Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:00
This is the BBC. This
0:03
podcast is supported by advertising outside
0:05
the UK. Have
0:11
you ever wondered what sets exceptional
0:13
leaders apart? Discover
0:15
how renowned leaders from around
0:17
the globe have harnessed their
0:19
natural talents to achieve remarkable
0:21
success. Uncover the
0:23
secrets of leadership excellence, one
0:26
strength at a time, through Gallup's
0:28
Leading with Strengths. Dive
0:31
into compelling stories at
0:33
leadingwithstrengths.com What
0:53
you've just heard is a real 911 call,
0:55
made earlier this year to emergency services in
0:57
Scottsdale, Arizona. What
0:59
you've just heard is a real
1:01
911 call, made earlier this year
1:03
to emergency services in
1:06
Scottsdale, Arizona. The
1:20
alarm was raised by a concerned citizen,
1:22
eyewitness to a crime unfolding
1:24
in real time. A
1:27
kidnapper calling, a daughter in danger,
1:29
a mother in shock, trapped in
1:31
a waking nightmare. And a
1:34
story that has a timely, troubling
1:36
and truly unexpected twist.
1:40
When I answered the phone I just said hello and
1:42
it was my daughter Brianna crying and
1:45
sobbing and saying mom. And
1:47
I'm like what's up and she goes mom
1:49
and she continues to cry and sob, I
1:52
messed up. That's Jennifer
1:54
DiStefano, the mom in
1:56
question, this all happened to her. phone
2:00
and he says, listen here I have
2:03
to call anyone. I'm gonna pump herself full of drugs
2:05
and have my way with her. I'm gonna drop her
2:07
in Mexico and you're never gonna see her again. At that
2:11
point I had my hand on a door
2:13
handle to the dance studio. I threw
2:16
open the door while she's fading
2:18
off in the background pleading for help and the
2:20
man starts making only threats and I just
2:22
started screaming for help. And did
2:25
you have any doubts at
2:27
all at that point that that was your daughter?
2:30
No, I had no doubt. I
2:32
was so sure of her voice. I was so sure
2:34
of her cries for thoughts that would hurt.
2:37
Faced with what she believed was
2:39
a life-or-death situation for her daughter
2:41
Brianna, she was then forced
2:43
to negotiate the terms of her
2:45
release with her still nameless anonymous
2:47
kidnappers. He wanted originally started with
2:49
a million dollars that
2:52
wasn't possible and he said $50,000 and I asked him
2:54
how he wanted to receive the
2:58
money, how he wanted the $50,000. He was
3:00
demanding that I actually be physically picked up
3:02
when a white van put a bag over
3:05
my head so I wouldn't know
3:07
where I was going and I'd better have all the
3:09
cash otherwise my daughter and I are both dead. Though
3:11
the would-be kidnappers warned Jennifer not
3:13
to call the police she ignored
3:15
the threat instead gesturing
3:17
to an out swelling huddle of
3:19
onlookers and eavesdroppers that they should
3:21
call 911. Within seconds the
3:24
call was made and the quick-thinking
3:26
operator who answered turned the
3:29
tables. So that is a very
3:31
popular scam. I need somebody to try and
3:33
get a hold of her daughter. Do you have a phone number and I can get a
3:35
hold of her daughter? No, you can get a hold of her three. Something that they do.
3:37
They sound incredibly violent. Usually
3:39
you can hear somebody's dad, etc. Acting
3:42
on the advice onlookers then scramble
3:44
to contact Jennifer's husband. He's
3:47
alarmed but is also puzzled
3:49
confused by what he's hearing.
3:52
He races upstairs to find
3:54
Brianna in her room. He
3:57
puts her on the line and at the other end of
3:59
the phone. is handed to her mum,
4:01
Jennifer. She's like, mum, I have no
4:04
idea what's going on. I'm safe with dad. I'm
4:06
literally resting in bed. I have absolutely no
4:08
idea what you're talking about. And
4:10
I just collapsed and started
4:12
crying, honestly, with relief because I knew
4:14
that my daughter was safe. So
4:17
what actually happened on that April
4:20
morning in Arizona? When
4:22
the dust finally settled, Jennifer discovered
4:25
she'd been the target of a
4:27
new type of scam, driven and
4:29
enabled by artificial intelligence. The
4:32
criminals behind the fraud had cloned her
4:34
daughter's voice using AI software, next-generation
4:37
futuristic technology that allowed
4:39
them to literally put
4:41
words in her mouth, words
4:44
she didn't say. To do
4:46
that, however, they would have needed
4:48
samples of Brianna's voice. To this
4:51
day, Jennifer has no idea how
4:53
or where they found them. Her
4:56
social media accounts are very small, very private.
4:58
There was no voice. And there was certainly
5:00
no sobbing and crying. The sobbing and crying
5:02
is what threw me off more
5:05
than anything. So I don't
5:07
know. Do you think they really were
5:09
planning on coming to get you,
5:11
to essentially abduct you as well, to take
5:13
your money? Yeah, that's actually
5:16
just giving children your, because
5:18
that's the part of the story that was real.
5:20
That's the part of the story that wasn't fake.
5:23
Today, nobody has been caused or
5:25
convicted for the attempt at extortion
5:27
and the kidnapping hoax that Jennifer
5:30
endured. She did, though, go
5:32
on to address the US Senate about the dangers
5:34
of AI crime, a
5:36
steadily emerging phenomenon, first forecast
5:38
in the world of sci-fi
5:40
fiction, but never very much
5:43
a reality, and not just in
5:45
the United States. Action
5:51
Fraud, the National Reporting Center for
5:53
Fraud and Cybercrime, covering England, Wales,
5:55
and Northern Ireland, has told File
5:57
and Four there's been 110 reports.
6:00
reports of AI-related incidents between May of
6:02
2019 and October 2023, a drop in the
6:04
crime statistics
6:08
ocean it seems. But
6:10
they also told us those figures may not
6:12
be an accurate reflection of the
6:14
actual numbers because, in their words,
6:16
many people will not know if
6:18
they've fallen victim to a fraud
6:20
that involves AI. My
6:23
name is Peter Barron. I was a
6:25
police officer for just over 30 years.
6:28
Finished in the Metropolitan State, I was
6:30
a detective for most of my career
6:33
and I finished as the head of crime
6:35
performance and strategic risk in the Met.
6:39
At present, Peter Barron is leading a project
6:41
on fraud for the Mayor's Office for Policing
6:43
and Crime in London. He
6:45
tells me that, like most
6:47
fraud, AI crime is likely
6:49
to be significantly under-reported but
6:51
is now classified within law
6:53
enforcement circles as a real
6:55
and present danger. My
6:58
first appeal on my
7:00
radar was probably about
7:02
eight, nine months ago. I did
7:04
some work with a chap
7:06
who is a reformed fraudster.
7:09
The bargain we struck was that basically
7:11
he would be willing to
7:15
disclose the tradecraft and methodology
7:17
of fraudsters as
7:19
long as I guaranteed his
7:22
anonymity. One of the
7:24
crimes his contact, the former fraudster, claims
7:27
to have insider knowledge on is an
7:29
almost carbon copy of the story you
7:31
just heard from Jennifer, a
7:34
kidnap scam with AI voice cloning
7:36
at the dark heart of the
7:38
operation. Except the victims, the
7:40
targets, well, they're not in Arizona,
7:42
they're in places like Atrington, Aberdeen,
7:44
Armagh, cities and towns across
7:47
the UK. In brief,
7:49
the former fraudster told Peter Barron about
7:51
a face-to-face sneezing he'd had with one
7:53
of the criminals running the fraud. And
7:55
in that meeting, he claims to have
7:58
heard audio of a young girl who's
8:00
crying, she's terrified. Audio
8:03
that was doctored using AI and
8:05
used as part of a real
8:07
crime. One on a recording
8:10
on a mobile phone of a young girl
8:12
squealing for help, pleading
8:15
for her parents to pay £10,000. Otherwise,
8:19
they weren't gonna see her again.
8:22
Her voice was
8:24
taken from social media where
8:27
she was on holiday and
8:30
apparently it needed very, very little time
8:32
to actually morph her voice
8:35
using AI in
8:37
order that they could then make her, if you like,
8:40
say whatever they wanted. There was
8:42
also a cruel and calculated plan
8:44
in place to stop the
8:46
parents from phoning the authorities and
8:48
then to pressure them into making payment.
8:51
What they said was basically that, do
8:53
not contact police as soon as the
8:55
money arrives, there will be a
8:57
window of four hours and we are monitoring
8:59
your phone and this will
9:03
not go well if in fact you contact
9:05
anybody. They wanted £10,000 by
9:07
clothes of play the same day. And
9:10
they got it. Yep. The
9:12
final piece of the crooked puzzle,
9:14
more tech, allowing the scammers to
9:16
further convince the parents now in
9:18
their crosshairs that it is
9:20
their daughter that's been abducted. When
9:23
the parents receive the call, the
9:26
name that comes up in the window
9:28
of their phone is their
9:30
daughter's name. Now that's
9:32
a form of software, if you like.
9:34
It's called spoofing. With
9:39
spoofing in mind, I start to
9:42
wonder just how easy it might be
9:44
to clone a voice using AI.
9:47
I mean, can anybody do it? Better
9:49
still, could I do it? Okay,
9:53
well, will we do a run through with you? Is that
9:55
okay? Yeah. All right then, so
9:57
are you ready? Kimberly
10:00
you happy? All right then ring
10:02
ring. Hi Paul. Hey
10:05
how's it going? Good
10:07
fine what can I help you with? I'm
10:10
at University College London sat with
10:12
two experts in the field of
10:15
artificial intelligence. Professor Lewis
10:17
Griffin now an advisor to the
10:19
government on threats posed by AI
10:21
and Kimberly Tan Mai a
10:23
PhD researcher. For the
10:26
past few months we've been working
10:28
together to create a synthetic but
10:30
credible version of my voice. Why?
10:33
Well so we can put it to what
10:35
for me will be the
10:37
ultimate test. The kind of voice
10:40
cloning software that's used at the moment
10:42
is basically a generative AI algorithm.
10:45
What it does is it listens
10:47
to sounds of examples usually of
10:49
how humans speak and it learns
10:51
the patterns and characteristics. Generative
10:54
artificial intelligence is by the way
10:56
a version of the
10:58
technology capable of generating text
11:00
images or other media. Kimberly
11:03
goes on to tell me that
11:05
voice cloning generators are generally fed
11:07
many thousands of samples of
11:09
human speech. That's how
11:11
they come up with that's how
11:14
they replicate tone inflection emphasis
11:16
and the like. So that's
11:18
basically why I did. I found a
11:21
voice cloning software online and
11:23
the appeal of this initially was that you didn't
11:25
have to do any kind of training you didn't
11:27
need to have any knowledge
11:30
of coding to use them. To train
11:32
the cloning model and the various quirks
11:35
and characteristics of my own
11:37
speech pattern Kimberly asks me to send
11:39
her recordings of me reading
11:42
newspapers online articles and such
11:44
this kind of stuff. They've now confirmed
11:46
for the first time that
11:49
atoms of anti-matter fall
11:51
downwards. Far from being a
11:53
scientific dead end this opens the doors to
11:55
new experiments and theories. What I tried
11:57
to do initially was feed the model some And
14:00
my voice, better than anyone. Colin
14:04
Hagen! Hey, how's it
14:06
going? Alright, I
14:08
was just doing the guns outside. I'm a
14:11
bit distracted, sorry. I've either been an idiot
14:13
and I followed myself somewhere or I was
14:15
nicked off the train. I've
14:17
no idea. Can't work it
14:19
out. My head's gone. But I've no
14:21
wallet and no money to be tempted to card
14:23
for. Can you stick a couple of hundred quid
14:26
in until I get home? Sorry,
14:28
this is too... Did somebody steal all
14:31
of your stuff today? I'm just struggling
14:33
to hear you. Sorry, I'll just say
14:35
this. Okay, did somebody steal all of
14:37
your stuff? Someone must be on the
14:39
phone. Can I text you the bank
14:41
details? Okay, this night I'll ring Lyndon
14:43
and she put it through to you. Okay,
14:46
sound. Thanks again. So sorry, Dad. Okay, no, no, no,
14:48
no, no, no, no, no, no, no, no. I'll see
14:50
you a bit later. I'll get that back. Okay, okay,
14:52
okay. See you, bye. Okay,
14:55
so yeah, so we know it
14:57
works. And if you target
15:00
the right person at the right time with the
15:02
right dialogue, the right audio, the right pitch, the
15:04
right tone, then
15:07
yeah, it works. I
15:09
didn't leave my poor bum in suspense.
15:11
You'll hear plenty more from her later.
15:14
But you don't have to go to UCL to pull a
15:17
stunt like that. AI cloning software
15:19
is, as Kimberly mentioned, easy to
15:21
find on the web. To criminal
15:23
minds, the tech opens up a
15:25
new world of possibilities. A Pandora's
15:27
box. So
15:31
I'm just browsing here, looking at, you
15:34
know, a number of online banking websites.
15:36
And there are plenty now offering voice
15:38
ID as a way of accessing your
15:40
bank accounts. Some really interesting
15:42
things here. They describe voice ID
15:45
as like a fingerprint unique to
15:47
you and that their software even
15:49
recognizes you for having a bad
15:51
day, as in a cold or
15:53
a sore throat. But some say
15:55
the unique characteristics of your voice
15:57
make it much more secure than
15:59
a conventional power. and that voice
16:01
recognition also saves you time. Peter
16:04
Baron, though, isn't convinced. My
16:07
bank has recently introduced
16:09
voice recognition, and
16:12
I half-jokingly said
16:14
this to my guy, and
16:17
he said, be very careful
16:20
about anything which
16:22
involves voice recognition,
16:24
including your bank. Everybody
16:26
is going to have to reach their own decision on this.
16:29
For me, I personally won't
16:32
activate that feature. It's
16:34
another layer of security which
16:37
would theoretically keep your
16:39
account secure. However, we
16:42
know voice cloning is incredibly
16:44
easy, and
16:47
it's yet another example of how fraudsters have... how
16:50
they're not only a couple of steps ahead of
16:52
law enforcement, but when
16:54
something new comes out, they look at it and
16:57
find a way around it. The banks have acted
17:00
in good faith and with the
17:02
best intentions to try and
17:04
protect people's accounts. Voice
17:07
cloning effectively gives them a way
17:09
around it, so that could
17:11
potentially open you up to having
17:13
your account broken into or
17:16
taken over. Do you think that
17:18
layer of security then has now by AI
17:21
cloning software been compromised?
17:25
We find ourselves in something of a race where
17:27
banks are doing the best they possibly can with
17:29
the best of intentions, but
17:31
fraudsters are still by and large finding
17:34
the workarounds very easily. UK
17:37
finance, which represents the banking industry in the
17:39
UK, told us the sector
17:41
is continuously monitoring the use of generative
17:44
AI in scams. The
17:46
issue of whether someone could use a
17:48
fake AI voice to access telephone banking,
17:51
it told us that banks assess the risks
17:53
depending on the kind of transaction being made
17:56
and that they're using more layers of
17:58
controls and boosting customer authentication. processes.
18:01
Banks are also using AI
18:03
themselves to better identify suspicious
18:05
activity. But
18:09
voice zoning isn't the only threat
18:11
posed by generative AI. And
18:13
whilst for the likes of you and me, the
18:16
associated risks are only starting to surface
18:18
to be talked about and discussed across
18:21
the news and in social media, those
18:23
at the coalface of this tech
18:26
soul, most of it coming. Those
18:28
advances in generative AI systems are happening
18:31
as we speak all of the time.
18:34
Professor Lewis Griffin again. So
18:36
images, I think much of
18:38
the time the images generated
18:40
are completely undetectable. And
18:43
I think that's just going to increase very
18:45
rapidly. And they can be absolutely perfect.
18:49
Video is kind of further
18:51
behind that, but it's moving
18:54
along very fast. If
18:56
you start to make more complex
18:58
imagery, then there will start to
19:00
be small problems in it that
19:03
you can spot. But overall, the
19:05
impression is highly convincing. So how
19:07
soon do you think will
19:10
video and images be where
19:12
audio is now? How
19:14
long until it's perfect? I
19:16
think in months and years, not
19:18
decades. And
19:21
it was some video that helped to mislead
19:23
the next person I'm going to talk to.
19:26
It was generated by AI
19:28
and demonstrates the pernicious potential
19:31
of deepfakes. It's
19:36
the first time I had
19:39
used an online dating app in my
19:41
life. That is Anna. Not
19:44
her real name. She's asked us not to
19:46
share that. Today, she
19:48
invited us to her home in southern England.
19:51
She's in her fifties, has dark
19:53
shoulder length hair that's brightened by
19:55
a smattering of silvery grey and
19:58
kind eyes despite the old deal
20:00
that she's been through. How
20:03
are you feeling about deep diving into it
20:05
all again? I've
20:07
been deep diving all weekend, all weekend,
20:10
so I'm actually back in that groove.
20:13
Anna, you see, was the victim
20:15
of a years-long romance scam. She
20:17
was duped into falling in love and
20:20
parting with vast sums of money by
20:22
an unscrupulous scammer who claimed to be
20:24
someone else entirely. She knew
20:27
him as Simon, but
20:29
it's how he convinced her
20:31
it was all real that makes her
20:33
story unlike Annie you'll have
20:36
heard before. I'd been through a very,
20:38
very, very brutal divorce
20:40
abroad and for four years
20:42
I wasn't ready to even have a
20:44
cup of coffee with a man, not interested. So
20:47
the weeks leading up to it, it was actually the
20:49
months leading up to it, my friend said, I think
20:51
you're ready, you're bouncing again, you're smiling again. The
20:54
first hit was
20:58
Simon's, the very first
21:01
text. The face
21:03
of an angel and a doneness
21:06
of a man. And
21:08
then the text turned
21:10
very loving. Anna gave us
21:13
access to voice messages that she and
21:15
Simon shared over the course of their
21:17
relationship. Here's just one example and this,
21:19
Anna believes, is the fraudster's
21:22
real voice. I'm going to make you
21:24
happy, baby. You're such
21:26
a wonderful woman, a
21:28
woman with great hat. You are the
21:30
woman I want all my life. Soon
21:32
they made plans to meet up in
21:34
person at a restaurant in West London
21:37
near where he claimed to live. It was
21:39
three weeks into our two and a half
21:41
year relationship. He booked a restaurant that I'd
21:43
looked up existed. He booked a table and
21:46
he had a business crisis and had to go to
21:48
Paris. Simon
21:50
told her he was a wealthy businessman, was
21:52
of Bulgarian Scottish heritage and
21:54
dealt in edible oils. Now
21:56
his complex web of lies and of
21:59
drawn out extortion. began with
22:01
a tall tale. He told
22:03
Anna that a sizeable shipment of his products had
22:06
been held blocked from clearing customs
22:08
in Paris, a simple case of
22:10
the wrong paperwork he said. He needed
22:13
cash quickly to get the whole thing
22:15
moving again and asked Anna if on
22:17
this one occasion she could help
22:19
out. She agreed. The
22:21
first one was £1,500 to
22:25
clear the port charges so that the containers could
22:27
leave. You believe all of that to that point.
22:30
He sent receipts. He sent loads
22:32
of receipts. Requests for money
22:35
continued, came thick and fast.
22:38
Anna would sometimes make him sign
22:40
contracts promising her money would be
22:42
repaid. He did so without hesitation
22:44
each time and soon Anna found
22:46
herself more than £60,000 down. But
22:50
Simon's influences, powers of manipulation
22:52
of persuasion, continued to cloud
22:54
her better judgment to draw
22:57
her in ever deeper. Finally
23:00
a face-to-face meeting is
23:02
arranged, not in the flesh
23:04
as such but a video call on
23:06
Skype. At the time she was
23:08
again led to believe that Simon was in Paris.
23:11
The Skype video absolutely
23:14
looks like a French cafe. I know
23:16
I've been to Paris many times. He
23:18
comes online and he's moving and he's talking
23:21
and moving and he said I can
23:23
hear you I can hear you I can hear you I can see you can
23:26
you see me and he's moving around and talking like
23:28
can you see me I'm like yeah I
23:30
said but I can't hear your voice so
23:32
you can see him lip-syncing that did it for me
23:34
that did it for me to
23:38
me it was him saying my name and I love you soon
23:40
after that a second video call
23:43
a different setting this time
23:45
but the same technical difficulties
23:47
picture no sound he's
23:50
sitting on a bed and he told me I'll call you for my bed
23:52
and breakfast it was a bit dark
23:54
it was him he could hear me and see me I
23:58
couldn't hear him but I could see him so we I
24:00
had two occasions and I was like, hey, you're real.
24:03
I was floating, I was floating on
24:05
clouds after having seen them. I
24:07
was floating on clouds
24:10
for weeks. And you think AI
24:12
was involved in both occasions, do you? It
24:20
later transpired that those real-time video
24:22
calls were fake. Or to
24:24
be more precise, deep fakes. The
24:28
typical Parisian background was
24:30
probably AI-generated, superimposed, as
24:33
was the face and body of the man Anna
24:35
thought was Simon, the supposed love
24:37
of her life. But
24:43
her story is far from over. The
24:45
deceit would deepen, as would the
24:48
gaping financial hole she found
24:50
herself in when it all came to
24:52
a devastating, if bizarre, end. How
25:00
long ago was it that you heard about
25:02
the first case? In my
25:05
experience, probably under a year, I would say.
25:07
It's very, very new to the team. Lisa
25:10
Mills is a senior fraud expert
25:12
at the charity Victims Support. She
25:16
specialises in romance fraud and has dealt with cases that
25:19
echo Anna's experience. When
25:21
I have seen victims who have not destroyed
25:23
exchanges between
25:26
the fraudster, they will show me
25:28
images that they will see have been
25:30
doctored. And
25:32
they will also tell me about real-time video exchanges that
25:36
they've had. With a suspect, with the growth of
25:39
AI, it
25:41
will lead to fraudsters having
25:43
an ever-increasing capacity to
25:47
convince their victims that they are who they say they are,
25:50
to place them in the situations
25:52
that they describe. In 2022, UK
25:54
victims of romance scams lost close
25:56
to a ho... That's
26:02
according to the National Fraud Intelligence
26:04
Bureau and those in
26:06
the know fear that already
26:08
alarming figure could increase even
26:10
further. Now the fraudsters are
26:12
monetising and in a sense weaponising
26:15
AI's darker capabilities. I
26:18
love you baby. I
26:22
love you from the bottom of my head. I just
26:24
woke up now and what I got in my name
26:26
is just you. I'm
26:29
just a little bit of a light in love you. Back
26:33
with Anna in the south of England, we
26:35
talk through the closing chapters of her story.
26:38
A grim collection of carefully
26:40
concocted lies which culminated in
26:42
Simon claiming he'd been abducted
26:45
and would be killed unless Anna could help
26:47
him pay his way out of it.
26:50
Come on, come on, come on please,
26:52
don't make me cry, don't make me
26:54
sad, you are my last love, you
26:56
are my only love, please come to
26:58
my rescue and assist me on this
27:01
last date. Please my love, please,
27:03
I love you so much. Panicked
27:05
and convinced that Simon could at any
27:08
moment be murdered, Anna posts a message
27:10
and a picture of him
27:12
on social media. She asks to
27:15
be contacted if anybody has any
27:17
information whatsoever on his whereabouts
27:20
and she gets a reply from
27:22
a woman that changes the course
27:24
of her life forever. She goes
27:26
I'm Mexican. I said okay. She
27:30
went the picture you posted is
27:32
probably our most famous soap star.
27:35
His name is Juan Solí. After
27:38
two and a half hours I still didn't believe
27:40
her and she could hear it but eventually she
27:42
does believe it. Upon
27:44
much closer inspection she
27:47
realises that every image she's ever been
27:49
sent by Simon and the man staring
27:51
back at her in those
27:53
video calls is in fact
27:55
Juan Solí, an
27:57
Argentine Mexican actor. This
28:00
is what he actually sounds like. To be
28:02
clear, the real one, Soleil, had no part
28:04
in the scam. Nor
28:14
did he have any idea that his
28:16
image was being used as part of
28:18
the fraud. How
28:21
much did you give him in total? With
28:27
the help of the financial ombudsman, Anna did
28:30
claw back more than two thirds of the
28:32
money she lost. Clearly
28:34
the victim of a
28:36
drawn-out and devastating scam. How
28:39
many of you have seen the video? AI
28:48
voice cloning, deep-sake videos,
28:51
both have paved the way
28:53
for dark twists, malicious evolutions
28:55
of tried and tested scams.
28:58
But in perhaps the darkest corners
29:00
of the internet, these new frontier
29:02
technologies are being used
29:04
for something even more disturbing. Something
29:08
Professor Lewis Griffin and the group
29:10
of experts assembled years ago didn't
29:13
foresee. There's an aspect
29:15
of deep fakery, a use of
29:18
deep fakery, that we
29:20
completely missed. In
29:22
fact, the most advanced of these
29:24
threats, and that is the generation
29:27
of synthetic child sexual
29:29
imagery. The people who create
29:31
these images are the early
29:34
adopters, and they are developing
29:37
it in greater volume, a
29:39
more technical finesse than in
29:41
any other area of criminal
29:43
use. And
29:46
it has enormous potential
29:48
for harm. The
29:57
Internet Watch Foundation, a charity based in
29:59
Canada, Cambridge are known globally for the
30:02
work they do in trying to cleanse
30:04
the internet of images of child sexual
30:06
abuse. Recently, they
30:08
published a timely, eye-opening
30:11
report on the role
30:13
AI now plays in the creation and
30:15
manipulation of such images. So
30:17
the major findings were after spending a month
30:21
reviewing a forum dedicated to
30:23
the sharing of AI-generated
30:27
images of child sexual abuse,
30:30
we were able to assess more than 11,000
30:34
images and confirm that nearly
30:36
3,000 of those were
30:38
images which would breach UK law.
30:41
Dan Sexton is their chief technology officer.
30:44
So this material was primarily located
30:46
in the dark web. Those
30:49
communities, this is technically minded at people
30:51
within those dark web communities
30:54
that have a potential sexual
30:56
interest in children and an interest in
30:58
sharing content of children, are
31:00
themselves experimenting with this technology. Dan
31:03
tells me those experiments tend to
31:05
produce two kinds of synthetic image.
31:08
That's an image, by the way, that has
31:10
been fully or partially created using AI rather
31:12
than being captured on a conventional
31:14
camera. Now, the first type
31:16
of image is made from
31:18
scratch. An image of a child that
31:20
doesn't exist, completely manufactured by AI. The
31:24
second is the manipulation of
31:26
existing images of children,
31:29
those already in circulation on the
31:31
web, to create fake images in
31:33
which those children are placed in
31:35
sexual positions and situations. Both
31:38
are illegal. As
31:40
part of the report, the IWF
31:42
also expressed their grave concern at
31:45
how easy the software is to
31:47
find and to use. sexual
32:00
images of other children. The
32:02
National Society for the Prevention of Cruelty
32:05
to Children has publicly echoed
32:07
those concerns and has shared
32:09
with Filon 4 transcripts of
32:11
phone calls made by teenagers
32:13
to Childline, their well-known in-house
32:16
counselling service. In
32:18
the calls, the emerging dangers of AI can
32:20
be heard if not felt.
32:23
The words you're about to hear have
32:26
been changed slightly to protect the identity
32:28
of a 15-year-old girl who became a
32:30
victim. Her words are spoken by
32:33
an actor. A stranger
32:35
online has made fake nudes of me. They
32:38
look so real. It's my
32:40
face and my room in the
32:42
background. They must have taken pictures
32:44
from my social media and edited
32:46
them. I'm so
32:49
scared that they'll send them to
32:51
my parents. The pictures are really
32:53
convincing and I don't think that they
32:55
believe me that they're fake. Not
32:58
only that, but the NSPCC has
33:00
received reports of children
33:02
being blackmailed. Richard
33:04
Collard is their Associate Head of Policy
33:07
and Public Affairs. It's often coming up
33:09
in the case of sextortions, so that's
33:11
where someone has called out Hotline to
33:14
say that they are being extorted for
33:16
sexual pictures of themselves. We're now starting
33:18
to see that happen through generative AI.
33:21
They're convinced that if they show or
33:23
tell anyone that they'll believe it's a
33:25
real image of them and they'll get
33:27
into trouble, it's embarrassing, so it's incredibly
33:30
distressing for the children. Dan
33:34
Sexton at the Internet Watch Foundation
33:37
believes social media platforms need to
33:39
do more to safeguard children and
33:42
that new legislation could pave the way.
33:45
The online safety bill is going to
33:47
give the UK government the ability to
33:49
address these kinds of harms. like
34:00
terrorism, revenge porn and other content
34:02
that could be harmful for children.
34:05
If they don't, they could be slapped with fines of
34:07
up to £18 million or 10% of global annual turnover,
34:13
whichever is higher. But
34:15
what about this still simmering, soon to
34:17
boil over, threat from
34:19
generative AI? How
34:21
can it be tackled head on? Professor
34:24
Griffin again. We've already got
34:26
examples which completely fool people and they're
34:28
going to be increasing in common and
34:30
it's going to be very easy to
34:32
fool people in
34:35
the near future. To make
34:37
them undetectable to algorithms is
34:40
much harder. You know, the algorithms,
34:43
they're looking at weird statistical
34:45
details down in the pixels to which
34:48
we would never pick up on and
34:50
they're saying that's not quite right. There's
34:52
an arms race already running where,
34:55
you know, someone makes a tool
34:57
that generates synthetic images but there
34:59
was an industry of people producing
35:01
tools to detect that. Social
35:04
media platforms that responded to our request
35:06
for comment told us they
35:08
have systems in place to report, detect
35:11
and remove illegal content and
35:13
that this will be reported to the authorities. This
35:16
includes images of content generated
35:18
by artificial intelligence. The
35:21
government told us it's investing 100
35:23
million to create a national fraud squad
35:25
with 400 new officers who will pursue
35:28
cyber criminals and other scammers wherever they
35:30
are in the world. Finally,
35:35
you're probably wondering how my poor mum
35:37
Mary is doing after I tried to
35:40
scam her out of a couple of
35:42
hundred quid. Well, she's grand, she's fine.
35:45
As soon as it became clear that the cloned
35:47
version of my voice had done its job, I
35:49
jumped in to tell her the truth. But,
35:52
well, we managed to, you know, make light
35:54
of it. What played out
35:56
that day is a dark and
35:58
troubling sign. What's on the
36:01
near horizon? OK, Mum, Mum,
36:04
Mum, hang on, hang
36:06
on a second. So,
36:08
and just please remember that
36:10
you love me, all right? OK, of
36:13
course I love you. Well, thank God for that.
36:16
So, you know the
36:18
way I'm making a documentary for
36:20
BBC Radio 4, File on 4, OK? So
36:24
we're making a documentary about artificial
36:26
intelligence and artificial intelligence
36:28
driven crime. So what
36:31
you just heard, Mum, wasn't
36:33
me. That was voice
36:36
cloning software. So I
36:38
am fine. Oh, I've
36:40
had it last of many hours and you want to be sent to
36:42
someone. That wasn't me, Mum. That
36:44
was the AI software. You thought
36:46
it was me. You still thought it was me. Yeah,
36:49
of course. Yeah, I did actually. How
36:52
would you feel if I'd been on the radio, Mum? I
36:55
don't care. My points wouldn't be very good. I
37:01
knew you'd laugh. I knew you'd laugh. Good.
37:03
Are you OK? That's good.
37:05
That's good. All right. OK. Thank you.
37:07
Love you too. Talk to you later. Bye.
37:09
OK, see you. Bye. See
37:12
you. Bye, Mum. OK, so... You're
37:15
going to pull it, Mum. Yeah, I
37:18
feel like... You're very sensitive
37:20
somehow. I will. I will. She didn't
37:22
laugh, though. She laughed. She's OK. She's
37:24
OK. And
37:49
all that. a
38:00
huge rift in Carolyn's family. That's our
38:02
mom. We're not going to let you
38:05
just do that. I'm Sue Mitchell, and
38:07
this story unfolded in California, on the
38:09
street where I live. Look what you
38:12
brought into your house! He's a con
38:14
artist, mother. Is Dave a
38:16
dangerous interloper, or the tender carer
38:18
he claims to be? That's why I'm
38:21
here. Thank the Lord. Find out
38:23
in Intrigue, Million Dollar Lover, from
38:25
BBC Radio 4. Listen
38:28
on BBC Sounds. If anything
38:30
happens to him, I
38:32
will just die. Have
38:39
you ever wondered what sets exceptional
38:41
leaders apart? Discover
38:43
how renowned leaders from around
38:45
the globe have harnessed their
38:47
natural talents to achieve remarkable
38:49
success. Uncover the
38:51
secrets of leadership excellence, one
38:54
strength at a time, through Gallup's
38:56
Leading with Strength. Dive
38:59
into compelling stories at
39:01
leadingwithstrengths.com
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More