Hacker News Reader

Comment by 55555 | Hacker News Reader

This is about to get way worse with AI. They'll be able to call you, say nothing, you'll answer, "hello?" and then hang up and by then they've cloned your voice and left a message on your parents' answering machine that sounds like you begging for them to transfer money because you're in danger. They'll be video calling everybody on linkedin with a job in accounting and will have the face and voice of their boss, asking them to transfer money. The situation is about to be terrible, and the only thing keeping it remotely in check is that most people choose to be somewhat moral.

View on HN

> by then they've cloned your voice and left a message on your parents' answering machine that sounds like you begging for them to transfer money because you're in danger.

You don't need to clone someone's voice for that; that's a scam common enough that one of my parents almost fell for it. Cloning someone's voice might tick up the likelihood of success by some amount, but would the investment in the research-and-infrastructure (spinning up the machines to clone the voice, researching the person you're planning on scamming to find their relatives' phone numbers, etc) be worthwhile when you can just dial every phone number in the tri-state area and just try it with a panicked-sounding young woman's voice?

Several elderly members of my family have unfortunately fallen for this. Voice replication is not a requirement.

I consistently have to remind myself that a lot of the readers on this site are significantly younger than myself, and have decades yet before this is going to be something that might impact them. Although I suppose they have grandparents.

Scarier still is the knowledge that today, I can spot a scam call immediately - but a future is coming where I'm the one who'll be likely to read off my credit card number to someone on the phone.

I agree, finding who is the parent of who increase the time and cost too much.

They just call at 3 am, cry and cry to make the voice difficult to recognize and then a friend/kidnapper/corrupt-police-officer ask for the money.

Yes and no; there is a tipping point, past which hardware/software vendors will be forced to adopt countermeasures and new ways to verify identity.

We will lose a lot though; the most prosperous societies really rely on high levels of interpersonal trust (which allows people to easily do things together, whether that's leaving their buggy outside when they go to buy a coffee or commercial relations). It would be a shame to see that get destroyed.

why would countermeasures destroy trust? if everyone's phone had an actual keyword and key and other details wouldn't that increase trust?

I get a call and there is a green checkmark saying "GPG VEIFIED" and then I know it's who I expect.

> why would countermeasures destroy trust

The countermeasures won't destroy trust - it's the fact that you can't just pick up the phone and work out if you're speaking to a real human. Or trust that posts on an internet forum are from real people, and not just planted by bad actors using generative AI.

I find it difficult to believe that AI will ever be able to clone your voice with just one word being said. There is a lot of variety in someone's voice thats not captured with the word "hello". Additionally, people have been saying for years that they can simply record you saying "Yes" and use that to agree to things but I have yet to see an epidemic of that being an issue either.

This is also the danger with these - apparent - AI meeting attendant bots being a thing now; organizations should ban them wholesale, as if they're unknown / unauthorised 3rd party ones, who knows what they'll do with the voice data of e.g. the boss.

The time to make money on e-hanko usable for everything is now.

Meh.

On the capability side, I say hello _when you call me_ differently than _when I call you_. Anybody who knows me will certainly have a sense something is off.

Socially, we all should be used to the idea of calling back using official numbers, and not closing an agreement on the first call.

But this is a numbers game. I imagine this vector would be more successful on third-world migrant workers, than in the first-world economies.

Reply on news.ycombinator.com