses Anveo VOIP, which is the only provider I found using whitelisting, and programmable filters. This allows one to program a filter to whitelist, and more: listed callers can ring my home phone, unknown callers go directly to voicemail, and blacklisted calls get a "This number is not in service" message, and a hangup. I get email alerts identifying all calls except blacklisted. It's a great system.
Most spammers don't leave messages. I very rarely get a spam message, ~ 1-2 per year.
Occasionally, I do disable the filter to allow callers to ring my phone for, e.g. an important delivery where I must be contacted. This is rare too.
I really want to buy or rent AI Daisy, for the interminable months that are the "Medicare enrollment period" in the US, when every slimy *$(#$*# wants to tell you which supplemental old folks' insurance to sign up for. Twenty to thirty calls a day are not unusual, some spoofing numbers to try to get you to pick up thinking it's your neighbor.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
