"Building Protective Tech- Dealing with Leaked Intimate Imagery" with Madelaine Thomas

WEBVTT
00:00:04.719 --> 00:00:07.459
You're listening to a stage talk titled Building
00:00:07.459 --> 00:00:10.500
Protective Tech Dealing with Leaked Intimate
00:00:10.500 --> 00:00:13.080
Imagery. This week, we spoke with the founder
00:00:13.080 --> 00:00:16.160
of Image Angel, a company developing forensic
00:00:16.160 --> 00:00:19.160
watermarking technology to help trace the source
00:00:19.160 --> 00:00:22.199
of online image abuse. Madelaine explains how
00:00:22.199 --> 00:00:24.760
the technology works, the challenges of encouraging
00:00:24.760 --> 00:00:27.920
adoption by platforms and how her own experience
00:00:27.920 --> 00:00:30.879
led her to found the company. We also discuss
00:00:30.879 --> 00:00:34.159
why the technology isn't open source and explore
00:00:34.159 --> 00:00:36.840
alternative approaches to addressing this systemic
00:00:36.840 --> 00:00:39.939
issue. You can find links to all of the resources
00:00:39.939 --> 00:00:42.219
mentioned in the talk in the podcast description.
00:00:42.460 --> 00:00:44.539
This talk was hosted by me, Charlotte Maher, on
00:00:44.539 --> 00:00:47.539
Thursday the 2nd of July, 2026, in the Bellingcat
00:00:47.539 --> 00:01:02.030
Discord server. Welcome back to the Stage
00:01:02.030 --> 00:01:05.230
Talks podcast. Today we are diving deep into
00:01:05.230 --> 00:01:08.069
the frontier of digital safety, journalism integrity
00:01:08.069 --> 00:01:10.189
and the engineering of online accountability.
00:01:10.989 --> 00:01:12.969
A short trigger warning that in this discussion
00:01:12.969 --> 00:01:15.349
we may touch on themes of sexual exploitation
00:01:15.349 --> 00:01:18.709
and image -based sexual abuse. When sensitive
00:01:18.709 --> 00:01:21.510
media leaks, when war crime imagery is manipulated
00:01:21.510 --> 00:01:24.030
or when intimate images are shared without consent,
00:01:24.530 --> 00:01:28.060
the digital damage happens in seconds. For decades,
00:01:28.280 --> 00:01:31.159
our cultural and legal response has been entirely
00:01:31.159 --> 00:01:34.620
reactive. We try to flag, report and take down
00:01:34.620 --> 00:01:37.040
content after it's already spread across the
00:01:37.040 --> 00:01:39.840
internet, as many of you will know. But in an
00:01:39.840 --> 00:01:43.219
era of generative AI and automated disinformation,
00:01:43.700 --> 00:01:47.239
reactive safety is a failing strategy. Our guest
00:01:47.239 --> 00:01:49.500
today is tackling this problem by shifting the
00:01:49.500 --> 00:01:52.480
entire paradigm from reaction to provenance.
00:01:52.920 --> 00:01:55.980
Madelaine Thomas is the founder of Image Angel.
00:01:56.260 --> 00:01:58.859
Her company built advanced server -side forensic
00:01:58.859 --> 00:02:01.599
watermarking tools that embed invisible mathematical
00:02:01.599 --> 00:02:04.120
trackers into the structural frequency of images
00:02:04.120 --> 00:02:06.719
and videos the exact millisecond they are viewed.
00:02:07.260 --> 00:02:09.599
It means that no matter how much an asset is
00:02:09.599 --> 00:02:12.400
screenshotted, cropped or manipulated by deepfake
00:02:12.400 --> 00:02:14.759
algorithms, it carries a permanent invisible
00:02:14.759 --> 00:02:17.710
digital paper trail back to its source. Within
00:02:17.710 --> 00:02:20.069
this next hour, we'll be exploring the exact
00:02:20.069 --> 00:02:22.009
methodology of how this technology is built,
00:02:22.210 --> 00:02:24.250
the friction between open source transparency
00:02:24.250 --> 00:02:26.969
and defensive security, and how this math could
00:02:26.969 --> 00:02:29.289
reshape everything from creator safety to global
00:02:29.289 --> 00:02:32.310
information warfare. As usual, you can place
00:02:32.310 --> 00:02:34.550
your questions in the chat accessible on the
00:02:34.550 --> 00:02:36.650
right hand corner of your screen, but please
00:02:36.650 --> 00:02:39.050
remember that this is being audio recorded for
00:02:39.050 --> 00:02:41.270
the podcast, so if you don't want me to mention
00:02:41.270 --> 00:02:43.750
your username, please add that as a note in your
00:02:43.750 --> 00:02:46.259
question. It's also worth noting that Bellingcat
00:02:46.259 --> 00:02:49.740
does not endorse any individual company. And
00:02:49.740 --> 00:02:52.219
today we're talking about the technology rather
00:02:52.219 --> 00:02:54.759
than the company itself. Madelaine, your journey
00:02:54.759 --> 00:02:57.219
into building this technology is deeply personal.
00:02:57.419 --> 00:02:59.439
I want to hand the floor over to you. Take us
00:02:59.439 --> 00:03:01.900
back to the beginning, your story and how this
00:03:01.900 --> 00:03:05.620
technology works under the hood. Thank you for
00:03:05.620 --> 00:03:07.840
having me here. I'm really excited to go through
00:03:07.840 --> 00:03:11.020
this with you. I guess I need to just caveat
00:03:11.020 --> 00:03:13.180
what you've just said that we're changing the
00:03:13.180 --> 00:03:15.979
way the world's working. We're doing it slowly.
00:03:16.400 --> 00:03:20.000
And the person in charge of this company has
00:03:20.000 --> 00:03:23.400
a long and really successful career as a dominatrix.
00:03:23.599 --> 00:03:27.060
So tech is not my first language and business
00:03:27.060 --> 00:03:32.300
is not my first line of work. So everything I've
00:03:32.300 --> 00:03:35.900
achieved up to now has been hard fought. We've
00:03:35.900 --> 00:03:39.590
made mistakes and we've done things wrong. And
00:03:39.590 --> 00:03:41.349
I'm proud to say that we've got this far and
00:03:41.349 --> 00:03:43.870
I'm really excited to show it to you. So what
00:03:43.870 --> 00:03:46.849
I'm going to go through is from the very beginning,
00:03:46.870 --> 00:03:49.289
what happened to where we're at now. And then
00:03:49.289 --> 00:03:50.430
in the middle of that, we'll go through a bit
00:03:50.430 --> 00:03:53.069
of the techie side of things as well. So everyone
00:03:53.069 --> 00:03:55.110
can see my screen. I'm going to click through.
00:03:55.509 --> 00:03:58.150
So company is called Image Angel because I just
00:03:58.150 --> 00:04:01.050
needed a minute. I needed someone, I needed someone
00:04:01.050 --> 00:04:06.250
to help me. Because in 2017, my world completely
00:04:06.250 --> 00:04:10.009
shattered. I was doing amazing stuff in my line
00:04:10.009 --> 00:04:13.789
of work and winning awards and being on magazine
00:04:13.789 --> 00:04:17.629
covers and enjoying the successful career that
00:04:17.629 --> 00:04:20.509
I fought so hard to achieve. But under cover
00:04:20.509 --> 00:04:23.410
of that, I was battling with this secret shame
00:04:23.410 --> 00:04:26.910
that my content was just everywhere associated
00:04:26.910 --> 00:04:31.290
with my name, my place of where I live and where
00:04:31.290 --> 00:04:35.269
I grow up. And there was no way to stop it. There
00:04:35.269 --> 00:04:38.189
was no protection. There was no safety net and
00:04:38.189 --> 00:04:42.610
I got to the lowest point, the absolute worst
00:04:42.610 --> 00:04:44.990
moment of my life, really. I look back at it
00:04:44.990 --> 00:04:47.970
and I think it could not have got worse up to
00:04:47.970 --> 00:04:55.509
a point that I thought I am worth less. I don't
00:04:55.509 --> 00:05:00.449
need to be here anymore, almost. It was a sense
00:05:00.449 --> 00:05:05.769
of just exposure and helplessness and I had nowhere
00:05:05.769 --> 00:05:10.019
to turn. I did go to the revenge porn helpline
00:05:10.019 --> 00:05:14.000
and I asked for some help. But because my work
00:05:14.000 --> 00:05:18.259
was commercial, I was considered unable to be
00:05:18.259 --> 00:05:22.139
helped. And this is happening all the time to
00:05:22.139 --> 00:05:25.920
so many people and more and more. Unattributed,
00:05:26.339 --> 00:05:30.120
fraudulent, leak, fake. It causes irreversible
00:05:30.120 --> 00:05:33.519
damage. to an individual, but also just to truth
00:05:33.519 --> 00:05:36.600
and trust and your sense of self and belonging.
00:05:37.519 --> 00:05:40.000
And so I started looking at the world and I'm
00:05:40.000 --> 00:05:44.720
like, what can I possibly do as one woman who
00:05:44.720 --> 00:05:46.500
does not have a technical background? What can
00:05:46.500 --> 00:05:53.120
I do? So I wanted to focus on huge impact with
00:05:53.120 --> 00:05:56.480
nothing. So I started with no money, no experience,
00:05:56.639 --> 00:05:59.680
but I also wanted people to be protected without
00:06:01.290 --> 00:06:07.389
So I built a system that could potentially change
00:06:07.389 --> 00:06:10.430
the way harm is happening in these spaces in
00:06:10.430 --> 00:06:12.529
a way that could prevent. And I'm looking at
00:06:12.529 --> 00:06:15.509
all these things and I need to shape my company
00:06:15.509 --> 00:06:19.009
in a way that is going to survive the future.
00:06:19.189 --> 00:06:22.730
But like, what the hell is that? What does good
00:06:22.730 --> 00:06:24.949
cutting edge tech look like? It looks like all
00:06:24.949 --> 00:06:27.649
of these things. And I'm looking at the world
00:06:27.649 --> 00:06:30.959
and I'm thinking, Prevention is everywhere in
00:06:30.959 --> 00:06:34.600
that we prevent an image being displayed through
00:06:34.600 --> 00:06:38.399
hashing. And we tell people, don't express yourself
00:06:38.399 --> 00:06:40.759
online and don't go online and don't be part
00:06:40.759 --> 00:06:44.240
of this digital world because of the possibility
00:06:44.240 --> 00:06:48.920
for things that are bad to happen. So I thought
00:06:48.920 --> 00:06:54.259
I must be able to do something. And it took me
00:06:54.259 --> 00:06:57.000
a good few years to come up with the correct
00:06:57.000 --> 00:07:02.399
solution. Because as policy is evolving, as GDPR
00:07:02.399 --> 00:07:06.420
is evolving, as tech is evolving, things became
00:07:06.420 --> 00:07:11.300
possible and then impossible. And I actually
00:07:11.300 --> 00:07:13.939
found a solution, which was forensic watermarking.
00:07:13.939 --> 00:07:17.959
And I genuinely had the thought of, oh, I can
00:07:17.959 --> 00:07:22.839
do that. Like it's hard. So I'm really pleased
00:07:22.839 --> 00:07:25.459
to be able to say that I have developed Image Angel.
00:07:25.930 --> 00:07:30.750
So it is not perfect, but it's a start. It adds
00:07:30.750 --> 00:07:35.370
invisible robust data in a private and cheap
00:07:35.370 --> 00:07:41.870
way into media. So live broadcast, imagery, and
00:07:41.870 --> 00:07:46.769
also video. And it makes the risk of leaking
00:07:46.769 --> 00:07:51.759
something outweigh the reward. And throughout
00:07:51.759 --> 00:07:54.240
this presentation, I'm going to explain things
00:07:54.240 --> 00:08:00.060
to you guys in the sense of, as a consensual
00:08:00.060 --> 00:08:04.259
adult content creator, selling consensual content
00:08:04.259 --> 00:08:08.079
to a consensual buyer. That's my viewpoint. It's
00:08:08.079 --> 00:08:13.720
my origins, although this can be applied to so
00:08:13.720 --> 00:08:16.060
many different places. And I would love it if
00:08:16.060 --> 00:08:19.399
you could... Think of it from your perspective
00:08:19.399 --> 00:08:21.779
and see how it would work in your setting and
00:08:21.779 --> 00:08:23.519
what the barriers might be or what the changes
00:08:23.519 --> 00:08:29.259
may be. So Image Angel is more than a technical
00:08:29.259 --> 00:08:32.679
integration. It is a software as a service, but
00:08:32.679 --> 00:08:36.860
it's also a network where creators, regulators,
00:08:38.100 --> 00:08:41.759
abusers can see that this platform doesn't tolerate
00:08:41.759 --> 00:08:44.419
abuse. This place uses this tech, which means
00:08:44.419 --> 00:08:47.840
it doesn't tolerate anything. being used to abuse.
00:08:49.019 --> 00:08:50.759
We also have a network where people can seek
00:08:50.759 --> 00:08:53.820
help and then give platforms the data so that
00:08:53.820 --> 00:08:56.460
they can act upon that verified data. So I see
00:08:56.460 --> 00:09:00.480
it as not tech. It's a revolution. It's me using
00:09:00.480 --> 00:09:03.200
my harm for the good of the world. And I also
00:09:03.200 --> 00:09:06.500
see myself as a little bit like Liam Neeson.
00:09:06.720 --> 00:09:09.200
I will look you, I will find you, and there will
00:09:09.200 --> 00:09:11.620
be consequences. Because when we do something
00:09:11.620 --> 00:09:16.320
and there are no consequences, it feels powerless.
00:09:16.799 --> 00:09:18.860
So that's what I wanted to do is give the power
00:09:18.860 --> 00:09:21.159
back a little bit. But enough about the background.
00:09:21.179 --> 00:09:24.320
I'm going to go into how it works and I want
00:09:24.320 --> 00:09:26.460
to show you what it looks like and what our system
00:09:26.460 --> 00:09:30.559
does and how it's built. So we're going to use
00:09:30.559 --> 00:09:34.220
this beautiful little pup. This is a AI image
00:09:34.220 --> 00:09:38.779
of a pup. And this is the data that is embedded
00:09:38.779 --> 00:09:41.679
into the image. This is what the computer would
00:09:41.679 --> 00:09:45.789
read. You as an individual would see it. as it
00:09:45.789 --> 00:09:50.330
is when it turns over. You'll see it as it is
00:09:50.330 --> 00:09:56.409
without any data embedded in there. But it is
00:09:56.409 --> 00:10:01.889
essentially adding super pixels into the image,
00:10:02.429 --> 00:10:05.669
overlaying it on top, depending upon the system
00:10:05.669 --> 00:10:08.509
structure or injecting it into the image itself.
00:10:09.570 --> 00:10:14.370
Those pixels contain data. They rotate. They
00:10:14.370 --> 00:10:16.990
change. And we also have the ability to change
00:10:16.990 --> 00:10:20.789
the density and intensity. We also have the ability
00:10:20.789 --> 00:10:24.490
to adjust the amount of data that it holds. But
00:10:24.490 --> 00:10:29.889
at the moment, 64 bits is our limit. And what
00:10:29.889 --> 00:10:35.029
happens then? So you're joining a platform, you're
00:10:35.029 --> 00:10:38.629
loading media and somebody comes along and screenshots
00:10:38.629 --> 00:10:42.980
it, which is so common. And they may edit it.
00:10:43.159 --> 00:10:46.179
They may add something to the top of it. They
00:10:46.179 --> 00:10:48.299
may scribble all over it. They may print it out.
00:10:48.620 --> 00:10:51.899
They may invert the color or they may change
00:10:51.899 --> 00:10:57.399
it, prop it. You will eventually find that out.
00:10:57.779 --> 00:11:02.559
That's how this world works. We always find content
00:11:02.559 --> 00:11:05.919
out there somewhere in the world. And at that
00:11:05.919 --> 00:11:10.429
point, that's where I was. that, oh my God, what
00:11:10.429 --> 00:11:13.289
the hell am I going to do moment? And I felt
00:11:13.289 --> 00:11:16.950
so powerless. And when I was creating this product,
00:11:17.590 --> 00:11:20.750
I needed to be able to say, even if nothing was
00:11:20.750 --> 00:11:24.490
possible, something must be possible. Even if
00:11:24.490 --> 00:11:27.429
there is no legal recourse or no change that
00:11:27.429 --> 00:11:30.710
can be made, it's important at this point to
00:11:30.710 --> 00:11:33.649
be able to go somewhere and get the help that
00:11:33.649 --> 00:11:36.389
you need, even if that is just a couple of lines
00:11:36.389 --> 00:11:39.370
of advice. in one centralized place because you're
00:11:39.370 --> 00:11:43.730
panicking. So we created the ImageAngel report
00:11:43.730 --> 00:11:47.190
and support system. So as well as creating a
00:11:47.190 --> 00:11:50.669
system that injects data into imagery, we also
00:11:50.669 --> 00:11:52.649
created a system where you could get the help
00:11:52.649 --> 00:11:55.210
that you needed if you needed it. So at that
00:11:55.210 --> 00:11:56.730
point, you would find your image out there in
00:11:56.730 --> 00:12:00.210
the wild and you would report it to us. And there's
00:12:00.210 --> 00:12:04.360
lots of people that ask the question, But how
00:12:04.360 --> 00:12:07.940
do I find where that piece of information is?
00:12:09.480 --> 00:12:14.799
There are numerous ways of finding data and information
00:12:14.799 --> 00:12:17.299
out there in the world. What I didn't want to
00:12:17.299 --> 00:12:20.440
do was recreate a system that already exists.
00:12:20.940 --> 00:12:23.340
So I've created the two parts of a system that
00:12:23.340 --> 00:12:26.620
compliment the tech that already exists in finding
00:12:26.620 --> 00:12:31.100
and tracking content out there. We are developing
00:12:31.100 --> 00:12:34.100
a system. But I will go into that a bit later,
00:12:34.360 --> 00:12:38.320
but we'll search the internet for watermarked
00:12:38.320 --> 00:12:41.299
imagery. So then what happens? You submitted
00:12:41.299 --> 00:12:44.080
that piece of information to ImageAngel. You
00:12:44.080 --> 00:12:47.539
could report it to an NGO. You could report it
00:12:47.539 --> 00:12:50.019
to the platform itself. You could report it to
00:12:50.019 --> 00:12:52.620
policing or anywhere that works in conjunction
00:12:52.620 --> 00:12:55.399
with ImageAngel. That report will come to us.
00:12:55.700 --> 00:12:59.179
We then triage it. We take that report and we
00:12:59.179 --> 00:13:01.789
see how serious is this. Is this life -threatening?
00:13:01.830 --> 00:13:05.029
Have you reported to us that this is danger status,
00:13:05.070 --> 00:13:06.929
in which case it would go right to the top of
00:13:06.929 --> 00:13:09.250
the pile? Or is this just, I'm a bit annoyed,
00:13:09.690 --> 00:13:13.210
my commercial work has gone missing and I'm frustrated?
00:13:13.590 --> 00:13:15.429
So we like to find out what's going on, what's
00:13:15.429 --> 00:13:18.649
the situation, so we can really triage. Then
00:13:18.649 --> 00:13:21.190
we extract the data that's in there. So there's
00:13:21.190 --> 00:13:24.230
two watermarks that are in there and they both
00:13:24.230 --> 00:13:28.480
contain a small piece of information. One contains
00:13:28.480 --> 00:13:31.220
the platform data. So that was where you uploaded
00:13:31.220 --> 00:13:35.559
it as the user. And in this instance, let's say
00:13:35.559 --> 00:13:38.860
it's a place called Fan Vault. Then Fan Vault,
00:13:39.200 --> 00:13:41.320
we contact them and we give that other piece
00:13:41.320 --> 00:13:44.000
of information that is held within the watermark.
00:13:44.460 --> 00:13:46.960
And we again, don't know what this is. It's just
00:13:46.960 --> 00:13:51.720
a random hexadecimal numbers and letters. And
00:13:51.720 --> 00:13:53.960
we look at that and go, this belongs to this
00:13:53.960 --> 00:13:57.220
platform. you'll be able to decode that. So even
00:13:57.220 --> 00:13:59.940
if we wanted to, we wouldn't be able to tell
00:13:59.940 --> 00:14:03.539
anyone what that data pertains to. All we know
00:14:03.539 --> 00:14:08.220
is where it came from. And then we advise the
00:14:08.220 --> 00:14:11.820
platform of Origin what should happen next. And
00:14:11.820 --> 00:14:15.360
this information is gathered because the user
00:14:15.360 --> 00:14:18.820
has told us what they would prefer. I'm furious.
00:14:19.259 --> 00:14:23.860
I'm upset. I need therapy. I need help. I need
00:14:23.860 --> 00:14:27.500
takedown help. I need this person to be blocked
00:14:27.500 --> 00:14:30.019
from the system. I need the police involved.
00:14:30.940 --> 00:14:34.179
Or I need nothing. I don't want to do anything.
00:14:34.240 --> 00:14:36.259
I just want to let you know that this has happened.
00:14:37.279 --> 00:14:40.100
So the extracted data is only decodable by the
00:14:40.100 --> 00:14:43.440
source platform when a verified case is raised.
00:14:43.759 --> 00:14:46.299
No exposed personnel identifying information.
00:14:46.340 --> 00:14:49.000
It is just a technical identifier. So it's privacy
00:14:49.000 --> 00:14:52.000
preserving. GDP are compliant at every step.
00:14:52.860 --> 00:14:55.559
That is vital. And one thing that we're still
00:14:55.559 --> 00:14:58.179
working out and something that may come up in
00:14:58.179 --> 00:15:01.340
questions is, well, how do you verify what a
00:15:01.340 --> 00:15:04.100
genuine request is? And that's a really good
00:15:04.100 --> 00:15:05.919
question. And we're still working through that.
00:15:06.500 --> 00:15:09.100
A genuine request we would imagine comes straight
00:15:09.100 --> 00:15:12.039
from a platform or policing and we can verify
00:15:12.039 --> 00:15:14.429
those. But sometimes you might get a request
00:15:14.429 --> 00:15:16.370
that comes in from an individual and that needs
00:15:16.370 --> 00:15:19.549
to be treated with care. But how do we make sure
00:15:19.549 --> 00:15:22.590
that it isn't an ex -partner trying to find out
00:15:22.590 --> 00:15:25.970
if someone is broadcasting on a platform or working
00:15:25.970 --> 00:15:29.409
on, let's say, OnlyFans? So we have to still
00:15:29.409 --> 00:15:32.730
work on that and build that part out. And then
00:15:32.730 --> 00:15:36.029
what happens? What happens then? So you've found
00:15:36.029 --> 00:15:37.909
this piece of leaked information out there in
00:15:37.909 --> 00:15:40.149
the wild. You've reported it to us and we've
00:15:40.149 --> 00:15:41.950
identified the person. Well, what happens then?
00:15:42.029 --> 00:15:45.009
And at this point, there is so many different
00:15:45.009 --> 00:15:48.330
options available. It could be small claims.
00:15:48.450 --> 00:15:51.230
It could be that it is against the Sexual Offences
00:15:51.230 --> 00:15:54.409
Act. It could be a copyright infringement. Is
00:15:54.409 --> 00:15:58.169
the user in the US? Are they in India? How do
00:15:58.169 --> 00:16:01.340
the two jurisdictions work together? When you
00:16:01.340 --> 00:16:06.539
regenerate a piece of material entirely, you
00:16:06.539 --> 00:16:10.779
regenerate it with the pixels that we also embedded
00:16:10.779 --> 00:16:15.879
in there. So it is somewhat possible to be able
00:16:15.879 --> 00:16:23.200
to say the presence of those pixels could denote
00:16:23.200 --> 00:16:28.659
that that came from a private channel. And of
00:16:28.659 --> 00:16:31.940
course, It's not going to be possible to use
00:16:31.940 --> 00:16:34.919
this as an evidence of its in and of itself,
00:16:34.940 --> 00:16:39.019
but it could stand as supporting evidence if
00:16:39.019 --> 00:16:41.340
you have the rest of the data to back it up.
00:16:42.000 --> 00:16:45.399
And of course, we've tested it. We want to know
00:16:45.399 --> 00:16:47.080
what people think. We don't want to put something
00:16:47.080 --> 00:16:50.539
out there that people don't want. And the response
00:16:50.539 --> 00:16:54.059
has been that third party neutrality, lived experience
00:16:54.059 --> 00:16:56.740
is really important. And this is the kind of
00:16:56.740 --> 00:16:59.389
perpetrator accountability that creators are
00:16:59.389 --> 00:17:03.009
desperate for, but no single mechanism is enough.
00:17:03.129 --> 00:17:08.109
Because when a harm that is this complex is happening,
00:17:08.509 --> 00:17:11.390
you need complex responses. You need layered,
00:17:11.829 --> 00:17:15.329
holistic responses. So when used together with
00:17:15.329 --> 00:17:17.630
things like hash databases, it can provide a
00:17:17.630 --> 00:17:20.069
safety infrastructure that is both preventative
00:17:20.069 --> 00:17:22.910
and reactive. But if someone is determined to
00:17:22.910 --> 00:17:26.670
do harm, then there is nothing that will stop
00:17:26.670 --> 00:17:32.069
them. It's a regulatory ally, not a big full
00:17:32.069 --> 00:17:36.269
stop. And then of course, we've asked exactly
00:17:36.269 --> 00:17:38.650
the market that we want to sell to the content
00:17:38.650 --> 00:17:42.250
creators on their platforms. What do you want?
00:17:42.269 --> 00:17:45.210
Does it work for you? And then there's a really
00:17:45.210 --> 00:17:46.950
important part that I haven't really gone over,
00:17:46.970 --> 00:17:51.869
which is who pays for this? The last thing an
00:17:51.869 --> 00:17:55.349
individual needs to do is pay for their own safety.
00:17:55.950 --> 00:17:58.490
when someone else is profiting from their work.
00:17:59.829 --> 00:18:04.150
It is vital that creating safety infrastructure
00:18:04.150 --> 00:18:07.849
is not accessible only to those who know about
00:18:07.849 --> 00:18:11.630
it or who can afford it. And that is why in creating
00:18:11.630 --> 00:18:14.190
this tech, we embed directly into the platforms.
00:18:14.750 --> 00:18:19.069
This also means that the data that is held on
00:18:19.069 --> 00:18:21.289
an individual doesn't need to be touched by the
00:18:21.289 --> 00:18:24.890
creator. But if we want to have real -world impact,
00:18:25.079 --> 00:18:28.640
We need real -world data. So we embed a platform
00:18:28.640 --> 00:18:32.200
level. They have the consumer data. They embed
00:18:32.200 --> 00:18:35.539
it into their system. And then we give them the
00:18:35.539 --> 00:18:38.119
key that unlocks that information. So I hope
00:18:38.119 --> 00:18:40.000
there's loads of questions. I'm sure there are.
00:18:40.700 --> 00:18:45.900
And that's it from me. Thank you very much. Thank
00:18:45.900 --> 00:18:49.880
you, everybody, for your many questions in the
00:18:49.880 --> 00:18:51.880
chat, actually. And thank you, Madeline, for
00:18:51.880 --> 00:18:54.960
taking us through that. You touched on this very
00:18:54.960 --> 00:18:56.960
quickly. We had one question, which I thought
00:18:56.960 --> 00:18:59.000
we could just kick off with actually. Somebody
00:18:59.000 --> 00:19:04.339
said, is this a stop NCII .org direct alternative?
00:19:04.579 --> 00:19:07.500
I was wondering if you could maybe explain about
00:19:07.500 --> 00:19:10.319
how you could work, for example, with the revenge
00:19:10.319 --> 00:19:14.500
porn helpline or how does that work? How do you
00:19:14.500 --> 00:19:17.500
differ from those kinds of services? Yeah. So
00:19:17.500 --> 00:19:21.240
there's kind of three things. There's forensic
00:19:21.240 --> 00:19:24.240
watermarking, which this is. which adds user
00:19:24.240 --> 00:19:27.180
data in there, or it could add origin data, or
00:19:27.180 --> 00:19:29.099
it could add the price that someone paid for
00:19:29.099 --> 00:19:30.579
something. It could be used in many different
00:19:30.579 --> 00:19:36.259
ways. There is hash matching, and that's both
00:19:36.259 --> 00:19:39.920
exact hash matching and perceptual. Like, is
00:19:39.920 --> 00:19:42.460
it potentially the same image? And then there's
00:19:42.460 --> 00:19:44.539
fingerprinting. And fingerprinting is a way that
00:19:44.539 --> 00:19:47.279
you can then find that exact copy of that thing
00:19:47.279 --> 00:19:51.700
out there in the world. We work in conjunction
00:19:51.700 --> 00:19:55.420
with all of these wonderful things. We would
00:19:55.420 --> 00:19:59.259
love to work more with Stop NCII, but we're completely
00:19:59.259 --> 00:20:02.880
self -funded and it costs money to do that, even
00:20:02.880 --> 00:20:05.839
though they're a charity. And to kind of touch
00:20:05.839 --> 00:20:08.619
on the kind of commercial side of this, I've
00:20:08.619 --> 00:20:11.819
paid for this development and this tech completely
00:20:11.819 --> 00:20:14.759
from my own pocket because this is so important
00:20:14.759 --> 00:20:19.480
to me. charge the platforms because that's when
00:20:19.480 --> 00:20:21.900
the harm is happening. We don't have a penny
00:20:21.900 --> 00:20:25.799
to waste and we are super, super, super tight
00:20:25.799 --> 00:20:29.880
on what we spend our money on. So to work with
00:20:29.880 --> 00:20:33.759
a charity in partnership with them, we need to
00:20:33.759 --> 00:20:36.799
be able to fund ourselves to grow big, to be
00:20:36.799 --> 00:20:39.400
able to prevent this harm happening. And then
00:20:39.400 --> 00:20:42.299
if they need funding, then we can give it to
00:20:42.299 --> 00:20:44.579
them because we've earned enough and we've made
00:20:44.579 --> 00:20:47.180
sure our ecosystem is running. The last thing
00:20:47.180 --> 00:20:49.319
I want to do is make a profit -making machine.
00:20:50.000 --> 00:20:51.940
If any profit that we make is going to go straight
00:20:51.940 --> 00:20:54.720
back into the system, because this is what needs
00:20:54.720 --> 00:20:57.039
it. And when I was thinking of setting up a company,
00:20:57.099 --> 00:21:00.519
I was like, how? How do I do this? I don't have
00:21:00.519 --> 00:21:04.299
any money. I could set up a company that's a
00:21:04.299 --> 00:21:06.559
charity. I could set up a community interest
00:21:06.559 --> 00:21:13.039
group. And to do that, I needed to have so many
00:21:13.039 --> 00:21:16.150
kind of grants and things like that. And then
00:21:16.150 --> 00:21:20.950
there was also the model, do I sort of do this
00:21:20.950 --> 00:21:22.769
with fundraising and stuff like that? And I thought,
00:21:22.769 --> 00:21:25.349
I don't want to pull from the community. Like
00:21:25.349 --> 00:21:28.730
I need this protection now. I don't need to fundraise
00:21:28.730 --> 00:21:32.049
and get money in to be able to do this. So I
00:21:32.049 --> 00:21:34.009
need this now. So I'm just going to pay for it
00:21:34.009 --> 00:21:39.970
to happen. So we do work directly with Stop NCII,
00:21:39.990 --> 00:21:42.750
but we are not a partner because we can't afford
00:21:42.750 --> 00:21:45.359
to pay that fee. That's a long winded answer,
00:21:45.500 --> 00:21:50.680
hey. No, it's a good one. It's a good one. As
00:21:50.680 --> 00:21:52.859
you've just mentioned, obviously this is a commercial
00:21:52.859 --> 00:21:57.059
entity and we've also spoken about how Bellingcat
00:21:57.059 --> 00:21:59.539
is open source and how, to be honest, a lot of
00:21:59.539 --> 00:22:03.180
the tech that we do spotlight in this platform
00:22:03.180 --> 00:22:08.200
is open source oriented. There is some protective
00:22:08.200 --> 00:22:11.980
tech. that is open source source code for public
00:22:11.980 --> 00:22:14.920
audit and transparency. For example, Teller is
00:22:14.920 --> 00:22:17.480
a really cool tool that helps automatically encrypt
00:22:17.480 --> 00:22:20.259
a reporter's data on the front lines if they
00:22:20.259 --> 00:22:23.700
are filming within it. Or we have Bellingcat's
00:22:23.700 --> 00:22:27.259
auto -archiver tool, which uses one -way cryptographic
00:22:27.259 --> 00:22:30.079
hashing to archive an image, video, or post,
00:22:30.160 --> 00:22:32.779
even when it's taken offline. But ImageAngel
00:22:32.779 --> 00:22:35.940
is closed source, so why have you chosen... to
00:22:35.940 --> 00:22:38.599
be closed source and if your algorithm relies
00:22:38.599 --> 00:22:41.180
on the source code staying a secret to remain
00:22:41.180 --> 00:22:44.259
secure, isn't that a structural vulnerability?
00:22:44.640 --> 00:22:47.980
Like how do you address why it's closed source?
00:22:48.839 --> 00:22:51.099
It's closed source at the moment because we're
00:22:51.099 --> 00:22:53.900
using a specific type of tech that we have to
00:22:53.900 --> 00:22:55.980
pay for because we want to be the best. So we
00:22:55.980 --> 00:22:58.920
outsource, we buy that and we use it. What we
00:22:58.920 --> 00:23:02.559
need to do to remain robust against manipulations
00:23:02.559 --> 00:23:06.990
going forward is get rid of that dependency poly
00:23:06.990 --> 00:23:11.710
third party and build our own. But we also then
00:23:11.710 --> 00:23:14.210
have that dependency on us. And what if someone
00:23:14.210 --> 00:23:16.549
works out how to manipulate that? So what we
00:23:16.549 --> 00:23:20.769
are now doing is taking a mixture of open source,
00:23:21.150 --> 00:23:25.730
closed source, and combining each of those in
00:23:25.730 --> 00:23:28.730
various platforms so that it is harder to then
00:23:28.730 --> 00:23:31.569
decode in the long run. We're always fighting
00:23:31.569 --> 00:23:35.799
against the next thing to kind of break and and
00:23:35.799 --> 00:23:38.420
close the stuff, right? So it's always a constant
00:23:38.420 --> 00:23:40.619
fight. So what we want to do is just always change
00:23:40.619 --> 00:23:44.579
that model so it's harder to decode. But also
00:23:44.579 --> 00:23:49.380
what we do is depending on a platform's risk
00:23:49.380 --> 00:23:51.859
appetite, vulnerability to this kind of thing,
00:23:52.119 --> 00:23:54.160
we work with them directly to figure out the
00:23:54.160 --> 00:23:58.660
best model and method for adjusting the algorithm
00:23:58.660 --> 00:24:01.789
of the implementation of the watermark. Yeah,
00:24:01.809 --> 00:24:04.130
it is interesting, the technicality of everything.
00:24:04.289 --> 00:24:06.710
And I really want to dig into that, but we've
00:24:06.710 --> 00:24:08.789
had a few, we've still got quite a few questions
00:24:08.789 --> 00:24:11.410
to work through, so I want to quickly get them
00:24:11.410 --> 00:24:13.710
in. Someone's just asked how much of this is
00:24:13.710 --> 00:24:16.309
about risk awareness and accepting that content
00:24:16.309 --> 00:24:19.470
placed online may be misused versus preventing
00:24:19.470 --> 00:24:21.809
misuse or responding to it. For example, through
00:24:21.809 --> 00:24:24.970
techniques like that hex technique. In some cases
00:24:24.970 --> 00:24:28.569
is understanding and accepting the risks of sharing
00:24:28.569 --> 00:24:31.089
content, such as having an OnlyFans account,
00:24:31.450 --> 00:24:35.750
more important than focusing on response. Yeah,
00:24:35.930 --> 00:24:40.009
I mean, there is a risk to everything. And honestly,
00:24:41.089 --> 00:24:44.369
everyone says to me, oh, well, what do you expect?
00:24:45.190 --> 00:24:46.369
You know, having an OnlyFans account, what do
00:24:46.369 --> 00:24:48.309
you expect? Well, I expect a little bit of decency
00:24:48.309 --> 00:24:51.569
and privacy. I would never share the images that
00:24:51.569 --> 00:24:53.869
are sent to me from a user. There is that implicit
00:24:53.869 --> 00:24:57.529
expectation of being in a private chat that this
00:24:57.529 --> 00:25:00.430
communication doesn't go elsewhere. And our watermark
00:25:00.430 --> 00:25:03.309
will only ever apply to content that is private.
00:25:03.789 --> 00:25:06.470
There's no point watermarking stuff that's out
00:25:06.470 --> 00:25:09.890
there in the world and free to view. That's not
00:25:09.890 --> 00:25:12.930
what we do. We're not necessarily here to enforce
00:25:12.930 --> 00:25:15.549
copyright. If someone wishes to do that, we can
00:25:15.549 --> 00:25:20.220
support them. The main aim of the company is
00:25:20.220 --> 00:25:23.039
to prevent harm, and that is with a deterrence
00:25:23.039 --> 00:25:27.640
method. Adding in data of consumers into the
00:25:27.640 --> 00:25:30.660
content they watch could be seen as a surveillance
00:25:30.660 --> 00:25:33.900
sort of, well, that means every piece of content
00:25:33.900 --> 00:25:36.640
contains my information and what if this and
00:25:36.640 --> 00:25:39.240
what if that? And the simple answer is if you
00:25:39.240 --> 00:25:43.339
do no wrong, nothing happens. It is not illegal
00:25:43.339 --> 00:25:46.640
to share your own content. You can do what you
00:25:46.640 --> 00:25:49.339
want with that image once you've taken a screenshot
00:25:49.339 --> 00:25:51.460
of it or a screen recording. You can do what
00:25:51.460 --> 00:25:53.980
you want, but you shouldn't be doing that. And
00:25:53.980 --> 00:25:56.920
if you do do a negative thing, you could potentially
00:25:56.920 --> 00:26:00.859
be caught and understand what the repercussions
00:26:00.859 --> 00:26:03.500
of that could be anything from just being banned
00:26:03.500 --> 00:26:05.980
or kicked off that platform, for example. So
00:26:05.980 --> 00:26:09.359
for example, if you were a content creator who
00:26:09.359 --> 00:26:13.359
had a private paid for space, like for example,
00:26:13.400 --> 00:26:16.440
people had to pay to gain content and that content
00:26:16.440 --> 00:26:19.339
was leaked, would your company help in that sense?
00:26:20.339 --> 00:26:23.759
Yeah, definitely. If a consumer was talking with
00:26:23.759 --> 00:26:26.099
a content creator and a content creator threatened
00:26:26.099 --> 00:26:28.660
to leak or something, their private material
00:26:28.660 --> 00:26:32.619
in the same regard, that's a private space. Interesting.
00:26:33.000 --> 00:26:37.299
We've had quite a few questions focusing on whether
00:26:37.299 --> 00:26:40.460
you've had any success with the big tech platforms
00:26:40.460 --> 00:26:44.460
adopting your tech or showing any interest in
00:26:44.460 --> 00:26:49.099
it. I wonder if you can speak about that. So
00:26:49.099 --> 00:26:53.539
we've had, because of my experience in the adult
00:26:53.539 --> 00:26:56.420
space, I used my connections and was like, look,
00:26:56.420 --> 00:26:58.759
I've been working on your platform for XYZ time.
00:26:58.799 --> 00:27:02.400
Do you think you could use this tech? And mostly
00:27:02.400 --> 00:27:08.180
they said no. But there was one platform called
00:27:08.180 --> 00:27:11.119
Friends To Follow who were like, we love this,
00:27:11.140 --> 00:27:15.519
we want this, make it happen. They've been really,
00:27:15.519 --> 00:27:20.220
really great in paying. They paid us, they gave
00:27:20.220 --> 00:27:23.440
us money to put this system in their platform
00:27:23.440 --> 00:27:25.119
and it's live and it's active and it's protecting
00:27:25.119 --> 00:27:29.319
half a million users. And that's the kind of
00:27:29.319 --> 00:27:35.369
good tech I see. I've approached, you could pretty
00:27:35.369 --> 00:27:38.170
much name every tech platform. We have in some
00:27:38.170 --> 00:27:42.130
way approached them. I can't say who has and
00:27:42.130 --> 00:27:43.829
who hasn't decided to go ahead with it because
00:27:43.829 --> 00:27:50.210
there are some people who are under NDA. The
00:27:50.210 --> 00:27:53.349
big, big ones. But I guess I can say, can I say
00:27:53.349 --> 00:27:55.210
I can say if someone has decided not to go ahead
00:27:55.210 --> 00:27:57.190
with something. Yeah. Yeah. Meta said that they
00:27:57.190 --> 00:27:59.569
didn't have a problem with stop NCII. So I was
00:27:59.569 --> 00:28:05.900
like, okay. If you think that, cool. I think
00:28:05.900 --> 00:28:09.400
there is a certain amount of unwillingness to
00:28:09.400 --> 00:28:13.539
accept that this is a good solution, partly because
00:28:13.539 --> 00:28:17.400
of my background. In fact, I had somebody who
00:28:17.400 --> 00:28:19.759
pitched the deck to, and they said, yeah, you've
00:28:19.759 --> 00:28:22.440
not done bad for a whore. And I thought, yeah,
00:28:22.819 --> 00:28:26.299
I've not done bad for a whore. Fuck you. But
00:28:26.299 --> 00:28:28.240
there are some people who don't like what I do
00:28:28.240 --> 00:28:32.089
because of where I've come from, I suppose. I'm
00:28:32.089 --> 00:28:34.329
not a native in the tech land, I don't speak
00:28:34.329 --> 00:28:38.750
in the same way. So there has been some people
00:28:38.750 --> 00:28:41.230
who don't want to be anything to do with this.
00:28:41.630 --> 00:28:43.470
And I've actually experienced quite a lot of
00:28:43.470 --> 00:28:47.450
negativity from people who are really, really
00:28:47.450 --> 00:28:52.990
big voices in violence against women and girls
00:28:52.990 --> 00:28:57.529
spaces. Even though I felt that harm, you know?
00:28:57.809 --> 00:29:00.920
And I created this to be able to prevent... content
00:29:00.920 --> 00:29:03.900
from getting out there and leaking and that kind
00:29:03.900 --> 00:29:07.019
of harm that comes through. So the response has
00:29:07.019 --> 00:29:11.059
been really mixed. Some places are really keen,
00:29:11.480 --> 00:29:13.819
others don't even want to talk about it. And
00:29:13.819 --> 00:29:17.900
that's probably because trust and safety is very
00:29:17.900 --> 00:29:21.480
different to compliance. Compliance is, oh, I
00:29:21.480 --> 00:29:25.079
made my system this safe. Look, we made it. Whereas
00:29:25.079 --> 00:29:27.299
trust and safety is anticipating harm, adding
00:29:27.299 --> 00:29:31.170
in prevention instead of reaction. I hope that
00:29:31.170 --> 00:29:35.230
that will change. There's some quite sarcastic
00:29:35.230 --> 00:29:40.250
shock in the chat about Metta saying no. I am
00:29:40.250 --> 00:29:42.690
so sorry if somebody said that to you as well.
00:29:42.910 --> 00:29:46.869
That's disgusting. You mentioned the tech platforms.
00:29:47.009 --> 00:29:49.829
What about governments? For example, why is forensic
00:29:49.829 --> 00:29:52.549
watermarking not part of the UK government strategy?
00:29:53.250 --> 00:29:55.250
They recently declared intimate image abuse a
00:29:55.250 --> 00:29:57.450
national emergency, yet regulations like the
00:29:57.450 --> 00:30:00.299
Online Safety Act focus almost entirely on reactive
00:30:00.299 --> 00:30:02.539
measures, giving platforms 48 hours take down
00:30:02.539 --> 00:30:04.900
content after it's reported, for example. Why
00:30:04.900 --> 00:30:06.900
do you think the government and Ofcom in the
00:30:06.900 --> 00:30:10.339
UK and more globally as well are hesitant to
00:30:10.339 --> 00:30:13.240
mandate preventative server -side forensic tracking?
00:30:14.759 --> 00:30:21.240
I wish I knew. I wish I knew. I have been, I'm
00:30:21.240 --> 00:30:25.599
not going to lie, bugging Ofcom particularly,
00:30:25.680 --> 00:30:28.849
but also the EU commission. about this tech.
00:30:30.089 --> 00:30:32.529
I think they don't understand it. If I read,
00:30:32.930 --> 00:30:36.269
and if you search through Ofcom's guidance and
00:30:36.269 --> 00:30:39.910
you search the word watermarking, in one of them,
00:30:40.650 --> 00:30:42.390
probably the most, not the most recent, I haven't
00:30:42.390 --> 00:30:45.190
checked that one before that, you search the
00:30:45.190 --> 00:30:47.690
word watermarking, it comes up with one thing
00:30:47.690 --> 00:30:50.569
referencing, you know, like the old Getty Images
00:30:50.569 --> 00:30:53.690
watermark, like Getty Images, proof of origin.
00:30:54.549 --> 00:30:58.210
I don't think they get it. But I have been thankfully
00:30:58.210 --> 00:31:03.309
invited to teach Ofcom about the tech itself.
00:31:03.609 --> 00:31:05.089
Now, of course they have to be really impartial
00:31:05.089 --> 00:31:09.250
and I respect that. I'm not interested in kind
00:31:09.250 --> 00:31:13.529
of dominating the market and making loads of
00:31:13.529 --> 00:31:16.829
kind of moves and money. I want people to be
00:31:16.829 --> 00:31:19.289
safe. I don't care if there's another company
00:31:19.289 --> 00:31:22.609
out there that can do this. Go ahead and do it.
00:31:23.690 --> 00:31:28.009
I need Ofcom to back me, but I don't feel like
00:31:28.009 --> 00:31:31.849
I have that. And I don't know why. There you
00:31:31.849 --> 00:31:33.890
go. There's your call to action guys. If you
00:31:33.890 --> 00:31:36.190
want to make an open source product that takes
00:31:36.190 --> 00:31:45.269
the image, go wild. There is someone else who
00:31:45.269 --> 00:31:46.529
does this. In fact, there's a couple of companies
00:31:46.529 --> 00:31:50.029
that do this and I've raised with them that I
00:31:50.029 --> 00:31:56.079
would love to collaborate. I see a world in which
00:31:56.079 --> 00:31:59.140
we decode someone else's watermark and we can
00:31:59.140 --> 00:32:01.920
say, hey, your content came to us. It actually
00:32:01.920 --> 00:32:03.799
belongs to you. You know how to decode that.
00:32:03.880 --> 00:32:08.140
And we work as a collaborative approach. It wasn't
00:32:08.140 --> 00:32:14.259
welcomed. I tell you what, I'm surprised. I really
00:32:14.259 --> 00:32:18.420
am. I grew up in a world where I thought when
00:32:18.420 --> 00:32:20.259
I grow up, it's never going to be like that.
00:32:20.779 --> 00:32:24.369
And I've grown up and I've realized It's capitalist
00:32:24.369 --> 00:32:29.170
hell out here. I hate it. I hope I can weave
00:32:29.170 --> 00:32:32.170
a little path out here and do things a bit differently.
00:32:32.930 --> 00:32:36.210
Well, we're all about collaboration here, hence
00:32:36.210 --> 00:32:39.670
this space. And we do have, by the way, as I'm
00:32:39.670 --> 00:32:42.069
on this subject, we do have a space in the server.
00:32:42.250 --> 00:32:44.250
If you want to contribute to Bellingcat tools,
00:32:44.450 --> 00:32:47.710
you can. It's called Tech Contributors. I'm sure
00:32:47.710 --> 00:32:50.009
one of the mods can link it. And then we also
00:32:50.009 --> 00:32:52.210
have a space called hashtag tool development,
00:32:52.349 --> 00:32:55.789
where you can work together building tools, open
00:32:55.789 --> 00:32:58.170
source ones, hopefully. But you can have a little
00:32:58.170 --> 00:33:01.150
chat in that space too. Somebody's asked, do
00:33:01.150 --> 00:33:02.950
you think that there is some merit in taking
00:33:02.950 --> 00:33:06.569
this product to T &S terms and services conferences
00:33:06.569 --> 00:33:09.730
that happen around the globe, just as a way to
00:33:09.730 --> 00:33:11.990
show the teams firsthand on how this can add
00:33:11.990 --> 00:33:14.230
value for users on their platforms? Have you
00:33:14.230 --> 00:33:17.309
ever thought about doing that? Yeah, we are actually
00:33:17.309 --> 00:33:22.099
presenting at Trustcon in the next few weeks.
00:33:23.680 --> 00:33:25.259
It's tough though, because it costs money, doesn't
00:33:25.259 --> 00:33:27.460
it, to go to these things. And we're only earning
00:33:27.460 --> 00:33:28.839
a little bit of money from the platforms that
00:33:28.839 --> 00:33:31.680
we're installed on. Look, it all kind of comes
00:33:31.680 --> 00:33:33.579
down to money, doesn't it? It sucks. I hate it.
00:33:33.799 --> 00:33:36.140
But yeah, it's something we're doing. It's something
00:33:36.140 --> 00:33:38.900
we're doing. The thing is I'm only as powerful
00:33:38.900 --> 00:33:44.319
as my team and as the presence of people who
00:33:44.319 --> 00:33:46.359
are kind of in the movement with us. We're only
00:33:46.359 --> 00:33:50.720
as powerful as those voices. And we're a minority
00:33:50.720 --> 00:33:53.079
voice. Like sex workers want this. Sex workers
00:33:53.079 --> 00:33:56.539
are a minority voice. No one listens, right?
00:33:57.420 --> 00:34:00.279
So to get where we've got to, I'm damn proud.
00:34:02.140 --> 00:34:05.960
And to be at TrustCon is pretty cool. But it
00:34:05.960 --> 00:34:09.380
all takes time or takes money, you know, effort
00:34:09.380 --> 00:34:13.510
and... I swear I have not seen my husband for
00:34:13.510 --> 00:34:17.690
a couple of bloody months. Poor guy. He's such
00:34:17.690 --> 00:34:20.269
a wonderful guy. He's really supported. This
00:34:20.269 --> 00:34:22.010
wouldn't be where it is if it weren't for him.
00:34:23.909 --> 00:34:27.489
In regards to money, you were talking at the
00:34:27.489 --> 00:34:33.710
beginning as to why you keep the service free
00:34:33.710 --> 00:34:38.929
for those that have been subjected to online
00:34:38.929 --> 00:34:42.239
image abuse. But platforms like Telegram, for
00:34:42.239 --> 00:34:45.920
example, are primary battlegrounds for information
00:34:45.920 --> 00:34:49.019
warfare. I mean, a lot of the Bellingcat investigations
00:34:49.019 --> 00:34:53.739
start in places like Telegram, but they fiercely
00:34:53.739 --> 00:34:56.039
guard their architectural privacy and refuse
00:34:56.039 --> 00:34:59.539
server -side moderation. If a platform refuses
00:34:59.539 --> 00:35:04.940
to integrate your API, is there a way that journalists
00:35:04.940 --> 00:35:10.949
or client -side solutions could work? For example,
00:35:11.389 --> 00:35:17.449
allowing journalists or investigators to use
00:35:17.449 --> 00:35:20.309
the tech. Is there a way that it could work on
00:35:20.309 --> 00:35:22.869
the client side where it doesn't necessarily
00:35:22.869 --> 00:35:27.130
have to have the platform's approval? I don't
00:35:27.130 --> 00:35:30.130
know. I'm going to say I don't know because I
00:35:30.130 --> 00:35:34.190
don't come from a tech background. We have an
00:35:34.190 --> 00:35:37.389
amazing tech team working with us. they are so
00:35:37.389 --> 00:35:39.329
much better place to answer those questions.
00:35:39.909 --> 00:35:44.869
But if a platform does not want a system like
00:35:44.869 --> 00:35:50.510
this, that's okay because we do have to have
00:35:50.510 --> 00:35:53.630
spaces and places that can be anonymous and you
00:35:53.630 --> 00:35:55.849
can go via VPN. You need to be able to do these
00:35:55.849 --> 00:35:59.690
things to be critical and to share things that
00:35:59.690 --> 00:36:03.389
need to be shared in kind of anonymous places.
00:36:03.900 --> 00:36:08.219
This tech is not about changing the whole infrastructure
00:36:08.219 --> 00:36:12.960
of the internet. It's about changing the mindset
00:36:12.960 --> 00:36:15.840
of people who view content that is privileged
00:36:15.840 --> 00:36:19.519
and private. How that happens in a space that
00:36:19.519 --> 00:36:23.719
won't do that is a tech question I cannot answer.
00:36:23.960 --> 00:36:24.980
And there's probably people in the chat that
00:36:24.980 --> 00:36:29.679
can answer that other than I can, for sure. Yeah,
00:36:29.679 --> 00:36:32.300
I think there's a cripple of questions about
00:36:32.300 --> 00:36:36.940
the technicalities of all of this. I'd love to
00:36:36.940 --> 00:36:40.139
dig into that a little bit more. Like for example,
00:36:41.059 --> 00:36:47.360
how do you ensure that if somebody's computer
00:36:47.360 --> 00:36:50.000
is hacked, for example, and these are leaked,
00:36:50.139 --> 00:36:52.559
how do you prevent those kind of false positives
00:36:52.559 --> 00:36:55.829
from happening? because as you mentioned, the
00:36:55.829 --> 00:36:58.690
kind of watermark gives the user ID. So has there
00:36:58.690 --> 00:37:01.329
been cases where that's happened or is there
00:37:01.329 --> 00:37:05.389
a way to deal with that kind of situation? So
00:37:05.389 --> 00:37:07.949
there's been no cases yet that have happened
00:37:07.949 --> 00:37:11.769
in that way. We've worked with a charity called
00:37:11.769 --> 00:37:16.329
Refuge to try and understand the way this could
00:37:16.329 --> 00:37:20.730
be used in a negative. What could happen to try
00:37:20.730 --> 00:37:24.219
and attack this? And we know that it is perfectly
00:37:24.219 --> 00:37:27.300
possible that someone will use this method to
00:37:27.300 --> 00:37:29.699
try and catch someone out to find data on someone
00:37:29.699 --> 00:37:33.159
and create cases that are malicious or get content
00:37:33.159 --> 00:37:35.579
taken down. And that's something that we still
00:37:35.579 --> 00:37:38.900
need to figure out ourselves. How do we verify
00:37:38.900 --> 00:37:42.000
the person who is giving us the information is
00:37:42.000 --> 00:37:44.320
the person they say they are? Is that their content?
00:37:44.460 --> 00:37:45.719
What information are they trying to find out?
00:37:45.980 --> 00:37:48.460
So in the first instance, what we do is if someone
00:37:48.460 --> 00:37:53.199
comes directly to us, as a company, we give them
00:37:53.199 --> 00:37:55.699
as much information as we feel is right at that
00:37:55.699 --> 00:37:58.440
time, which is, thank you very much for submitting
00:37:58.440 --> 00:38:01.480
this. We're looking into it. And that's all we
00:38:01.480 --> 00:38:04.420
do at that point. We then have a little look.
00:38:04.500 --> 00:38:07.019
We see what's in there. If there is no data,
00:38:07.559 --> 00:38:11.440
we don't say there is no data. We just say, we
00:38:11.440 --> 00:38:13.199
can't do anything with this at this time. We
00:38:13.199 --> 00:38:15.019
don't say whether or not there is, or there isn't,
00:38:15.019 --> 00:38:17.639
or we found anything. We just sort of say, you
00:38:17.639 --> 00:38:19.800
know, sorry, we can't help you today kind of
00:38:19.800 --> 00:38:22.920
thing. If we do find data in there, we don't
00:38:22.920 --> 00:38:26.500
tell them that either. We then go and find the
00:38:26.500 --> 00:38:28.280
information on the platform that it has been
00:38:28.280 --> 00:38:31.500
flagged so that we then know, is this a legitimate
00:38:31.500 --> 00:38:34.440
request? Because we know that there will be people
00:38:34.440 --> 00:38:37.940
who are trying to find content that is not watermarked
00:38:37.940 --> 00:38:40.699
so that they can then use that in a space for
00:38:40.699 --> 00:38:44.219
abuse or even find content that is watermarked.
00:38:44.300 --> 00:38:46.900
There's also that privacy around I'm a user.
00:38:47.119 --> 00:38:49.159
I've had my content leaked. I don't want to be
00:38:49.159 --> 00:38:51.460
seen as troublesome for the platform that I'm
00:38:51.460 --> 00:38:53.940
earning my money on. I'm already feel like I'm
00:38:53.940 --> 00:38:56.059
a target or I'm shadow banned. I don't want to
00:38:56.059 --> 00:38:59.139
be causing this company more stress and them
00:38:59.139 --> 00:39:01.820
think it's a me problem. So we also don't tell
00:39:01.820 --> 00:39:05.380
the platform who that end user is. It's a very
00:39:05.380 --> 00:39:09.239
complex dance between data and finding the right
00:39:09.239 --> 00:39:11.179
information without giving the right information
00:39:11.179 --> 00:39:15.440
over and it's complex. But we have a person entirely
00:39:15.440 --> 00:39:19.880
dedicated to both GDPR and then we have that
00:39:19.880 --> 00:39:22.800
support service who works on making sure that
00:39:22.800 --> 00:39:24.780
there is no harm happening in these kind of edge,
00:39:25.159 --> 00:39:29.219
weird, odd cases. In regards to, you mentioned
00:39:29.219 --> 00:39:31.800
at the beginning that, you know, if it's screen
00:39:31.800 --> 00:39:36.840
-shotted or posted across platforms, this forensic
00:39:36.840 --> 00:39:41.539
watermarking still tracks. Yeah. However, we
00:39:41.539 --> 00:39:45.579
see a lot of manipulation. If somebody mentioned
00:39:45.579 --> 00:39:48.360
earlier that one manipulation that's done is
00:39:48.360 --> 00:39:50.960
video inversion, for example, horizontal flipping.
00:39:52.079 --> 00:39:55.969
We also obviously see... deep fakes now as well
00:39:55.969 --> 00:40:00.610
of people taking original video of, for example,
00:40:01.210 --> 00:40:03.969
a professional sexual performer and putting a
00:40:03.969 --> 00:40:09.550
different face on that. In those cases, is the
00:40:09.550 --> 00:40:12.869
watermark strong enough to follow that kind of
00:40:12.869 --> 00:40:16.769
paper trail? Is there circumstances, for example,
00:40:16.829 --> 00:40:21.349
with AI manipulation where that trail could meet
00:40:21.349 --> 00:40:26.239
a dead end? Absolutely. If you're determined
00:40:26.239 --> 00:40:30.460
enough, you can break this. There is nothing
00:40:30.460 --> 00:40:33.659
that this will survive through if you are determined
00:40:33.659 --> 00:40:36.960
enough. We know people who are determined to
00:40:36.960 --> 00:40:40.019
do harm will do everything they can. This is
00:40:40.019 --> 00:40:45.739
designed for casual, I'm going to call it just
00:40:45.739 --> 00:40:51.139
opportunistic and silly leaks. People being stupid,
00:40:51.380 --> 00:40:54.849
opportunistic and thinking they might get away
00:40:54.849 --> 00:41:00.570
with it. This is not for creating a system that
00:41:00.570 --> 00:41:03.389
is unbreakable. It's impossible to create a system
00:41:03.389 --> 00:41:06.170
that's unbreakable. I would say that there's
00:41:06.170 --> 00:41:10.730
a spectrum of it. The more you do to an image,
00:41:10.989 --> 00:41:13.130
the less likely we are to be able to extract
00:41:13.130 --> 00:41:18.889
data. But also the more information we put in
00:41:18.889 --> 00:41:22.409
there, the less likely it is that we can extract
00:41:22.409 --> 00:41:26.780
it. because there's a payoff between how invisible
00:41:26.780 --> 00:41:29.980
it looks and how much data we can embed in there.
00:41:30.179 --> 00:41:35.079
So it's a dance again of how much information
00:41:35.079 --> 00:41:37.059
can we put so we can still identify you but you
00:41:37.059 --> 00:41:40.400
still can view the image of disruption versus
00:41:40.400 --> 00:41:43.840
how much manipulation can happen to it and every
00:41:43.840 --> 00:41:47.760
case. when it comes down to anything outside
00:41:47.760 --> 00:41:53.840
of that very generic crop flip minor edit will
00:41:53.840 --> 00:41:56.039
then be an individual case that we have a human
00:41:56.039 --> 00:42:00.099
review. And that's so important because we also
00:42:00.099 --> 00:42:04.500
don't want false negatives. Absolutely. There's
00:42:04.500 --> 00:42:09.119
always a hole in every method, in every tech.
00:42:09.579 --> 00:42:12.389
If you are looking to the future, In regards
00:42:12.389 --> 00:42:15.250
to this space, in regards to protective tech,
00:42:16.090 --> 00:42:19.110
what is the biggest technical or cultural hurdle
00:42:19.110 --> 00:42:24.829
you predict you might face in making this technology
00:42:24.829 --> 00:42:31.070
an uncompromised standard? What are the biggest
00:42:31.070 --> 00:42:34.010
obstacles facing protective tech in this space?
00:42:35.570 --> 00:42:37.670
You're talking a bit of quantum and we're all
00:42:37.670 --> 00:42:40.630
up a creek without a paddle, right? So if we
00:42:40.630 --> 00:42:43.130
were to go that far, we're done for, we're cooked.
00:42:43.909 --> 00:42:47.750
But in short term, I think the biggest hurdle
00:42:47.750 --> 00:42:52.889
is adoption. If, for example, every, let's say,
00:42:53.150 --> 00:42:55.789
easily findable space on the internet that our
00:42:55.789 --> 00:42:58.429
friends and family use, uses a system like this,
00:42:58.489 --> 00:43:00.510
then it will start to become colloquial knowledge
00:43:00.510 --> 00:43:03.969
that you don't do that. And then the more people
00:43:03.969 --> 00:43:06.989
who get kind of caught for doing negative things
00:43:06.989 --> 00:43:09.530
in spaces like this, it will kind of become known,
00:43:09.630 --> 00:43:12.250
oh yeah, that was like, did you know what happened
00:43:12.250 --> 00:43:14.750
to him? And it becomes kind of that colloquial
00:43:14.750 --> 00:43:16.190
understanding that you just don't do that. And
00:43:16.190 --> 00:43:19.210
that starts to create a social change of, we
00:43:19.210 --> 00:43:23.289
respect people in this kind of space. So adoption
00:43:23.289 --> 00:43:27.429
is probably the biggest thing. And I need people
00:43:27.429 --> 00:43:30.050
to want it. You know, is this something that
00:43:30.050 --> 00:43:35.320
people even want? Do we need people having data
00:43:35.320 --> 00:43:37.900
injected into imagery? Our research shows that
00:43:37.900 --> 00:43:40.599
it does, but if the wider world doesn't want
00:43:40.599 --> 00:43:44.340
to adopt this, then it doesn't go anywhere. Yeah,
00:43:44.480 --> 00:43:47.280
absolutely. I mean, you need to make sure that
00:43:47.280 --> 00:43:48.659
you're building it for the right people, right?
00:43:48.699 --> 00:43:50.579
But you did mention that you were saying that
00:43:50.579 --> 00:43:54.219
you, at the beginning, did ask that exact question
00:43:54.219 --> 00:43:57.179
to those that have been subjected to this abuse
00:43:57.179 --> 00:44:01.050
previously. Sarah, one of our amazing mods in
00:44:01.050 --> 00:44:03.429
the chats just asked, regarding humans reviewing
00:44:03.429 --> 00:44:07.010
complicated cases, are the people doing so compiling
00:44:07.010 --> 00:44:08.989
anything they're learning as they review these
00:44:08.989 --> 00:44:11.909
cases? EG, if they are able to see a pattern
00:44:11.909 --> 00:44:14.489
in the way Image Angel is outsmarted, is that
00:44:14.489 --> 00:44:18.059
being documented for future improvements? Yes,
00:44:18.059 --> 00:44:20.099
it is. Yeah. Everything that we do, we need to
00:44:20.099 --> 00:44:21.619
keep a record of it because then it improves
00:44:21.619 --> 00:44:24.000
our system. But also everything we do to our
00:44:24.000 --> 00:44:28.440
imagery also makes the kind of the anti, the
00:44:28.440 --> 00:44:30.400
people who were kind of trying to attack it stronger.
00:44:30.579 --> 00:44:32.199
So we're always, it's just cat and mouse. We're
00:44:32.199 --> 00:44:34.039
always trying to adjust and change. And that
00:44:34.039 --> 00:44:37.099
will always be the case. And I think that's great
00:44:37.099 --> 00:44:42.840
for innovation. We have ideas. for changing the
00:44:42.840 --> 00:44:45.320
way this works, for adding in extra data, for
00:44:45.320 --> 00:44:47.840
additional products that could support systems
00:44:47.840 --> 00:44:50.639
and take downs and searches and added in signals.
00:44:51.320 --> 00:44:53.599
It's all a part of development and they all will
00:44:53.599 --> 00:44:55.699
come in time and it all has to be for the right
00:44:55.699 --> 00:44:58.829
reasons. But yeah, everything. to be like, how
00:44:58.829 --> 00:45:02.610
are we learning? But also importantly, it's really
00:45:02.610 --> 00:45:05.929
vital that the people in that human in the loop
00:45:05.929 --> 00:45:09.469
don't view this kind of information and content.
00:45:09.489 --> 00:45:12.010
So we have to go through a really stringent process
00:45:12.010 --> 00:45:15.909
of keeping as much of that content blurred and
00:45:15.909 --> 00:45:18.250
blocked out as possible. So even if we do need
00:45:18.250 --> 00:45:20.670
to take that piece of information, how do we
00:45:20.670 --> 00:45:23.289
make sure that that user is protected and safe?
00:45:23.289 --> 00:45:25.409
And also that your nudes aren't being seen by
00:45:25.409 --> 00:45:29.190
some tech guy. We've had a couple of questions
00:45:29.190 --> 00:45:32.750
about court cases. I don't know if this tech
00:45:32.750 --> 00:45:36.889
has ever been used in a legal case, but they've
00:45:36.889 --> 00:45:38.809
asked if this is used in court, does having a
00:45:38.809 --> 00:45:41.050
watermark have positive effect on the outcome?
00:45:41.170 --> 00:45:43.309
Similar to having a lock on a property, how it's
00:45:43.309 --> 00:45:45.409
approving that the thief saw it and knew he is
00:45:45.409 --> 00:45:47.909
breaking in. I feel like regardless of the technical
00:45:47.909 --> 00:45:49.809
part being a hundred percent, this could be a
00:45:49.809 --> 00:45:52.289
big benefit in the legal part of the issue. Have
00:45:52.289 --> 00:45:56.190
you interacted with that? legal procedure world
00:45:56.190 --> 00:45:58.829
yet? Because obviously that's something that
00:45:58.829 --> 00:46:01.750
a lot of we do with war crimes, for example,
00:46:01.889 --> 00:46:04.929
war crime imagery and manipulated imagery. We
00:46:04.929 --> 00:46:10.650
have previously provided information for legal
00:46:10.650 --> 00:46:13.469
processes. So is that something that you kind
00:46:13.469 --> 00:46:17.750
of work with on your side? Yeah, and it's something
00:46:17.750 --> 00:46:19.949
that we want to do more of. The problem is, again,
00:46:20.090 --> 00:46:23.030
resources. We're still a startup. We're still
00:46:23.030 --> 00:46:30.949
scaling. What we do is test, ask, nudge everyone
00:46:30.949 --> 00:46:35.369
all the time. What we know is that data embedded
00:46:35.369 --> 00:46:40.750
in imagery is not unusual in court cases. There's
00:46:40.750 --> 00:46:44.769
metadata, there's even information like a username,
00:46:44.789 --> 00:46:47.489
a handle that could be kind of linked back to
00:46:47.489 --> 00:46:49.070
the person who created the deepfake. And there
00:46:49.070 --> 00:46:50.969
have been cases that have been successfully prosecuted
00:46:50.969 --> 00:46:53.969
in those cases. With the Image Angel, not yet.
00:46:54.010 --> 00:46:56.210
But hopefully that'll be the case. And then when
00:46:56.210 --> 00:46:58.150
it comes to AI manipulation and how much data
00:46:58.150 --> 00:47:01.070
is still left in there, it is possible to still
00:47:01.070 --> 00:47:05.230
see information held within an image. Only we
00:47:05.230 --> 00:47:08.429
can't read the exact data, but the presence of
00:47:08.429 --> 00:47:11.230
the regeneration of those pixels proves that
00:47:11.230 --> 00:47:13.829
there was an original image that came from that
00:47:13.829 --> 00:47:16.059
non -conceptual space. and just the presence
00:47:16.059 --> 00:47:19.880
of it alone. Even though we cannot record or
00:47:19.880 --> 00:47:23.139
find enough robust data to pinpoint exactly who
00:47:23.139 --> 00:47:27.099
it was, the presence of it there, even visually,
00:47:27.619 --> 00:47:30.440
denotes that there was something there and that
00:47:30.440 --> 00:47:33.320
can go along to supporting. I know a lot of people
00:47:33.320 --> 00:47:35.519
are going to say, oh, well, I had my device stolen.
00:47:35.719 --> 00:47:39.579
I was packed. The biggest defense when somebody
00:47:39.579 --> 00:47:43.719
is being accused of intimate image abuse or deep
00:47:43.719 --> 00:47:46.199
fake creation, for example, is I was hacked,
00:47:46.539 --> 00:47:48.820
I had my phone stolen. That may very well be
00:47:48.820 --> 00:47:53.280
the case and life is like shit. Things happen,
00:47:53.460 --> 00:47:57.659
bad things happen. What we could also do in this
00:47:57.659 --> 00:48:03.780
evidence gathering is prove a kind of a pattern
00:48:03.780 --> 00:48:07.019
of abuse and we may be able to track somebody's
00:48:07.019 --> 00:48:10.039
abusing across a matter of months or weeks. So
00:48:10.039 --> 00:48:11.760
it is quite possible that someone had the device
00:48:11.760 --> 00:48:14.340
taken and stolen and accessed then, but would
00:48:14.340 --> 00:48:17.440
that pattern continue? Who knows? It's all adding
00:48:17.440 --> 00:48:22.239
evidence to support people. If you are interested
00:48:22.239 --> 00:48:25.400
in this particular subject, I definitely recommend
00:48:25.400 --> 00:48:28.039
listening back to our last stage talk with the
00:48:28.039 --> 00:48:31.000
CNN team who spoke about some of the work that
00:48:31.000 --> 00:48:33.900
they've done, uncovering communities who were
00:48:33.900 --> 00:48:36.739
sharing, online communities who were sharing
00:48:36.739 --> 00:48:43.519
intimate imagery. that was either filmed or was
00:48:43.519 --> 00:48:47.420
leaked. They were into detail about the accountability
00:48:47.420 --> 00:48:50.380
there and how they conducted those investigations.
00:48:50.619 --> 00:48:52.480
It's definitely worth listening back to and those
00:48:52.480 --> 00:48:54.400
links are in the chat and also will be in the
00:48:54.400 --> 00:48:55.739
description of the podcast if you're listening
00:48:55.739 --> 00:48:59.199
back. And thank you, Sarah, for pinning that
00:48:59.199 --> 00:49:03.940
again. We've only got a couple of minutes left,
00:49:03.940 --> 00:49:06.519
so I'm going to have to squeeze in the last few
00:49:06.519 --> 00:49:09.320
questions now. We had a question earlier. saying,
00:49:09.320 --> 00:49:12.079
has there been any resistance in adoption from
00:49:12.079 --> 00:49:14.599
the OnlyFans sex worker space? I figured that
00:49:14.599 --> 00:49:16.599
would be an audience open to watermarking. You
00:49:16.599 --> 00:49:19.980
mentioned that you consulted a few people from
00:49:19.980 --> 00:49:23.059
that space, but has there been any resistance
00:49:23.059 --> 00:49:28.280
in adoption from those spaces? No, all of my
00:49:28.280 --> 00:49:31.179
friends I would say now are sex workers. I had
00:49:31.179 --> 00:49:34.099
such a wonderful community around me, sex workers,
00:49:34.340 --> 00:49:38.239
creators, you name it. artists of all sorts of
00:49:38.239 --> 00:49:41.380
walks of life. There has been no resistance to
00:49:41.380 --> 00:49:43.940
it. The only sort of thought was, oh, wait, wait,
00:49:44.019 --> 00:49:46.360
wait, are you putting my data in it as a creator?
00:49:46.519 --> 00:49:48.619
You know, can that person track who I am? But
00:49:48.619 --> 00:49:50.619
no, that's, you don't need to prove who you are.
00:49:50.920 --> 00:49:52.860
It's your face or it's, you know, you've got
00:49:52.860 --> 00:49:56.880
the original. No, and even in the consumer space
00:49:56.880 --> 00:50:00.000
as well. A lot of people I've been speaking to
00:50:00.000 --> 00:50:02.619
are consumers of pornography and adult content
00:50:02.619 --> 00:50:07.130
and social media and stuff. Nobody has said,
00:50:07.130 --> 00:50:10.710
I don't want that. And I think it's because we're
00:50:10.710 --> 00:50:14.289
used to data doing things. We know that if we
00:50:14.289 --> 00:50:16.510
are responsible for something online, it can
00:50:16.510 --> 00:50:18.869
be traced back to us. Our search history is,
00:50:18.869 --> 00:50:20.630
you know, it's possible to link these things
00:50:20.630 --> 00:50:23.150
back to us. We're never entirely anonymous. I
00:50:23.150 --> 00:50:25.030
think it's not beyond our understanding that
00:50:25.030 --> 00:50:27.690
this is a thing. And the only people who would
00:50:27.690 --> 00:50:31.130
potentially be opposed to it, and I welcome opposition,
00:50:31.750 --> 00:50:33.869
are people who would want to do something negative
00:50:33.869 --> 00:50:36.199
with that. The one thing that we need to make
00:50:36.199 --> 00:50:40.719
sure is that this cannot be used in a negative
00:50:40.719 --> 00:50:43.019
way. And that's something that will always be
00:50:43.019 --> 00:50:46.199
evolving. Yeah, absolutely. I would potentially
00:50:46.199 --> 00:50:49.900
challenge your notion that everybody knows how
00:50:49.900 --> 00:50:52.199
much information is behind each thing that they
00:50:52.199 --> 00:50:56.110
upload. The stuff that we find in metadata is
00:50:56.110 --> 00:50:59.489
sometimes hilarious or source code of websites.
00:51:00.550 --> 00:51:02.989
If you want a reference, go check out our work
00:51:02.989 --> 00:51:07.570
on the Kinahan Cartel and see kind of what professional
00:51:07.570 --> 00:51:11.170
criminal organizations are sharing online. You'll
00:51:11.170 --> 00:51:15.590
see the people's understanding of what is being
00:51:15.590 --> 00:51:18.690
shared. Right, we are nearly end of time. So
00:51:18.690 --> 00:51:22.059
I wanted to ask, as this talk is really about
00:51:22.059 --> 00:51:24.000
building protective tech. What would be your
00:51:24.000 --> 00:51:27.199
advice to people who perhaps are listening and
00:51:27.199 --> 00:51:30.519
want to go out and build something similar themselves?
00:51:30.739 --> 00:51:33.780
What would your advice be to those who are looking
00:51:33.780 --> 00:51:38.659
to develop protective tech? I think it's important
00:51:38.659 --> 00:51:40.579
that we stop and pause and think, who am I building
00:51:40.579 --> 00:51:42.340
this for and what are the potential negatives?
00:51:42.760 --> 00:51:44.360
Well, how could this be manipulated and used
00:51:44.360 --> 00:51:46.719
in a negative way? If we're going to build good
00:51:46.719 --> 00:51:48.579
stuff, we have to make sure that it's truly good.
00:51:48.679 --> 00:51:50.840
It really, really is good and not just good in
00:51:50.840 --> 00:51:53.659
our minds or in our projection of our world.
00:51:53.780 --> 00:51:56.320
Is this good for the global South? Is this good
00:51:56.320 --> 00:52:00.639
for old people? Is this good for more people
00:52:00.639 --> 00:52:04.719
than I can conceive? And that's what we try to
00:52:04.719 --> 00:52:07.159
do is stress test in as many different environments
00:52:07.159 --> 00:52:11.820
as possible. But also I would say, don't do what
00:52:11.820 --> 00:52:15.719
I did and wait. I took this idea to hundreds
00:52:15.719 --> 00:52:17.860
of people, I would think, over the time, you
00:52:17.860 --> 00:52:19.460
know, hundreds of platforms like, oh, I've got
00:52:19.460 --> 00:52:21.480
an idea, would you like to develop this? Everyone
00:52:21.480 --> 00:52:24.480
said, no, it's not possible. And I thought, no,
00:52:25.260 --> 00:52:28.119
it really is possible and it's really important.
00:52:28.119 --> 00:52:31.380
And being ready to do that was a choice. I had
00:52:31.380 --> 00:52:32.960
to make that choice. Like I'm never, you're never
00:52:32.960 --> 00:52:34.699
going to be financially ready, emotionally ready.
00:52:34.840 --> 00:52:36.500
Your home life is never going to be settled enough.
00:52:36.760 --> 00:52:37.940
If you want to go out and make a difference,
00:52:38.000 --> 00:52:41.480
you got to go and grab it. It's a choice. You
00:52:41.480 --> 00:52:45.420
need to make the choice. And if you're not feeling
00:52:45.420 --> 00:52:48.260
inspired enough already, as I said, do go look
00:52:48.260 --> 00:52:51.280
at hashtag tech development in here because there's
00:52:51.280 --> 00:52:54.199
lots of people trying to do the same. We had
00:52:54.199 --> 00:52:56.079
a question earlier and I promised them that I
00:52:56.079 --> 00:52:58.820
would ask this towards the end. Are you hiring
00:52:58.820 --> 00:53:01.880
and or supporting contributors to this project
00:53:01.880 --> 00:53:05.340
as you're building it out? Because I think you
00:53:05.340 --> 00:53:10.420
have a lot of sans in this space now. I need
00:53:10.420 --> 00:53:13.579
fans. Every day I struggle with like, am I doing
00:53:13.579 --> 00:53:17.099
the right thing? I've got imposter syndrome to
00:53:17.099 --> 00:53:21.219
the max. Are we hiring? We're not directly hiring
00:53:21.219 --> 00:53:24.059
because we are growing slow and growing well,
00:53:24.579 --> 00:53:27.960
but we want to build a database of people who
00:53:27.960 --> 00:53:30.000
want to work for us so that when we have those
00:53:30.000 --> 00:53:33.349
money and that time, we give back. I want to
00:53:33.349 --> 00:53:35.190
hire the best brains. If you want to work for
00:53:35.190 --> 00:53:38.730
us, please come and say hi. If you have a thought,
00:53:39.010 --> 00:53:41.449
I'd even love to know, what do you hate about
00:53:41.449 --> 00:53:44.550
it? What do you challenge? What do you want to
00:53:44.550 --> 00:53:47.949
do? What do you want to hear and say? I'm putting
00:53:47.949 --> 00:53:51.469
my heart out to the Bellingcat community. Just
00:53:51.469 --> 00:53:53.349
tell me what you think. I'd love to hear your
00:53:53.349 --> 00:53:59.030
feedback. And how can they do that would be my
00:53:59.030 --> 00:54:02.940
follow up question. Probably Instagram is probably
00:54:02.940 --> 00:54:07.199
the best way. We also have TikTok. We have LinkedIn.
00:54:07.619 --> 00:54:12.400
We have an X, but I don't like to use it. And
00:54:12.400 --> 00:54:14.480
all of it will be linked on our website. But
00:54:14.480 --> 00:54:16.519
if you search Image Angel with an underscore
00:54:16.519 --> 00:54:20.559
at the end, you'll likely find us. I'm just dropping
00:54:20.559 --> 00:54:23.280
that Instagram link into the chat now, and I'll
00:54:23.280 --> 00:54:24.980
include it in the description of the podcast.
00:54:26.039 --> 00:54:27.880
Finally, finally, finally, as I mentioned at
00:54:27.880 --> 00:54:30.739
the start, Bellingcat doesn't endorse individual
00:54:30.739 --> 00:54:33.239
companies. So I wanted to ask you if there's
00:54:33.239 --> 00:54:36.360
anybody else in this space that you would want
00:54:36.360 --> 00:54:39.039
to shout out? For example, is there any other
00:54:39.039 --> 00:54:42.320
resources that have been vital to your building
00:54:42.320 --> 00:54:45.599
of this project or any other companies that are
00:54:45.599 --> 00:54:48.380
doing something cool that you think should be
00:54:48.380 --> 00:54:53.400
on people's radars? Oh, my God. Oh, my God. Oh,
00:54:53.480 --> 00:54:57.239
my God. That's completely blank. That probably
00:54:57.239 --> 00:55:03.500
is. I would say Not Your Porn are a really good
00:55:03.500 --> 00:55:07.840
group of people. Not Your Porn is a group that
00:55:07.840 --> 00:55:11.619
victims, allies, and they just jump in and try
00:55:11.619 --> 00:55:16.320
and lobby and get change to happen. It's a not
00:55:16.320 --> 00:55:20.340
for profit. They are volunteer run. They are
00:55:20.340 --> 00:55:24.590
awesome. Go and support Not Your Porn. I just
00:55:24.590 --> 00:55:27.010
dropped their LinkedIn in the chat there. I didn't
00:55:27.010 --> 00:55:29.369
have, I couldn't find the website that quickly,
00:55:29.590 --> 00:55:33.570
but it is there. And hopefully you can explore
00:55:33.570 --> 00:55:35.630
that and I'll make sure to find the proper link.
00:55:36.010 --> 00:55:39.130
Oh, thank you people in the chat for the description
00:55:39.130 --> 00:55:42.449
of the podcast. All right. We are about time.
00:55:42.670 --> 00:55:45.429
Thank you so much Madelaine, for your time today
00:55:45.429 --> 00:55:49.389
and for going through this amazing tech. It's
00:55:49.389 --> 00:55:51.250
really fascinating to hear a little bit about.
00:55:51.530 --> 00:55:55.670
your solution for this massive problem. And hopefully
00:55:55.670 --> 00:55:58.730
it's inspired others in the audience to think
00:55:58.730 --> 00:56:01.130
about how they can shape their own solutions
00:56:01.130 --> 00:56:04.889
for this issue. A lot of love in the comments.
00:56:05.849 --> 00:56:08.210
In two weeks time, we will be back with another
00:56:08.210 --> 00:56:11.090
stage talk. It is not going to be on this subject.
00:56:11.670 --> 00:56:13.590
We have covered this subject quite a bit over
00:56:13.590 --> 00:56:15.929
the last few weeks. We are going to be moving
00:56:15.929 --> 00:56:19.150
to environmental. investigations in two weeks
00:56:19.150 --> 00:56:22.929
time. We have a fantastic environmental journalist
00:56:22.929 --> 00:56:26.210
coming in who recently won environmental journalist
00:56:26.210 --> 00:56:28.269
of the year awards, and they're going to be coming
00:56:28.269 --> 00:56:33.230
in to talk about how you can do environmental
00:56:33.230 --> 00:56:36.489
investigations digitally. So that's what the
00:56:36.489 --> 00:56:38.969
next stage talk will be about. But for now, thank
00:56:38.969 --> 00:56:42.190
you again, Madelaine. Thank you for having me.
00:56:42.650 --> 00:56:45.519
Thank you for listening to the stage talk. If
00:56:45.519 --> 00:56:48.500
you'd like to catch a stage tour live where you
00:56:48.500 --> 00:56:51.400
can ask the guest questions, join the Bellingcat
00:56:51.400 --> 00:56:57.780
Discord server by visiting www .discord .gg slash
00:56:57.780 --> 00:57:00.679
Bellingcat. The music you've heard is titled
00:57:00.679 --> 00:57:04.619
Dawn by Newer Self and is courtesy of Artlist.