1
00:00:02,260 --> 00:00:06,370
[Auto-generated transcript. Edits may have been applied for clarity.]
I. To say.

2
00:00:10,460 --> 00:00:14,240
All right along with you. Good afternoon.

3
00:00:15,890 --> 00:00:21,650
So let me start by a quick introduction. Um, I'm even Martino, which I'm acting head of department.

4
00:00:21,650 --> 00:00:25,040
This academic here in in computer science department.

5
00:00:25,040 --> 00:00:30,169
And it is my great pleasure to introduce to you, professor surgeon champion, um, visiting us.

6
00:00:30,170 --> 00:00:36,980
Um, from Keith Turek, a fellow Croatian, um, surgeon is a full professor at MPH.

7
00:00:37,010 --> 00:00:40,820
He's also chair of Zurich, uh, Information Security Institute.

8
00:00:41,300 --> 00:00:42,740
Uh, um, for security.

9
00:00:42,740 --> 00:00:55,250
Privacy associate by, uh, associate VP for, uh, digital transformation and government at and uh, among all these roles, you also do some research.

10
00:00:55,250 --> 00:01:00,799
So you just, I believe, finished your ERC consolidator grant on secure position in various networks.

11
00:01:00,800 --> 00:01:06,709
So, um, his work is, uh, the boundary between the physical and digital domain,

12
00:01:06,710 --> 00:01:12,740
starting with analysis of, uh, physical properties, meaning signal propagation,

13
00:01:12,740 --> 00:01:18,229
wireless communication, and actually using physical laws to create security properties such as,

14
00:01:18,230 --> 00:01:22,220
for example, something we're going to hear about, um, called distance ball.

15
00:01:22,820 --> 00:01:30,350
And, uh, working all the way up through smartphone sensors to the satellites and digital infrastructures.

16
00:01:30,800 --> 00:01:35,540
Maybe just to give you a quick example, around 2010, you know each other for some time.

17
00:01:35,540 --> 00:01:45,500
So around 2010, he published the first paper on the vulnerabilities of so-called PCA system of a PKC suppressive keyless entry system.

18
00:01:46,580 --> 00:01:51,049
Now, this key but many high end cars back then. Now it's more of a commodity.

19
00:01:51,050 --> 00:01:58,040
But back then it was um, it was quite famous and expensive to have a key that would allow you higher level of usability.

20
00:01:58,220 --> 00:02:04,760
Basically, you would just need to be near to open the car and enter the vehicle and, and such.

21
00:02:04,760 --> 00:02:09,889
And most of the first one to have a systematic evaluation of such a system, showing, of course,

22
00:02:09,890 --> 00:02:19,130
that there is a logical breakage in design because, uh, ability to communicate does not imply physical proximity.

23
00:02:19,310 --> 00:02:23,540
And so he was the first one to introduce relay attacks on that system.

24
00:02:24,380 --> 00:02:29,180
At the same time. When are you asking what the industry has done about it?

25
00:02:29,180 --> 00:02:35,149
Because as a security researcher, you are you are required to do this, uh, something called the responsible disclosure.

26
00:02:35,150 --> 00:02:42,080
So he and his team went to all the car manufacturers, and actually the car industry did nothing.

27
00:02:42,080 --> 00:02:45,680
Right. They said, well, the usability is a great important sales factor.

28
00:02:45,860 --> 00:02:49,040
And, you know, um, these events won't happen today.

29
00:02:49,550 --> 00:02:55,700
You can go to YouTube or, you know, there are millions of videos how expensive cars,

30
00:02:55,700 --> 00:03:01,550
especially Teslas and similar ones, are basically being taken away in the middle of the night.

31
00:03:01,700 --> 00:03:05,960
Interestingly, my neighbour also got his brand new BMW that way.

32
00:03:06,350 --> 00:03:12,260
It was picked up by my Rinke, uh, smart, uh, right, uh, doorbell.

33
00:03:13,190 --> 00:03:18,200
But it turned off notifications. So yeah. So so yeah, it was late.

34
00:03:18,770 --> 00:03:27,799
Anyway, um, because the industry did nothing certain, uh, created a secure distance measurement system there.

35
00:03:27,800 --> 00:03:33,709
Then in the spinoff company that eventually, later on was a client and then successfully.

36
00:03:33,710 --> 00:03:38,960
And today in most of the car keys, you have surgeons secure distance measurement protocol.

37
00:03:39,860 --> 00:03:43,339
Okay. So, um, with this, I will stop.

38
00:03:43,340 --> 00:03:49,010
And, uh, before I finish a bit of housekeeping, first of all, uh, the strategy lecture series,

39
00:03:49,010 --> 00:03:56,510
our distinguished series is generously funded by Oxfam, Oxford Asset Management and, uh, yeah, thankful for that.

40
00:03:56,810 --> 00:04:02,180
And also, please stay after the lecture for some tea, coffee and biscuits just around the corner here.

41
00:04:02,480 --> 00:04:06,170
And with this, uh, let's welcome together Professor Chapman.

42
00:04:12,760 --> 00:04:16,390
So thank you, Yvonne, and thank you all for, uh, for coming.

43
00:04:16,900 --> 00:04:23,740
I'm I'm going to try to cover two topics. One is more kind of mundane and related to, uh, smartphones.

44
00:04:24,070 --> 00:04:32,469
And the other one is, is related to, uh, maybe our, our future and cyber physical systems that will be deployed and the and the system and

45
00:04:32,470 --> 00:04:35,830
what I believe are the systems that we need to build in order to secure our future.

46
00:04:35,830 --> 00:04:38,920
Because otherwise, I, I think it's going to be very scary.

47
00:04:39,400 --> 00:04:43,540
So let me, uh, start with, uh, with one simple question.

48
00:04:44,480 --> 00:04:47,810
And that's do you truly own your your phone and.

49
00:04:47,840 --> 00:04:58,490
And here the question is, is somehow similar to the one that you or the to the test that you can you can do when you ask yourself the if you own,

50
00:04:58,880 --> 00:05:04,940
if you truly own a house that you live in. And a good test is to try to nail a picture on the wall.

51
00:05:05,480 --> 00:05:11,470
Right. So the similar thing you can do with a phone, you can ask yourself, what can I actually do with this phone?

52
00:05:11,480 --> 00:05:17,270
So how much access do I have, uh, to to what I want to do on this phone.

53
00:05:17,270 --> 00:05:27,259
So whether there is some hardware that you want to modify or that you want to configure in some way, uh, you sit over there.

54
00:05:27,260 --> 00:05:31,760
So you being the apps potentially that you can develop and install,

55
00:05:32,120 --> 00:05:41,149
whereas there is this whole ecosystem below not only of operating systems but potentially hypervisors.

56
00:05:41,150 --> 00:05:43,880
There's a lot of firmware that's being deployed on these phones.

57
00:05:44,150 --> 00:05:50,750
So there is a lot of control that is being exercised for good or bad by different entities in the in this ecosystem.

58
00:05:50,750 --> 00:05:56,480
And these could be either big tech entities like Google, Samsung, Apple, uh,

59
00:05:56,510 --> 00:06:02,989
Qualcomm and, and so forth, or even a lot of, uh, third party, uh, entities.

60
00:06:02,990 --> 00:06:11,600
And what you can see here already, if you're unfamiliar with, uh, with the phones, is that the phones are divided into or at least this arm, uh,

61
00:06:11,600 --> 00:06:19,640
platform is divided into what's called the normal world and the secure world, which is, uh, typically running something like arm transition.

62
00:06:19,910 --> 00:06:21,500
Right. So that's the architecture.

63
00:06:22,040 --> 00:06:33,620
And, um, there you already see a division between secure services and potentially, uh, normal, uh, execution of of, let's say, users users code.

64
00:06:35,630 --> 00:06:40,850
That part is also heavily controlled by by different commercial entities.

65
00:06:40,850 --> 00:06:46,460
And as a user, you typically have no access to these secure services you can't easily configure.

66
00:06:49,240 --> 00:06:54,190
So this is not only about control. So this ecosystem is is quite is quite large.

67
00:06:54,200 --> 00:07:04,770
We have hundreds of mil. I mean it's hard to estimate, but there are 100 over 100 million lines of code that are that your application sits on top of.

68
00:07:04,780 --> 00:07:10,060
So if even if you have something like a secure messenger running on top of your your code,

69
00:07:10,420 --> 00:07:17,320
it's, it sits on this large ecosystem of, of um of code which.

70
00:07:18,430 --> 00:07:22,330
Also then results in vulnerabilities being being disclosed.

71
00:07:22,840 --> 00:07:25,239
Uh, every once in a while, a lot of them are zero day,

72
00:07:25,240 --> 00:07:33,010
meaning that effectively you don't really know about these vulnerabilities before they are publicly released,

73
00:07:33,010 --> 00:07:38,050
but they have been out there in the wild and and potentially have compromised your phone.

74
00:07:38,950 --> 00:07:48,310
And clearly, with these AI tools, tools coming up that can analyse large code bases, we can expect more such vulnerabilities to to appear.

75
00:07:49,700 --> 00:07:52,910
This is not surprising or shocking to any computer scientist.

76
00:07:53,360 --> 00:07:58,910
This is just a link to some of these zero days published by Google.

77
00:07:59,360 --> 00:08:07,939
So they keep on coming. And so you could even even say that at any given point in time, you're probably exposed to some of them on your on your phone,

78
00:08:07,940 --> 00:08:13,880
and they might be weaponized against you, or it depends on what kind of target you tend to be.

79
00:08:15,380 --> 00:08:20,150
And this is I'm not the first one to notice. Reports about security vulnerabilities are everywhere.

80
00:08:20,700 --> 00:08:25,130
Uh, we can see politicians, people being targeted.

81
00:08:25,520 --> 00:08:30,560
Also entire organisations being being targeted. The same applies to control.

82
00:08:31,400 --> 00:08:41,330
Um, in terms of controlling this, this ecosystem and effectively some of these entities acting as potential gatekeepers to,

83
00:08:41,390 --> 00:08:44,990
uh, to the functionality of the phone. There are numerous reports on this.

84
00:08:45,470 --> 00:08:48,890
Uh, the latest one you can maybe see by graphene, for example,

85
00:08:48,890 --> 00:08:54,890
which is a secure operating system that you can run on your on your phone complaining about attestation,

86
00:08:54,890 --> 00:09:05,090
being, uh, being forced into some, um, application ecosystems and effectively breaking some of the functionality on their, on their phones.

87
00:09:05,510 --> 00:09:14,780
So this is this is clearly a problem, right, that we have that we have entities that provide a lot of service.

88
00:09:15,800 --> 00:09:22,370
So they create these rich ecosystems. But in some sense we are dependent on them.

89
00:09:22,700 --> 00:09:26,719
Right. So we are dependent on them to control access to resources.

90
00:09:26,720 --> 00:09:33,920
Because the way that computer systems are designed is that is that someone needs to manage access to shared resources like CPUs,

91
00:09:33,920 --> 00:09:36,380
memory, uh, peripherals and so on.

92
00:09:36,590 --> 00:09:47,720
And we put this big fat operating system in between, plus other layers in order to, to mediate this access and through a a set of privacy, security,

93
00:09:47,750 --> 00:09:56,450
uh, arguments, you can then then try to argue why certain access of an application is to an A from an application is actually restricted.

94
00:09:59,060 --> 00:10:05,270
So this means effectively that in today's uh, architectures, access to peripherals such as screens, networks,

95
00:10:05,540 --> 00:10:17,180
every sensor needs to be mediated by these uh, OEM, uh operating system providers like Apple, Apple, uh, Google, Samsung and so on.

96
00:10:17,780 --> 00:10:20,660
It is not only about control, it's also about availability.

97
00:10:20,870 --> 00:10:31,790
So if those entities decide to, to, um, stop their service, your functionality, whatever that is, would, would equally, equally stop.

98
00:10:32,240 --> 00:10:40,700
And also it's about configurability, which is lack of developer control over what they want to to do with the phone.

99
00:10:41,000 --> 00:10:50,810
I was involved in a broader, um, effort by ePFl and back then during the pandemic, where we were working on contact tracing frameworks.

100
00:10:51,620 --> 00:10:56,660
One of them was, in the end, adopted and deployed by Apple and Google.

101
00:10:57,080 --> 00:11:01,820
And one of our main challenges was just access to Bluetooth and access to Bluetooth beacons.

102
00:11:02,120 --> 00:11:08,540
And this was heavily guarded for, for, uh, some good reasons by, by Apple and Google.

103
00:11:08,540 --> 00:11:17,869
And this led to a lot of frustration, um, among among researchers because many ideas that we had even in that state of,

104
00:11:17,870 --> 00:11:23,360
of, let's say, uh, global panic, we could not implement, um, our ideas.

105
00:11:24,710 --> 00:11:28,370
So clearly you can run alternative ecosystems, right?

106
00:11:28,400 --> 00:11:33,830
You could say, I don't like this ecosystem dominated by my big tech players.

107
00:11:34,130 --> 00:11:37,160
I'm going to switch to two alternatives.

108
00:11:37,160 --> 00:11:40,160
And these might be some Linux phones. It might be graphene.

109
00:11:40,160 --> 00:11:45,500
If I built alternative ecosystems that give you potentially more control over over the phone.

110
00:11:46,220 --> 00:11:53,510
But they also have a restricted functionality because you might not no longer be able to run an app that, um,

111
00:11:53,960 --> 00:11:59,510
of your bank or maybe some favourite app that potentially requires a different type at the station,

112
00:12:00,080 --> 00:12:03,710
or wants to be run only within a Google or Apple ecosystem.

113
00:12:04,280 --> 00:12:07,280
So this is you lose a lot of functionality.

114
00:12:09,860 --> 00:12:14,310
So this means that we essentially are facing sort of a choice.

115
00:12:14,450 --> 00:12:16,850
I do carry two phones with you.

116
00:12:16,880 --> 00:12:26,840
You have a, a secure phone and, and a sort of general phone, uh, on, on which you can install some less critical application or,

117
00:12:26,960 --> 00:12:31,580
or you just suffer with the consequences of, of each of these ecosystems.

118
00:12:32,120 --> 00:12:36,709
So we raised this question some, some years ago. Why does it have to be like this?

119
00:12:36,710 --> 00:12:41,510
And can we actually design something else. So can we have nice nice things in a nutshell.

120
00:12:41,750 --> 00:12:49,510
So a phone that is actually secure, open, developer friendly and can actually be controlled by the user.

121
00:12:49,520 --> 00:12:56,120
But I'm going to refine this, this statement because I, we didn't want full control by the users.

122
00:12:56,360 --> 00:13:02,700
We want actually shared control. So this phone needs to be controlled by your user,

123
00:13:02,700 --> 00:13:08,610
but at the same time functional and compatible with these large ecosystems like Android, iOS and so on.

124
00:13:10,450 --> 00:13:18,610
So we started this in 2020, started thinking about this, and then published the first paper in in 2021.

125
00:13:19,450 --> 00:13:23,630
It was a really difficult publication process, I can tell you, because, uh,

126
00:13:23,740 --> 00:13:29,140
some of the ideas that we wanted to push, they didn't find immediate acceptance in the, in the community,

127
00:13:29,410 --> 00:13:33,969
but effectively we wanted to create a phone that contains several phones, for example,

128
00:13:33,970 --> 00:13:44,710
to where you can run one E.M.S. or more and one or more domains that execute some secure apps.

129
00:13:45,280 --> 00:13:49,540
But those apps should have full access to the hardware peripherals.

130
00:13:49,540 --> 00:13:59,919
They should be able to run bare metal, essentially. So we wanted to move from this ecosystem on the left where we have this monolithic, let's say,

131
00:13:59,920 --> 00:14:09,490
monolithic design to this system on the right where we would create ecosystems that are equally privileged.

132
00:14:09,640 --> 00:14:21,450
So we didn't want to create a phone where the user dominates these large OEM ecosystems, because this might break their their security model.

133
00:14:21,460 --> 00:14:21,880
Right.

134
00:14:22,210 --> 00:14:34,300
So we wanted to build a phone where you can execute your workflow bare metal with full access to peripherals in parallel with the existing commercial,

135
00:14:34,840 --> 00:14:38,229
uh, phones. And what can you do that there?

136
00:14:38,230 --> 00:14:44,470
You could create domains, for example, where you run in parallel with, let's say, Android, another secure operating system,

137
00:14:44,740 --> 00:14:52,450
or even a self-contained bare metal application that runs directly on the on on the course.

138
00:14:53,170 --> 00:14:59,680
So one of these, one of these um, domains needs to be protected from the user.

139
00:15:00,010 --> 00:15:04,000
So whatever the user does on the phone and needs to be controlled by um,

140
00:15:04,360 --> 00:15:08,890
and others potentially need to be controlled by the user or in any combination of those.

141
00:15:11,980 --> 00:15:15,220
So we are talking about really equal privilege.

142
00:15:15,790 --> 00:15:22,510
And we wanted to do it in a way that that also doesn't over privilege any of the layers on the phone.

143
00:15:23,230 --> 00:15:30,980
Now come to the to the how we we did it. But just to set aside a little bit of a of of a scene,

144
00:15:31,000 --> 00:15:37,090
what we wanted to achieve is that we wanted to have secure domains that have direct access to screen and networks.

145
00:15:37,090 --> 00:15:44,740
So when I'm running in a secure domain, I want full screen access without having to trust Android, for example.

146
00:15:45,010 --> 00:15:55,240
So when I'm running in my domain, that peripheral is tied to my to talk to that code and cannot be modified or inspected by Android.

147
00:15:55,510 --> 00:16:01,630
The same with network, for example, or with any other sensor or any other peripheral that we want.

148
00:16:01,660 --> 00:16:09,460
Equally, we want that high, high availability. So meaning that if I if Android suddenly drops.

149
00:16:10,380 --> 00:16:19,140
That this other domain can take over the functionality and and connect to the network and continue running as, as if nothing happened.

150
00:16:19,560 --> 00:16:25,740
So we wanted to remove this dependency on these on these large operating systems while

151
00:16:26,100 --> 00:16:30,389
allowing them to still run on the platform fully isolated from the from the rest,

152
00:16:30,390 --> 00:16:36,180
because the same properties that we want for a secure domain. We also want for for the OEM code.

153
00:16:37,860 --> 00:16:43,200
And at the same time, we didn't want to introduce a huge layer below these,

154
00:16:43,200 --> 00:16:50,250
these systems and a large TCB because that wouldn't that wouldn't make sense to introduce yet another dependency.

155
00:16:52,590 --> 00:17:00,630
We also wanted to do it in a usable fashion, meaning that so how would you now even even interact between these these domains?

156
00:17:01,200 --> 00:17:08,750
Well, this is not a very complicated idea, but but it's it's, uh, it needed to be to be to be done.

157
00:17:08,760 --> 00:17:14,780
So a simple press of a button where when you're running your, your phone, you can just press a button.

158
00:17:14,790 --> 00:17:22,770
This button is, uh, so the, the firmware has been configured such that this button press cannot be interrupted by,

159
00:17:23,280 --> 00:17:26,430
uh, or modified by any of the of the running domains.

160
00:17:27,210 --> 00:17:30,180
You get into a mode where you can switch between different domains.

161
00:17:30,540 --> 00:17:35,560
So either you press a button and you keep on switching between domains or anything of that sort.

162
00:17:35,580 --> 00:17:43,799
So at a press of a button, you switch between, uh, let's say your Android and your secure, uh, domain running, for example,

163
00:17:43,800 --> 00:17:50,520
a signal messenger and wireless and, and then you can enable all kinds of interactions between these domains.

164
00:17:50,880 --> 00:18:01,470
For example, your Android could be receiving a notification that you have a message in a secure domain without presenting a message.

165
00:18:01,470 --> 00:18:05,250
And then you can press a button switch and see the message in a secure domain.

166
00:18:05,340 --> 00:18:10,770
You can allow interactions between these domains, but clearly according to any policy that you would that you would like.

167
00:18:14,870 --> 00:18:28,240
So of course, all this sounds very, very nice when you present this kind of a vision and, um, and then, um, clearly somehow it needs to be built.

168
00:18:28,250 --> 00:18:32,389
Or the question is, can you even build this on today's, um, um, platforms?

169
00:18:32,390 --> 00:18:40,820
And, and what I've learned over these 20 years as a, as a professor is that there is nothing that dedicated students cannot do.

170
00:18:41,450 --> 00:18:50,370
Um, so it took us only for only more than 40 years going through these different, uh, prototypes.

171
00:18:50,370 --> 00:18:56,660
So first, for some arm board that we that we worked on, there was a, uh, an emulator that we used.

172
00:18:56,900 --> 00:19:01,160
And then you see the first examples of, of phones.

173
00:19:01,190 --> 00:19:05,180
These were some older phone models that we, that we built.

174
00:19:05,780 --> 00:19:08,930
Uh, I mean, we didn't build a phones, but a firmware that we updated.

175
00:19:09,440 --> 00:19:15,920
And this was in summer 2024 when, when some hope that this can actually run on,

176
00:19:15,950 --> 00:19:23,000
on phones without without much, um, uh, let's say, uh, much disturbance to the ecosystem.

177
00:19:23,390 --> 00:19:30,290
And then finally in, in March this year, uh, a spinoff company from our,

178
00:19:30,650 --> 00:19:35,810
from our department presenting this prototype at the Mobile World Congress in Barcelona,

179
00:19:36,740 --> 00:19:42,530
where we, we showcased essentially that such a system can be, can be built.

180
00:19:43,490 --> 00:19:52,220
And so how does it work? Well, here's the latest prototype, which is from now from well, effectively from a week ago or so.

181
00:19:52,640 --> 00:20:02,690
So a prototype of a phone where you switch between Android to a Linux, uh, operating system running some web apps.

182
00:20:03,140 --> 00:20:06,890
Three Ma here that you see is, uh, is a sweet, secure messaging app.

183
00:20:06,900 --> 00:20:14,120
So we reported it there. So every time that you switch, you switch potentially to a locked, let's say lock screen.

184
00:20:14,840 --> 00:20:21,139
Um, now you switch back to Android and you can do that as many times as you, as you like.

185
00:20:21,140 --> 00:20:28,360
And you can run several of these of these domains. So that's that essentially tells you what we built.

186
00:20:28,360 --> 00:20:32,020
What can you do. But then the question remains is how?

187
00:20:32,770 --> 00:20:35,500
How did you actually build this. Right. I mean, what do you do?

188
00:20:36,280 --> 00:20:42,910
Well, all problems in computer science clearly can be can be solved by another level of indirection.

189
00:20:43,330 --> 00:20:50,950
This is clearly not me who said this? Apparently. Roger Needham, uh, and a few other people also also, uh, said this.

190
00:20:53,040 --> 00:21:00,149
So the first idea would be to to Virtualise you put a hypervisor underneath all these different domains,

191
00:21:00,150 --> 00:21:04,350
these operating systems, this hypervisor is going to virtualise access to peripherals.

192
00:21:04,350 --> 00:21:09,030
And there you go. This is exactly the path that we didn't want to take.

193
00:21:09,180 --> 00:21:13,170
The reason is that you will replace this control.

194
00:21:14,300 --> 00:21:17,450
That you wanted to, um, to give up.

195
00:21:17,750 --> 00:21:27,710
You didn't want to have within these operating systems with now a hypervisor which will again manage access to these, to these peripherals.

196
00:21:28,190 --> 00:21:31,700
So we did not want to virtualise, uh, peripherals.

197
00:21:31,700 --> 00:21:35,630
We didn't want to virtualise access to, uh, to hardware.

198
00:21:36,900 --> 00:21:52,170
So instead we wanted to design a lean and simple firmware which will simply connect or dedicate peripherals to different domains.

199
00:21:52,950 --> 00:21:59,459
And so it is the sort of simple approach to, to designing such a, such a system.

200
00:21:59,460 --> 00:22:06,870
It of course need to be tested because one wouldn't believe that it would actually work without breaking all kinds of functionalities on the phone,

201
00:22:07,230 --> 00:22:14,309
but it but it works. So we introduced the simple layer, and I wouldn't even call it a layer,

202
00:22:14,310 --> 00:22:21,600
because it's rather a simple piece of code that assigns resources to the operating system and domains.

203
00:22:22,260 --> 00:22:26,610
It simply says, you get these these calls at this time.

204
00:22:26,610 --> 00:22:30,240
At this point in time, when you run, you get exclusive access to the screen,

205
00:22:30,870 --> 00:22:37,319
you get this memory, you get access to GPUs, you get access to, uh, to something else.

206
00:22:37,320 --> 00:22:41,190
So some of these peripherals can be shared, some of them not.

207
00:22:41,190 --> 00:22:47,190
But generally the principle is that we assign instead of virtualise.

208
00:22:48,780 --> 00:22:50,670
An arm actually allows you to do that.

209
00:22:51,890 --> 00:23:00,560
So, uh, first thing that we that we did is if you look at this, this side here, you see that that hypervisors are gone.

210
00:23:00,560 --> 00:23:04,370
So virtualisation is not needed there anymore.

211
00:23:04,460 --> 00:23:07,580
Right. This is not something that we want that we want to do.

212
00:23:07,910 --> 00:23:13,400
And the only thing that we did is we extended the secure monitor which runs on El L3.

213
00:23:13,760 --> 00:23:23,209
So if you look at these, uh, these uh, levels, these privileged levels, uh, the L3 would be the, the highest privilege level, uh,

214
00:23:23,210 --> 00:23:34,840
layer here where which we extended with, let's say, 6000 lines of code, which can then enforce this functionality that I, that I just spoke.

215
00:23:34,980 --> 00:23:38,990
And these lines of code are fully transparent and open. Anyone can inspect them.

216
00:23:38,990 --> 00:23:44,120
There is no there are no secrets there that that you that one needs.

217
00:23:44,750 --> 00:23:51,170
So first thing that we are doing is domains run concurrently on separate call of course.

218
00:23:51,170 --> 00:23:58,940
So we are not pausing Android we are not or iOS by the way this works on on iOS as well is just

219
00:23:58,940 --> 00:24:04,430
that access to Apple devices is more difficult and then than working with Android devices.

220
00:24:05,900 --> 00:24:09,240
And so the and the secure monitor. Right.

221
00:24:09,290 --> 00:24:18,070
Also uses, um, other space controllers in order to assign memory to each to each of these, of these course.

222
00:24:18,080 --> 00:24:24,260
So it's kind of a coarse grained access control in terms of scheduling one domain.

223
00:24:24,860 --> 00:24:35,810
It can be any of these domains. It can be an independent domain can also be be in charge of of of scheduling these the the execution of these domains.

224
00:24:36,350 --> 00:24:44,510
This also can be configured by by the user. It depends on the on the setup, on the, on the phone and the properties that you that you want.

225
00:24:46,010 --> 00:24:53,180
In terms of peripherals, arm arm is also pretty good for this in the sense that your peripherals are memory mapped.

226
00:24:53,450 --> 00:24:58,579
And now you can essentially use again, uh,

227
00:24:58,580 --> 00:25:06,260
the address space controllers in order to grant domain specific access to, to um, to individual peripherals.

228
00:25:06,260 --> 00:25:14,780
So here on this on this figure, you can see that memory is just mapped so that, uh, GPS belongs to this, uh, domain one.

229
00:25:15,200 --> 00:25:21,170
And the screen domains domain two and domain three runs runs the network.

230
00:25:23,120 --> 00:25:31,459
So maybe this sounds a little bit too exclusive, but when we did the analysis of the peripherals and the workflows on the phone,

231
00:25:31,460 --> 00:25:39,710
it actually works pretty well in the sense that the screen, for example, is something that you need direct access to and exclusive access to.

232
00:25:40,070 --> 00:25:42,680
Right? Because when you switch from one domain to another,

233
00:25:42,680 --> 00:25:49,250
no other domain should see what you are doing or take input or see what you are, what you are presenting on the screen.

234
00:25:50,540 --> 00:25:56,930
You also, however, for a network, maybe you can have one domain that you trust for availability.

235
00:25:57,500 --> 00:26:01,399
So this domain can proxy the traffic for all the other domains.

236
00:26:01,400 --> 00:26:11,120
So you don't need to need to even, even um, worry let's say about about the network that much of your traffic is end to end encrypted for example.

237
00:26:11,270 --> 00:26:16,430
So this really depends on, on the peripheral and on the configuration that you, that you like.

238
00:26:19,420 --> 00:26:27,100
One thing that actually surprised us, and it's always it's always a bit of a, uh, over time, you'll refine what you're what you're doing is,

239
00:26:27,610 --> 00:26:35,500
is what turned out to be the biggest and the so far security issue that we, that we have identified is, uh, where interrupts.

240
00:26:36,610 --> 00:26:46,570
So, uh, there are a lot of peripherals clearly that work with, with interrupts and interrupt management through, uh, a general interrupt controller.

241
00:26:48,440 --> 00:26:55,510
Turned out to be the the biggest issue. So one of one of my collaborators, or one of the colleagues in on this project,

242
00:26:56,690 --> 00:27:03,410
has published a number of works showing how interrupts across different trusted execution environments and so on,

243
00:27:03,800 --> 00:27:07,590
can trigger attacks on those on those environments.

244
00:27:07,650 --> 00:27:13,880
Right. This you can see them as, as, uh, a type of micro architectural or side channel attacks.

245
00:27:14,180 --> 00:27:16,370
They're not really that's a different class of attacks,

246
00:27:16,700 --> 00:27:26,720
but essentially allowing domains to manipulate interrupts that they are not in charge of causing security, security issues.

247
00:27:27,560 --> 00:27:32,810
So one of one of the things that we need to to solve in this turn out to be quite interesting,

248
00:27:32,810 --> 00:27:41,060
interesting work was to to then introduce this, uh, GSC protection so that we call the GSC guard,

249
00:27:42,050 --> 00:27:52,310
uh, that we implemented there in order to make sure that only those domains that own those peripherals can also,

250
00:27:53,120 --> 00:27:57,259
uh, work with the interrupts, uh, from those from those peripherals.

251
00:27:57,260 --> 00:28:03,590
So essentially that this interrupt routeing is, is, is is protected that the owners are,

252
00:28:03,950 --> 00:28:11,299
are um properly, um, um properly restricted to only what they will down and how this is done.

253
00:28:11,300 --> 00:28:21,470
It's maybe a little bit too, too much of, uh, uh, to get into, but uh, luckily this GSC guard plus the,

254
00:28:21,470 --> 00:28:27,080
um, the way that the GSC can be configured with these secure non-secure,

255
00:28:27,590 --> 00:28:36,169
uh, uh, labelling of, uh, of interrupts and the awareness of the of the interrupt ID's of of individual, individual.

256
00:28:36,170 --> 00:28:43,460
Cause you can then you can then, um, isolate this properly and and protect your protect your system.

257
00:28:45,030 --> 00:28:50,760
So in a nutshell. We created a new a new architecture.

258
00:28:50,780 --> 00:28:59,210
I don't think the the core idea is not that new. Clearly assigning assigning peripherals exclusively to particular pieces of code.

259
00:28:59,330 --> 00:29:11,479
I don't think that's a terribly new idea in itself, but somehow going back or even I would say daring according to our given,

260
00:29:11,480 --> 00:29:17,870
given the feedback that we got from the reviewers to go back and revisit an almost the default

261
00:29:17,870 --> 00:29:23,209
idea that we should always virtualise all of our peripherals and and all of our resources,

262
00:29:23,210 --> 00:29:26,570
and then stack things on top of on, on top of each other.

263
00:29:27,260 --> 00:29:33,680
That that idea of virtualising actually didn't really solve the problem that we were trying to solve here,

264
00:29:33,680 --> 00:29:42,230
which is a problem of trying to, to, uh, satisfy together satisfy all the constraints of the users.

265
00:29:42,740 --> 00:29:51,049
OEMs align incentives and, and build a platform where we have a minimal or at least as,

266
00:29:51,050 --> 00:29:57,240
as reduced as possible level of trust in these common components that enforce isolation.

267
00:29:57,260 --> 00:30:00,740
In this case, this, uh, secure monitor running on on the R3.

268
00:30:01,550 --> 00:30:06,050
So we, we with a set of these measures like the, uh, eSIM.

269
00:30:06,050 --> 00:30:16,730
So secure monitor, extend it with less of less than 10,000 lines of code, uh, using these memory protection mechanisms, uh,

270
00:30:16,730 --> 00:30:22,840
to isolate peripherals and memory extending, uh, on GA cigar, uh, and GSC, um,

271
00:30:23,090 --> 00:30:28,640
well, essentially extended some this GSE guard reprogramming these these buttons.

272
00:30:29,060 --> 00:30:36,410
We were able to to create an ecosystem where anyone is able to essentially run whatever

273
00:30:36,410 --> 00:30:41,960
they want on their phone with full security isolated from their primary operating system.

274
00:30:43,610 --> 00:30:51,170
And I believe that if we can and we started doing this now, if we can push this to every phone,

275
00:30:51,680 --> 00:30:57,049
we can really open up, uh, open up development on each phone,

276
00:30:57,050 --> 00:31:05,360
because suddenly these domains get much closer to hardware and also enable more security and sovereignty on the phone,

277
00:31:05,360 --> 00:31:11,810
because now it's up to you to decide how much of the code you depend on for your critical applications.

278
00:31:13,160 --> 00:31:20,960
I'm clearly is not sitting sitting, um, let's say, um, I mean, they are clearly developing their platforms.

279
00:31:21,500 --> 00:31:26,809
So, um, if you're familiar with this, with this ecosystem, I see some faces.

280
00:31:26,810 --> 00:31:33,920
Here they are. Um, so, um, CA for example, is, uh, is arm's, uh, confidential computing.

281
00:31:34,430 --> 00:31:43,130
Uh, these are confidential computing extensions. Uh, we clearly then evolved also these design so that if in the future these can be deployed,

282
00:31:43,940 --> 00:31:51,800
we already have solutions that would, um, give us the properties that I talked about just on this, on this new platform as well.

283
00:31:54,360 --> 00:31:54,900
So yeah.

284
00:31:55,050 --> 00:32:06,210
So I'm really hoping that this will go somewhere and that, um, all of us will be able to have, um, more secure phones, more development freedom.

285
00:32:06,600 --> 00:32:14,520
And also we can we can give users, but also organisations more sovereignty over their critical workflows at the same time,

286
00:32:14,880 --> 00:32:18,570
um, preserving the benefits of this exciting ecosystem.

287
00:32:18,570 --> 00:32:27,299
And I, I have to say that the in terms of, um, the feedback that we got, especially at the Mobile World Congress and also in,

288
00:32:27,300 --> 00:32:32,760
uh, when talking to people, uh, I'm quite excited that we, we might be onto onto something.

289
00:32:32,760 --> 00:32:42,780
It's, uh, it's an interesting, uh, sort of academic and academic idea that, uh, that seems to be going, going now, uh, fully commercial.

290
00:32:44,800 --> 00:32:49,340
I am clearly not done with my talk. It's a little bit too early to to let you go yet.

291
00:32:49,940 --> 00:32:57,649
So this was the first part of my talk when I wanted to talk about something a little bit that touches on all of us,

292
00:32:57,650 --> 00:33:05,600
and it's a it's something that we all very often interact with and, and made a, made a case for how to make it better.

293
00:33:06,890 --> 00:33:15,980
But there is another topic that that worries me already for for maybe 20 years, but it seems to be more and more imminent, um, these days.

294
00:33:16,100 --> 00:33:23,780
So this is most likely I mean, clearly this is an AI generated image as you can as you can imagine,

295
00:33:23,930 --> 00:33:31,370
I asked I asked the AI to to generate a European city with some robots and drones in it.

296
00:33:31,370 --> 00:33:36,920
So this is this is how I imagined a European city with some robots and drones.

297
00:33:38,180 --> 00:33:40,760
So if you strip strip out all the technology,

298
00:33:40,970 --> 00:33:47,450
it might it might look really nice and and appealing if you put all the technology in that you see here, it might be,

299
00:33:47,450 --> 00:33:56,029
uh, it might be a little bit scary, maybe not the future that you want to live in, but it might be the future that that we build, we will experience.

300
00:33:56,030 --> 00:34:07,250
So a lot of self-driving cars, maybe humanoid robots, maybe drones flying around delivering things, uh, making our lives generally better.

301
00:34:08,870 --> 00:34:12,890
The question that you that I would ask myself, and maybe it's only me,

302
00:34:14,180 --> 00:34:19,280
is if I see a drone flying over my head, or if I see a robot passing next to me.

303
00:34:20,000 --> 00:34:24,080
My question is, is this robot supposed to be there? Right.

304
00:34:24,110 --> 00:34:26,690
Is it going to harm me? Is this a friend or foe?

305
00:34:27,890 --> 00:34:35,480
And this is a question that I would say most governments and most police forces these days are also thinking about.

306
00:34:36,470 --> 00:34:47,150
Are these devices supposed to be there? How do we register these devices like we we started with cars in a very, very, uh, let's say loose manner.

307
00:34:47,150 --> 00:34:53,720
And now we, we know exactly where these cars should drive with these robots and drones.

308
00:34:54,140 --> 00:34:58,640
We equally want to know where they are and are they supposed to be there.

309
00:34:58,640 --> 00:35:07,820
So we need to register them. They need to be able to securely and in a verifiable manner, uh, determine their location.

310
00:35:09,410 --> 00:35:15,620
So and then potentially reported to us or the infrastructure needs to be able to measure their, their location.

311
00:35:16,400 --> 00:35:26,060
So what we would actually want is some sort of spoofing resilience so that these devices that the locations of these devices can be securely,

312
00:35:26,840 --> 00:35:32,570
uh, calculated. We want to be able to verify the locations of these devices.

313
00:35:32,840 --> 00:35:37,970
And that's a slightly different property because it's about the claim, it's the location of a device that's trying to cheat.

314
00:35:39,580 --> 00:35:47,740
And we also want clearly increased resilience to interference because as we have seen, especially in the case of of Gnss.

315
00:35:49,480 --> 00:35:59,230
Janice. So our GPS, Galileo and so on are highly vulnerable to spoofing the different types of spoofing attacks and jamming attacks.

316
00:35:59,230 --> 00:36:06,730
And these is just one of the several maps of Europe where you can see at a particular date.

317
00:36:06,730 --> 00:36:14,350
This was in February this year. The locations where spoofing and jamming is is currently happening.

318
00:36:16,510 --> 00:36:26,020
And this is not very surprising. If you if you have been following the news, you might see this kind of news where drones on battlefields are.

319
00:36:26,470 --> 00:36:36,640
Instead of relying on on wireless communication, they are dragging wires behind them in order to be less susceptible to interference because

320
00:36:36,640 --> 00:36:41,470
of the of the of the interference based attacks on the functionality of these drones.

321
00:36:42,670 --> 00:36:49,120
Now, it's not only the devices that calculate their locations that are vulnerable to to these kind of attacks,

322
00:36:49,120 --> 00:37:00,970
but it's also the infrastructure as well. So if you read this carefully, you will see that we are already in an era where satellites, especially,

323
00:37:01,360 --> 00:37:10,209
uh, Gnss satellites, which are quite, uh, critical, are seen as a target and even a physical target.

324
00:37:10,210 --> 00:37:15,070
Right? I mean, they they're even facing a potential physical disruption.

325
00:37:17,170 --> 00:37:21,580
I mean, this is not very surprising, especially for for our group.

326
00:37:21,670 --> 00:37:24,760
Uh, I think this is a very, very old video, right.

327
00:37:25,360 --> 00:37:28,690
Uh, where we we were doing GPS spoofing.

328
00:37:29,270 --> 00:37:35,290
Not sure how to make this a bit. A bit faster, but I'll try.

329
00:37:36,010 --> 00:37:41,430
So. You will see GPU spoofing happening.

330
00:37:41,430 --> 00:37:44,190
So here this mobile phone is in is in Zurich.

331
00:37:44,640 --> 00:37:51,920
And back then what we did is we we spoofed the location from this, from this point where you saw it jump to another building in Zurich.

332
00:37:51,930 --> 00:37:58,670
Now it moved a little bit. Now it's going to move to another location if you see it now moving there.

333
00:37:58,700 --> 00:38:04,680
This was all done by spoofing that injected messages into the antenna of these of this phone,

334
00:38:05,040 --> 00:38:08,670
and therefore shifted the perceived location of this phone.

335
00:38:09,180 --> 00:38:12,270
And this doesn't only work in a small area.

336
00:38:12,600 --> 00:38:16,830
We now shifted the location to Berlin. Right. So this works pretty well.

337
00:38:17,580 --> 00:38:25,200
And clearly this this has been done with civilian GPS where authentication is, is um, doesn't really exist.

338
00:38:25,200 --> 00:38:30,600
So you can generate genuine GPS, um, authentic looking G.P.S. signals.

339
00:38:32,410 --> 00:38:36,549
This is not only a problem for mobile phones because as a as a user,

340
00:38:36,550 --> 00:38:42,160
you have corrective mechanisms and you can rely on other markers in order to figure out where you are.

341
00:38:42,640 --> 00:38:52,360
But this is clearly also an issue, uh, for Rob, for drones and other robots which rely on this spoofing, uh, on this, uh, location information.

342
00:38:52,360 --> 00:38:56,169
Right. So here you can see a, I mean, I'm not going to play a video,

343
00:38:56,170 --> 00:39:04,780
but this was done by my group of and and Jonathan in Northeastern, where they took a drone that was set to hover.

344
00:39:04,930 --> 00:39:08,140
And then by injecting, uh, spoofing signals,

345
00:39:08,290 --> 00:39:18,280
they managed to fully control the drone so they could essentially order the drone to go left or right by injecting appropriate GPS, uh, signals.

346
00:39:20,990 --> 00:39:26,540
So why is GPS um or Genesis more, more broadly, um, vulnerable?

347
00:39:26,780 --> 00:39:33,850
Well, because first of all distance. Right. It's easy to jam and overshadow and disrupt the signal.

348
00:39:33,920 --> 00:39:37,160
These are these are satellites that are 20,000km away.

349
00:39:37,730 --> 00:39:43,610
Uh, the the power of the of the signal on the surface of the Earth is very weak.

350
00:39:44,000 --> 00:39:48,620
And this you can't easily, easily fix in in with that constellation.

351
00:39:49,160 --> 00:39:57,440
Another thing is broadcast, meaning that these are broadcast only signals and they are susceptible to delays.

352
00:39:57,980 --> 00:40:05,600
And these selective delay attacks can cause a location to be shifted from one to another location,

353
00:40:06,320 --> 00:40:11,930
even if the if the signals are digitally signed or authenticated in some way, because, well,

354
00:40:11,930 --> 00:40:16,309
the adversary can just pick up those signals and shift them to another location,

355
00:40:16,310 --> 00:40:25,100
or shift them selectively and introduce delays and relays, which has been already shown in in in academic research, quite a bit.

356
00:40:26,180 --> 00:40:34,069
Another thing is that the types of waveforms and modulations and time of arrival mechanisms that were built for Gnss,

357
00:40:34,070 --> 00:40:44,210
at least for vanilla Gnss, they are not they're not really, um, fully resistant to the time of arrival manipulation, manipulation, attacks.

358
00:40:44,660 --> 00:40:54,450
And the protection measures do not really scale. So my my argument is that we need to rethink positioning.

359
00:40:55,290 --> 00:41:03,770
And we actually have a nice chance here. So. We are in an age of of Leo satellites being deployed heavily.

360
00:41:03,800 --> 00:41:07,130
They are much closer to the surface of the Earth 500km.

361
00:41:07,550 --> 00:41:12,260
We have great experience in running cellular networks, uh, at uh,

362
00:41:12,260 --> 00:41:19,639
at the massive scale so we can leverage these additional infrastructure abilities in order

363
00:41:19,640 --> 00:41:26,810
to secure those to secure positioning against both spoofing attacks and and interference.

364
00:41:27,170 --> 00:41:35,570
And how how will we do that? Well, these ingredients for those who who have been maybe following this research will not be very surprising.

365
00:41:35,900 --> 00:41:42,440
So we need to integrate two way ranging. So it turns out that the ability to to do two way ranging from roundtrip time

366
00:41:42,440 --> 00:41:48,320
of flight ranging can actually solve the problem of secure positioning and of,

367
00:41:48,410 --> 00:41:49,850
uh, location verification.

368
00:41:50,330 --> 00:41:59,090
And also a denser infrastructure gives us more opportunity to rethink our, uh, jamming resilience and, uh, time of arrival protection mechanisms.

369
00:42:00,660 --> 00:42:06,630
So what we are working on is to integrate terrestrial and satellite infrastructure.

370
00:42:07,710 --> 00:42:18,870
Uh, so meaning work, which typically, uh, Leo Leo constellations plus terrestrial, uh, infrastructure in order to solve, solve this problem.

371
00:42:19,890 --> 00:42:24,000
So the first thing that you have to think about in this space and, and this is,

372
00:42:24,000 --> 00:42:28,500
of course, a very high level, high level talks, I can't go into many details is that.

373
00:42:28,920 --> 00:42:33,780
So if you look at the broadcast systems like DNS and also like some of the

374
00:42:33,780 --> 00:42:37,500
some of the Leo constellations that are simply broadcasting for positioning,

375
00:42:37,860 --> 00:42:46,290
you can't really prevent spoofing. So that's really not a solution because of these selective delay attacks that I mentioned.

376
00:42:46,590 --> 00:42:48,060
But what you can do instead,

377
00:42:49,170 --> 00:42:58,440
you can do around the time of flight distance measurements that you can then in addition protect against distance reduction attacks.

378
00:42:58,620 --> 00:43:04,409
So if you can prevent distance reduction attacks, meaning that an adversary cannot reduce the distance,

379
00:43:04,410 --> 00:43:09,600
but only large, he just happens that within these this kind of geometry, like a,

380
00:43:09,840 --> 00:43:15,480
like a this pyramid here, if you do this from, from several, uh, vantage points,

381
00:43:16,710 --> 00:43:24,000
you can have a guaranteed provable, uh, secure positioning system and even location verification system.

382
00:43:25,690 --> 00:43:30,100
So and this is a result that's quite old. It's from 2004 2005.

383
00:43:31,270 --> 00:43:39,610
Recently a group of people are sitting at this point, and, um, our students want us.

384
00:43:39,610 --> 00:43:45,010
We're are extending this kind of an an idea with what they call trick.

385
00:43:46,380 --> 00:43:56,000
Where they are combining essentially broadcast with a response to so a broadcast from one once one of the systems.

386
00:43:56,010 --> 00:44:06,090
So for example NSE or Leo and then a respond a response to a two terrestrial system forming instead of these circles that constrain the geometry,

387
00:44:06,360 --> 00:44:15,900
they they create ellipses in order to and define a region within which spoofing is not is not possible.

388
00:44:16,800 --> 00:44:20,040
Right and into which spoofing is not possible either.

389
00:44:21,030 --> 00:44:26,069
So this is this seems like relatively, relatively simple geometric thinking,

390
00:44:26,070 --> 00:44:30,510
but there's actually quite interesting, interesting academic work behind it.

391
00:44:32,660 --> 00:44:39,080
So. But in order for these geometries to work and give us secure positioning, we need we need something else.

392
00:44:40,160 --> 00:44:46,220
We need to be able to protect these two ways ranging against distance reduction attacks.

393
00:44:47,060 --> 00:44:54,650
So if you have a satellite and, uh, and a drone or two drones running challenge response protocols that are authenticated,

394
00:44:54,920 --> 00:45:01,700
one would believe that these are sufficient to prevent distance reduction attacks, but they're really not.

395
00:45:01,850 --> 00:45:14,120
So we know from from the research literature that that specially crafted signals injected by the adversary into these two devices,

396
00:45:14,120 --> 00:45:20,550
measuring a distance, uh, can actually shorten the distance that is measured between these two devices.

397
00:45:20,570 --> 00:45:27,330
If we if an adversary can do that, there is practically no help in in building a secure positioning system.

398
00:45:27,350 --> 00:45:31,810
So we need to solve this problem. So how can this happen?

399
00:45:34,780 --> 00:45:43,600
So if you if you look at a bolt here, uh, a pulse which is wirelessly transmitted over a wireless channel when it.

400
00:45:44,860 --> 00:45:49,380
When it arrives at the receiver, it's going to be somewhat smudged, right?

401
00:45:49,390 --> 00:45:56,620
So it's going to be spread by the by this channel because these wireless channels are quite complex.

402
00:45:57,610 --> 00:46:01,330
And not only that, it's going to be spread, but counter-intuitively,

403
00:46:02,320 --> 00:46:13,120
the the time of arrival of this smudged pulse will not be the strongest peak that you see here, but it might be the one that's that's before four.

404
00:46:13,390 --> 00:46:16,450
You have to essentially determine which one is it,

405
00:46:16,600 --> 00:46:23,140
and that that's where a lot of crafting might go in terms of building these kind of arranging systems,

406
00:46:23,470 --> 00:46:32,650
which one of these earlier peaks that come before the the strongest peak is actually the one that determines the true the true distance.

407
00:46:33,790 --> 00:46:40,060
And by injecting signals in such a way that you can create this even earlier peak.

408
00:46:41,620 --> 00:46:47,820
You will be able to shorten the distance. So that's intuitively how this works.

409
00:46:47,840 --> 00:46:52,460
Not not very precise, but intuitively how some of these attacks work.

410
00:46:52,850 --> 00:46:58,880
And this was one of the attacks that we that we did on, on, on an ultra wideband ranging system where we,

411
00:46:59,090 --> 00:47:04,310
we paired, we looked at the, the ranging between an an NXP chip and an iPhone.

412
00:47:04,700 --> 00:47:13,730
So on an NXP ultra wideband chip and iPhone where we when we turned our spoofing, we were able to bring a distance that was measured at 8m to 0m.

413
00:47:16,600 --> 00:47:19,640
So this is clearly. And how did we do that?

414
00:47:19,670 --> 00:47:27,040
Well, we actually didn't know any of the, of the keys or, or random sequences that were exchanged between these two devices.

415
00:47:27,520 --> 00:47:35,799
And we injected what was essentially another random sequence into the on top of the packets that were being exchanged by these two devices.

416
00:47:35,800 --> 00:47:46,150
And we managed to cause those early peaks to appear in the in the receiver and therefore trigger a shorter, a shorter distance measurement.

417
00:47:46,300 --> 00:47:53,880
I mean, the measurement of a shorter distance. He's actually a very, very fresh out of the print result.

418
00:47:54,660 --> 00:48:02,020
Um. Where we have recently shown it's a very simple kind of kind of result.

419
00:48:02,320 --> 00:48:10,960
Again, very counter-intuitive. So if you see here on the left, there are you can use in a narrow band, uh, system.

420
00:48:12,140 --> 00:48:15,590
You can actually use a negative group delay.

421
00:48:16,040 --> 00:48:19,310
Um, filter that can.

422
00:48:19,760 --> 00:48:32,180
I'm not sure if I have a. Yeah. So here this blue this blue signal is the, um, is the signal that should come out of the that should arrive at the,

423
00:48:32,180 --> 00:48:36,020
at the receiver if we didn't introduce our attack.

424
00:48:36,020 --> 00:48:43,400
Right. So instead what we did is we created we, we implemented a filter that created this red.

425
00:48:44,640 --> 00:48:49,290
Red curve, but it seems as if it has arrived earlier.

426
00:48:49,400 --> 00:48:57,000
Right. So if you look at this envelope here, and if you measure the time of arrival of that envelope, you will measure an earlier time of arrival.

427
00:48:57,750 --> 00:49:03,299
And how did we do that? Well, this filter essentially, uh, aggressively amplifies these,

428
00:49:03,300 --> 00:49:10,530
these amplitudes here and then suppresses them later in order to create this shape without knowing what's coming,

429
00:49:10,530 --> 00:49:13,700
without knowing which signal, which which signal, these come.

430
00:49:13,710 --> 00:49:20,100
So it's a very, very elegant way to reduce a measured a measure distance by manipulating,

431
00:49:20,550 --> 00:49:29,660
uh, by manipulating, using in this case, something that wrote you. Another attack that that comes from, uh, already from 2006.

432
00:49:31,400 --> 00:49:40,250
Is what's called an early detect, late commit attack. Uh, in this case, I illustrated on ultra wideband, right where you have a drone sending some,

433
00:49:40,790 --> 00:49:45,860
some pulses to a drone, and they are supposed to arrive as indicated here.

434
00:49:46,040 --> 00:49:52,280
The attacker wants to transmit those pulses earlier, but doesn't know what they are in advance.

435
00:49:52,730 --> 00:50:02,510
Cannot guess them. Right? So they are completely random. So what the attacker can do instead the attacker can send something in advance.

436
00:50:04,370 --> 00:50:08,960
Some random random garbage within the slots that he wants.

437
00:50:09,860 --> 00:50:17,470
And then when he learns. A part of the signal that that, um, that the sender transmitted.

438
00:50:17,950 --> 00:50:23,410
He sort of he generates the remaining part of this, of this signal.

439
00:50:25,480 --> 00:50:29,290
And now some of the time of arrival systems will react in the following way.

440
00:50:29,320 --> 00:50:34,750
They will say, hey, you know, maybe half of the of this symbol is messed up,

441
00:50:34,750 --> 00:50:41,260
but the other half shows shows enough evidence that this is a correct, that this is a correct symbol.

442
00:50:42,730 --> 00:50:48,350
And in and when summarised. I think this is a fine symbol.

443
00:50:49,810 --> 00:50:56,710
And of course, this kind of behaviour that you see here, it's a bit exaggerated but can happen on the real wireless wireless channel.

444
00:50:57,580 --> 00:51:01,720
So you end up advancing again and shortening a distance.

445
00:51:04,790 --> 00:51:08,840
So how can one protect against these physically? Uh, attacks.

446
00:51:09,770 --> 00:51:15,470
So basic protections, like a secret spreading codes can help?

447
00:51:15,680 --> 00:51:20,060
Not entirely, but they can. They can assist there. There is a lot of work now.

448
00:51:20,090 --> 00:51:23,959
There was a work work from Apple, for example, in the ultra wideband space,

449
00:51:23,960 --> 00:51:30,170
showing how effectively using spreading spreading codes across different sequences

450
00:51:30,470 --> 00:51:35,270
can potentially mitigate some of these attacks in the context of ultra wideband GPS.

451
00:51:35,810 --> 00:51:42,799
Uh is introducing camera, which are physical layer markers, um, which are some sort of punctured codes,

452
00:51:42,800 --> 00:51:47,600
right, in order to, to detect on the physical layer whether weather signals have been manipulated.

453
00:51:48,200 --> 00:51:58,610
Uh, in ultra Wideband, we have we have done a lot of work securing that even mentioned resulting in a spinoff company and product in up, uh, in Wi-Fi.

454
00:51:58,820 --> 00:52:05,360
We have we have mechanisms for that that are being now introduced in standards, the same in Bluetooth.

455
00:52:06,470 --> 00:52:11,930
And well, that's something that I already mentioned how this could be and this could be done.

456
00:52:13,700 --> 00:52:23,629
So in a nutshell, we have different technologies, geometries and ranges that we could potentially use to derive a position of a device.

457
00:52:23,630 --> 00:52:29,870
But at the at the core of it for each of these technologies is how do we protect

458
00:52:29,870 --> 00:52:35,899
the time of arrival in order to avoid this kind of time of arrival manipulations,

459
00:52:35,900 --> 00:52:41,030
because at least against distance reduction, because if we can prevent distance reductions,

460
00:52:41,810 --> 00:52:46,130
distance enlargements will be prevented by the geometry itself.

461
00:52:46,940 --> 00:52:48,680
So we need to focus on that.

462
00:52:50,420 --> 00:52:58,460
And if you look at different technologies that are being used for ranging, uh, you can of course pick on which one you want to work.

463
00:52:58,640 --> 00:53:03,380
We worked already on Ultra wideband, on Bluetooth, uh, on 5G,

464
00:53:04,280 --> 00:53:11,780
and most recently we introduced a new a new system that we built specifically for, uh, space.

465
00:53:11,780 --> 00:53:19,759
So for arranging with, uh, Leo satellites, where we then focussed on all of the M, uh, modulation because that's, uh, modulation.

466
00:53:19,760 --> 00:53:30,210
That's, that's being used in, in a lot of communication systems, um, including with Leo and, and of course, this is now an overloaded overload,

467
00:53:30,260 --> 00:53:37,590
this slide of all the things that one needs to do in this space in order to in order to make this work right from the start,

468
00:53:38,000 --> 00:53:44,659
designing the waveforms that you want to use to design the packets and making sure that, uh,

469
00:53:44,660 --> 00:53:49,310
that you have a time of arrival verification that secure and provably secure and so on.

470
00:53:49,670 --> 00:53:55,460
I'm going to give you an intuition behind behind, um, those those results.

471
00:53:55,940 --> 00:54:01,690
So this is the intuition. No. So what are we doing?

472
00:54:01,690 --> 00:54:06,850
For those of you who don't? Who are familiar with all of the modulation or not?

473
00:54:07,390 --> 00:54:10,870
I mean, it's essentially a frequency kind of embedding.

474
00:54:11,770 --> 00:54:22,660
And with an inverse Fourier transform, you get, uh, you get the, um, the time, the time domain signal, then you, then you transmit and.

475
00:54:23,650 --> 00:54:28,450
What did we actually do? We took two consecutive symbols in the OFDM.

476
00:54:29,050 --> 00:54:32,620
Uh, in the URL of the a manipulation encoded in the in the frequency domain.

477
00:54:33,280 --> 00:54:35,859
And we created, uh, a verification,

478
00:54:35,860 --> 00:54:44,020
a verification function that is going to test whether the wireless channel is consistent across those two, those two symbols.

479
00:54:44,410 --> 00:54:54,520
And what does this on on a high level test. This test whether the the distortion of the wireless channel is plausible or not.

480
00:54:55,420 --> 00:55:00,520
So if it's plot if it's if it's inconsistent across these two consecutive symbols,

481
00:55:00,790 --> 00:55:06,579
this means that most likely there is there is some sort of an intervention by by

482
00:55:06,580 --> 00:55:11,830
an adversary and not an expected and expected behaviour of the of the channel.

483
00:55:15,310 --> 00:55:19,120
So, um, yeah.

484
00:55:20,230 --> 00:55:25,219
And this in the, in the, um, in this constellation diagram looks looks like this.

485
00:55:25,220 --> 00:55:31,410
So the. Typically when you are when you decode these symbols,

486
00:55:32,040 --> 00:55:38,579
what you expect is that due to the channel changing the phase and the amplitude of your of your arriving signals,

487
00:55:38,580 --> 00:55:41,480
you will need to do some form of a of a compensation.

488
00:55:41,490 --> 00:55:50,100
So if you if you can test that this compensation is inconsistent across two symbols, you can let's say build on top of that.

489
00:55:50,100 --> 00:55:55,770
And, and then um, uh, have your time of arrival verification function.

490
00:55:55,770 --> 00:56:00,990
And for that, of course, we produced the proof that is in our, in our research research work.

491
00:56:02,460 --> 00:56:08,310
But of course, you know, security, we need to, uh, argue through proofs that we, that we write in papers.

492
00:56:08,310 --> 00:56:12,660
But implementation wise, we wanted to see if such ideas could actually work.

493
00:56:13,170 --> 00:56:17,879
And for that we have built first then in lab in lab setup.

494
00:56:17,880 --> 00:56:25,800
Right. This large box here is essentially emulates a wireless signal between the ground and and the satellites.

495
00:56:25,800 --> 00:56:31,890
So so this is a full a full blown, uh, satellite link simulator, right,

496
00:56:31,890 --> 00:56:36,990
where you can plug in your, your signals from the ground and from the satellite and back,

497
00:56:36,990 --> 00:56:42,930
and then test alongside with all the constellations and all the Doppler shifts and everything that you have to cope with.

498
00:56:43,380 --> 00:56:52,260
Right, in order to to test whether this type of modulation and this type of verification function can actually survive such channels,

499
00:56:53,040 --> 00:56:56,340
and we have shown that functionally it can. Right.

500
00:56:56,730 --> 00:57:01,140
We were not very happy with this because we said we we need to go out and actually test this.

501
00:57:01,860 --> 00:57:09,960
And um, and then we went to, uh, to a tunnel because transmitting outside is not so easy if you,

502
00:57:10,410 --> 00:57:14,880
if you are, um, getting licenses to transmit in certain bands, it's not so obvious.

503
00:57:15,270 --> 00:57:20,009
A tunnel clearly is not a channel that's, that's representative of a satellite channel.

504
00:57:20,010 --> 00:57:28,770
You don't have a tunnel to the up in the sky, but we wanted to functionally test across 500m where there are actually set up works.

505
00:57:29,880 --> 00:57:38,880
And so what you can see at one end of the tunnel, there was uh, there is a, there's one, uh, transceiver, another one here on this cart.

506
00:57:39,540 --> 00:57:43,590
And, uh, and what you don't see students pushing this cart down the tunnel.

507
00:57:43,980 --> 00:57:52,680
And as a result, you can you can see this distance actually changing, being being reduced as a result of this, of this approach.

508
00:57:53,130 --> 00:57:57,510
And then when they pull back you will see the distance, the distance increase.

509
00:57:57,750 --> 00:58:01,470
Increase back. Um back to the.

510
00:58:01,860 --> 00:58:08,830
Yeah. Mhm. And of course we are not happy with that either.

511
00:58:09,460 --> 00:58:13,240
So we did the test in the lab.

512
00:58:14,320 --> 00:58:19,030
We did the uh terrestrial tests and we will do more open space tests.

513
00:58:19,870 --> 00:58:28,420
But we are also happy that ESA uh, will allow us to deploy our, our design within the, um, the cyber Cube.

514
00:58:28,510 --> 00:58:30,610
So there was a competition, and there will be.

515
00:58:30,670 --> 00:58:37,480
And a European Space Agency launched satellite that will implement several projects, including, uh, our own,

516
00:58:37,990 --> 00:58:43,180
uh, where we will then perform our experiments from the satellites down to the down to the ground.

517
00:58:46,020 --> 00:58:53,159
So in summary, I believe that we need to change the way that we that we work on,

518
00:58:53,160 --> 00:59:00,569
on these positioning systems in the sense that we need to change the the reference that only broadcast the broadcast,

519
00:59:00,570 --> 00:59:05,969
the only systems are sufficient, that we don't need terrestrial infrastructures to be built in our cities.

520
00:59:05,970 --> 00:59:09,780
I think we do. I think we will need more robustness in this space.

521
00:59:09,960 --> 00:59:18,210
Otherwise we will see a lot more attacks that involve spoofing of locations, duping on on own locations, by by devices.

522
00:59:18,540 --> 00:59:26,280
So we need to build. We need to be ready build fundamentals for, for for this uh, for this particular problem space.

523
00:59:27,150 --> 00:59:31,230
And if we can do that, maybe we'll have a more secure future.

524
00:59:32,160 --> 00:59:37,080
Um, of course, it's easy to be pessimistic about that, right?

525
00:59:37,560 --> 00:59:44,969
Most likely, how these things develop is that you first need attacks and problems to happen and then deploy these, these kind of solutions.

526
00:59:44,970 --> 00:59:54,780
But let's see, maybe we we get ahead of a curve. Uh, there is some hope because there are already a number of calls by, by relevant European,

527
00:59:54,930 --> 01:00:00,240
uh, and global entities calling for secure positioning to be, to be deployed.

528
01:00:00,540 --> 01:00:02,090
And also these, uh,

529
01:00:02,100 --> 01:00:11,400
the effects of the of the recent wars where spoofing and jamming has been quite prevalent are showing how relevant this topic is, is becoming.

530
01:00:13,880 --> 01:00:22,480
Uh, so final final musings for my from my side when it comes to generally system design and and

531
01:00:22,490 --> 01:00:30,990
security research is that I think we we can be really bold in hardening our infrastructures.

532
01:00:31,070 --> 01:00:36,800
I think it doesn't require us to be to, to design everything from scratch.

533
01:00:36,830 --> 01:00:46,040
I think we can integrate technologies. Uh, our choice of OFDM is it's not only a choice to go for OFDM,

534
01:00:46,040 --> 01:00:52,339
but we are clearly looking for integration within existing 5G and other other

535
01:00:52,340 --> 01:00:58,600
technologies to towards and to effectively simplify the implementation of of such um,

536
01:00:58,610 --> 01:01:05,360
protection measures into existing, uh, ecosystems in the same way for the sovereign smartphone.

537
01:01:05,870 --> 01:01:09,830
What we want wanted to do is to avoid the trap of of being exclusive.

538
01:01:09,860 --> 01:01:14,389
We wanted to be on a platform that's actually open to to everyone,

539
01:01:14,390 --> 01:01:21,230
including large ecosystems, because this is not an, uh, this is not a fight against someone.

540
01:01:21,230 --> 01:01:29,420
It's a it's a, it's a fight to have systems that we can actually control, either as a user or as as as as communities.

541
01:01:32,330 --> 01:01:36,980
And so of course, of course, that I did all of this alone.

542
01:01:37,730 --> 01:01:42,680
Um, and, uh, so these are some, some of the people on, uh,

543
01:01:42,680 --> 01:01:47,810
that I want to acknowledge that work quite hard on some of these, uh, projects that you have seen.

544
01:01:48,230 --> 01:01:50,210
And with this, I, I thank you.