1 00:00:00,200 --> 00:00:04,200 We're discussing Ai and cybersecurity with one of the 2 00:00:04,200 --> 00:00:08,000 foremost AI researchers in the world McCall. 3 00:00:08,000 --> 00:00:11,500 Pacu check is the CTO of Jen. Yeah. 4 00:00:11,500 --> 00:00:14,600 Thank you know I'm a special breed in a way you know I've 5 00:00:14,600 --> 00:00:18,600 been an AI computer scientist for a kind of big part of my 6 00:00:18,600 --> 00:00:23,600 life. I've been a professor went 7 00:00:23,600 --> 00:00:28,400 through all these postdocs, sabbaticals, getting grounds 8 00:00:29,100 --> 00:00:33,800 advising team. 80 students. All this Gizmo and I always was 9 00:00:34,000 --> 00:00:39,000 super excited about what a I can have as a positive impact to 10 00:00:39,000 --> 00:00:41,000 society. So I was going to one of those 11 00:00:41,000 --> 00:00:47,000 crab occasionally in fact researchers and can back in 12 00:00:47,100 --> 00:00:51,000 2005. We started couple of research 13 00:00:51,000 --> 00:00:55,700 projects with my PhD students in the use of AI in the field of 14 00:00:55,700 --> 00:01:01,500 cyber security at a time. You know, the AIO Is only coming 15 00:01:01,500 --> 00:01:05,400 up as an application area and people were using AI for image 16 00:01:05,500 --> 00:01:08,100 analytics. And the video analytics we just 17 00:01:08,100 --> 00:01:11,400 wanted to make a breakthrough in the field over how Ai and 18 00:01:11,400 --> 00:01:13,400 machine learning can be used in cybersecurity. 19 00:01:13,800 --> 00:01:19,400 And believe me those days, it was very difficult to sell AI am 20 00:01:19,400 --> 00:01:22,500 L to cyber security Specialists online today. 21 00:01:22,600 --> 00:01:27,700 Do they say, I is driving all the cyber security analytics in 22 00:01:27,700 --> 00:01:29,800 the majority of the systems that we use. 23 00:01:29,900 --> 00:01:34,800 Use these days. And but you're after a couple of 24 00:01:34,800 --> 00:01:39,900 startups and to working with the VCS, I was kind of invited that 25 00:01:39,900 --> 00:01:45,200 asked, by Andrea Bocelli, who is a former CTO of almost then a 26 00:01:45,200 --> 00:01:49,300 CEO. We was being promoted and he, he 27 00:01:49,300 --> 00:01:52,500 kept talking to me and kind of getting me, slowly excited about 28 00:01:52,900 --> 00:01:59,800 how fun it would be to use a I not for B2B cybersecurity for 29 00:02:00,000 --> 00:02:04,400 Enterprise sector but try to use my creativity and experience for 30 00:02:04,400 --> 00:02:06,700 building cybersecurity for consumers. 31 00:02:07,500 --> 00:02:11,600 Ai and they both are basically for consumers and he planted 32 00:02:11,600 --> 00:02:16,300 this bark. And here I am in the gym from a 33 00:02:16,300 --> 00:02:22,300 vast Building Systems and running the R&D departments and 34 00:02:22,300 --> 00:02:26,100 research labs and threat Labs. So that we build the best in 35 00:02:26,100 --> 00:02:29,800 class cyber security for consumers. 36 00:02:30,300 --> 00:02:32,300 Consumers are basically, this is what I do. 37 00:02:34,000 --> 00:02:39,300 Do you want to give us maybe a brief overview of the 38 00:02:39,300 --> 00:02:44,400 distinction between cyber security for the Enterprise and 39 00:02:44,400 --> 00:02:51,800 cybersecurity for consumers? This one is extremely exciting, 40 00:02:52,200 --> 00:02:57,100 especially these days when the cybersecurity as a field, is 41 00:02:57,100 --> 00:03:01,200 undergoing a major change because you're in the the past 42 00:03:01,800 --> 00:03:06,400 attackers and the other Us. 30 years, actually at the girls 43 00:03:06,400 --> 00:03:10,100 were writing Malabar, and we'll start get it at computer systems 44 00:03:10,100 --> 00:03:17,400 at the networks and the programs operating systems that people 45 00:03:17,400 --> 00:03:22,700 used it was it was the duty of the Enterprise of the industry, 46 00:03:22,700 --> 00:03:26,100 the business to make sure that there is of them, death cell 47 00:03:26,300 --> 00:03:28,400 every Hardware, they sell every net with it. 48 00:03:28,400 --> 00:03:31,700 Is there is a safe. Okay. 49 00:03:32,100 --> 00:03:35,300 So and then users. Consumers were pretty much users 50 00:03:35,300 --> 00:03:38,500 of this infrastructure. That the Enterprise sector have 51 00:03:38,500 --> 00:03:43,000 made to be safe with the with the recent change that we see in 52 00:03:43,000 --> 00:03:48,800 the industry were, it's not any longer, the vulnerability of the 53 00:03:48,800 --> 00:03:51,900 operating systems, and the networks and the computers on 54 00:03:51,900 --> 00:03:56,300 our devices, by it's more people that are the vulnerability. 55 00:03:56,300 --> 00:03:59,900 And the supply chain, people are not only victims of cyber 56 00:03:59,900 --> 00:04:05,000 security attacks, but people are also a conduit and Pieces on the 57 00:04:05,000 --> 00:04:09,400 supply chain that not only consume but also participate in 58 00:04:09,400 --> 00:04:13,800 deploying an attack. So people are getting in the 59 00:04:13,800 --> 00:04:18,100 front People Are People cognition and the way how people 60 00:04:18,100 --> 00:04:22,200 think and consume the internet is becoming the vulnerability 61 00:04:22,800 --> 00:04:27,000 because of this major change there is now a lot more interest 62 00:04:27,300 --> 00:04:29,500 in. Consumers are basic rate were 63 00:04:29,700 --> 00:04:35,000 the expectation of the industry is to build Old technologies 64 00:04:35,200 --> 00:04:39,800 that will be there covering people at the very end or when 65 00:04:39,800 --> 00:04:42,100 they touch the internet. And this is a very, and as 66 00:04:42,100 --> 00:04:47,300 recognition is what they see what they read the messages, 67 00:04:47,300 --> 00:04:50,500 they receive the emotions they post. 68 00:04:50,900 --> 00:04:54,200 So the consumer service that created these days needs to be 69 00:04:54,200 --> 00:04:57,000 at this very edge of the internet, which is different 70 00:04:57,000 --> 00:05:00,700 Edge. That was exciting than 50 years 71 00:05:00,700 --> 00:05:02,600 ago. And I'm not only going to 72 00:05:02,600 --> 00:05:08,200 talking about about my My what I think but your we have data, you 73 00:05:08,207 --> 00:05:16,900 know, Jen as a company is, is a technology company that is a 74 00:05:16,900 --> 00:05:21,900 house of a number of Technologic brands in the field of cyber 75 00:05:21,900 --> 00:05:25,000 security. I originally come from a vast, 76 00:05:25,300 --> 00:05:29,900 which is, which used to be the biggest European cyber security 77 00:05:29,900 --> 00:05:33,600 brand that merged last year was not LifeLock, which was the 78 00:05:33,700 --> 00:05:38,500 biggest consumer services with the brand and in others. 79 00:05:39,600 --> 00:05:43,200 We because of our freemium offering, you know, animals. 80 00:05:43,200 --> 00:05:46,000 We were going to do the first and the biggest and the premium 81 00:05:46,000 --> 00:05:51,000 and consumer cyber security. We see close to half a billion 82 00:05:51,000 --> 00:05:53,300 and points. So we see a huge part of the 83 00:05:53,300 --> 00:05:58,100 internet and in this is visibility in gives us a data 84 00:05:58,100 --> 00:06:02,100 where we see that. Currently, it's only 30% of the 85 00:06:02,100 --> 00:06:03,600 attacks that we see on the internet. 86 00:06:03,800 --> 00:06:06,500 And that are caused by a classical model board is 87 00:06:06,500 --> 00:06:12,700 targeting devices, and network infrastructure was 70% of all 88 00:06:12,700 --> 00:06:18,100 what we see are attacks like fishing and scamming attacks on 89 00:06:18,100 --> 00:06:23,600 human cognition. Give in that, weird is a I come 90 00:06:23,600 --> 00:06:30,500 into play to help prevent this aspect of the cybersecurity 91 00:06:30,500 --> 00:06:34,000 supply chain. As you described it, we see See 92 00:06:34,000 --> 00:06:38,500 attackers optimizing their costs and mocked and maximizing the 93 00:06:38,500 --> 00:06:40,500 effect of deployment of the attacks. 94 00:06:41,200 --> 00:06:44,400 And they were using a different automation techniques and 95 00:06:44,400 --> 00:06:46,800 methodologies, including artificial intelligence. 96 00:06:46,800 --> 00:06:49,800 Different kinds, not only emotional learning, but also 97 00:06:50,300 --> 00:06:53,300 automated planning, automated reasoning different types of a 98 00:06:53,300 --> 00:06:58,500 I4 G making detects as cheap as possible and as a large-scale 99 00:06:58,600 --> 00:07:02,300 deployment as possible, so there was a lots of AI under the hood 100 00:07:02,700 --> 00:07:06,900 or more Writing more attacks mobile deployment, in order to 101 00:07:06,900 --> 00:07:09,500 be able to respond and to protect our consumers, 102 00:07:09,500 --> 00:07:14,200 efficiently, the cybersecurity firms needed to deploy 103 00:07:14,200 --> 00:07:22,700 high-grade, AI to protect the consumers because as soon as you 104 00:07:22,700 --> 00:07:27,500 stop deploying the right level of automation against automated 105 00:07:27,500 --> 00:07:30,800 attacks, you would be losing your Warfare. 106 00:07:30,800 --> 00:07:38,200 There is no way how by even analysts you can defend the 107 00:07:38,200 --> 00:07:41,200 automated High scale attacks that are coming from the 108 00:07:41,200 --> 00:07:45,500 attackers. So, for this reason, AI has been 109 00:07:45,800 --> 00:07:50,500 very well designed to be used on the side of the the defense. 110 00:07:51,100 --> 00:07:55,700 But there have been a number of problems that we've experienced 111 00:07:55,800 --> 00:08:00,500 as the Defenders. One of those were that for each 112 00:08:00,500 --> 00:08:03,500 different type of attack that we see on the internet. 113 00:08:04,400 --> 00:08:08,700 We're gonna need it to build a new AI detector, new AI, 114 00:08:08,700 --> 00:08:13,200 classifier, we cybersecurity experts were building the 115 00:08:13,200 --> 00:08:17,000 classifiers by designing features and doing lots of 116 00:08:17,200 --> 00:08:22,000 training. And soon, we started to see that 117 00:08:22,300 --> 00:08:26,800 this doesn't scale because we need a lot of programmers, and 118 00:08:26,800 --> 00:08:31,500 cyber and subject matter experts who can help us to design those 119 00:08:31,500 --> 00:08:33,500 algorithm those algorithms do so much. 120 00:08:33,600 --> 00:08:36,799 E-learning tools. You're talking now about 121 00:08:36,799 --> 00:08:39,600 signature-based. Yes, I'm talking about the 122 00:08:39,600 --> 00:08:42,200 signature bass time over detection, exact. 123 00:08:42,200 --> 00:08:45,800 Okay. And in order to be able to kind 124 00:08:45,800 --> 00:08:49,900 of cope with this scalable scalability problem, explosion 125 00:08:49,900 --> 00:08:53,700 of the types of malware that that we are seeing online. 126 00:08:54,500 --> 00:08:58,500 We needed to deploy a deep learning kind of my thoughts. 127 00:08:58,600 --> 00:09:03,000 That makes the programmers and the cybersecurity on the list 128 00:09:03,000 --> 00:09:05,800 free. Of Designing the features. 129 00:09:06,000 --> 00:09:08,000 So there is one type of algorithm. 130 00:09:08,800 --> 00:09:12,700 If we're all design that you train on different types of 131 00:09:12,700 --> 00:09:18,200 data, large data, and classified the detector exhibits. 132 00:09:18,300 --> 00:09:23,900 More General detective detection capability and can be used 133 00:09:23,900 --> 00:09:28,900 across the pipeline and the, the cyber security company and, 134 00:09:29,100 --> 00:09:33,500 like, in a vast, what we did is we've build those a unique 135 00:09:33,600 --> 00:09:39,000 General AI based methods, that help the user has to be to be 136 00:09:39,000 --> 00:09:41,400 protected very well and you can imagine. 137 00:09:41,600 --> 00:09:45,800 You can imagine this kind of similar to a current way. 138 00:09:45,800 --> 00:09:51,000 How AI is effective in building, the large language models, what 139 00:09:51,000 --> 00:09:55,300 we did is we've built a similar deep learning methods that were 140 00:09:55,308 --> 00:10:00,400 not trained on natural language. But or Json files, internet is 141 00:10:00,400 --> 00:10:04,200 written in Json. So like 70% of all Fossil the 142 00:10:04,200 --> 00:10:08,800 intent, our Json files structured, but with a variable 143 00:10:08,800 --> 00:10:14,700 length, and to be able to train a eye on any type of Json file, 144 00:10:15,300 --> 00:10:18,500 I was a complexity that we were trying to resolve and we're a 145 00:10:18,508 --> 00:10:21,500 success success success for resolving so. 146 00:10:21,600 --> 00:10:27,000 So one Challenge in a cybersecurity is to be able to 147 00:10:27,000 --> 00:10:32,100 come up with generic enough AI, that can be effective across 148 00:10:32,200 --> 00:10:34,700 different types of current. And future threats. 149 00:10:35,200 --> 00:10:37,800 There is the other huge challenge that we have in cyber 150 00:10:37,800 --> 00:10:40,100 security. That is explained ability. 151 00:10:41,000 --> 00:10:44,700 Yo, cyber Security Experts are like medical doctors. 152 00:10:45,100 --> 00:10:48,100 They know the best. They don't need any AI to help 153 00:10:48,100 --> 00:10:52,200 them classify moreover, right. So to establish the 154 00:10:52,200 --> 00:10:55,900 understanding between the AI researchers and so basically the 155 00:10:55,900 --> 00:11:00,400 expert is a non-trivial and ever, which is why there is a 156 00:11:00,400 --> 00:11:03,300 need to build a lots of explain ability. 157 00:11:03,800 --> 00:11:07,800 It could blow to explain AI to the Marlborough on the list so 158 00:11:07,800 --> 00:11:11,700 that it's accepted. So and we were building a 159 00:11:12,100 --> 00:11:16,100 exciting novel tools, for example, expandability and 160 00:11:16,100 --> 00:11:21,000 cybersecurity, in order to accelerate the deployment of AI 161 00:11:21,000 --> 00:11:27,900 in cyber SEC, are the primary challenges here, lying with the 162 00:11:28,400 --> 00:11:32,900 ability to accumulate sufficient data for your models. 163 00:11:33,600 --> 00:11:39,600 Or in building the broader, General algorithms that will 164 00:11:39,600 --> 00:11:41,900 operate, effectively, on those models. 165 00:11:42,300 --> 00:11:45,200 It's one of the genetic. Let's say so we have an access 166 00:11:45,200 --> 00:11:48,300 to a fantastic data on the internet. 167 00:11:48,300 --> 00:11:51,500 So we see a lot, but the generality. 168 00:11:51,700 --> 00:11:56,700 So that the detectors are fine-tuned, to be able to detect 169 00:11:56,700 --> 00:11:59,600 different types of Marlborough campaigns. 170 00:11:59,800 --> 00:12:04,400 This has been a challenge. So as, as much as, In other 171 00:12:04,400 --> 00:12:07,200 applications of AI. There has been a push to build 172 00:12:07,200 --> 00:12:09,700 an algorithm that is capable of doing more things. 173 00:12:10,200 --> 00:12:14,100 Such as in game, playing the designers design algorithm that 174 00:12:14,300 --> 00:12:17,000 can play goal and shotgun at the same time. 175 00:12:17,000 --> 00:12:20,200 This is the aspect of generality and game playing similar. 176 00:12:20,200 --> 00:12:25,700 Aspects of generality is, is needed in building when we build 177 00:12:25,700 --> 00:12:29,900 AI for a cybersecurity today, the cybersecurity is different 178 00:12:29,900 --> 00:12:34,200 because it's not vulnerability of the operating Eames but is 179 00:12:34,200 --> 00:12:35,800 the vulnerability of people condition. 180 00:12:35,800 --> 00:12:39,300 So the attacks are different, and the attackers are writing 181 00:12:39,300 --> 00:12:41,600 something else. They are not writing Json files, 182 00:12:41,600 --> 00:12:44,800 there are not compiling assembler code with the 183 00:12:44,800 --> 00:12:48,000 attackers are doing there. There are writing text and 184 00:12:48,000 --> 00:12:50,900 natural language. That is supposed to be deceptive 185 00:12:50,900 --> 00:12:55,000 and believable so that the users are willing to open an 186 00:12:55,000 --> 00:12:59,400 attachment or a willing to share their a cyber say there are 187 00:12:59,400 --> 00:13:02,800 financial data there are willing to click on the link. 188 00:13:03,100 --> 00:13:10,500 So it's Totally different type of warfare and their artificial 189 00:13:10,500 --> 00:13:13,600 intelligence is much more successful and much more 190 00:13:13,600 --> 00:13:16,300 impactful but it comes to attacking. 191 00:13:17,100 --> 00:13:21,200 Because in order to be able to craft and deploy successful 192 00:13:21,200 --> 00:13:24,400 couldn't attack, you need three things. 193 00:13:24,600 --> 00:13:29,500 Number one is, you need lots of data about the victim. 194 00:13:29,900 --> 00:13:34,500 You need to kind of collect data about where Where the victim 195 00:13:34,500 --> 00:13:37,900 goes on the internet. Would they like what they did 196 00:13:37,900 --> 00:13:39,700 with the a show? Would they read? 197 00:13:40,200 --> 00:13:43,300 And through this? It's possible that the 198 00:13:43,300 --> 00:13:46,700 algorithms will create more personalized. 199 00:13:46,700 --> 00:13:49,400 Communication, more personalized, could attack. 200 00:13:49,600 --> 00:13:52,600 The second is that you need is to steal somebody's credentials. 201 00:13:52,700 --> 00:13:57,300 Identity theft are increasing the effectiveness of a couldn't 202 00:13:57,300 --> 00:13:59,900 attack. If you receive your the 203 00:13:59,900 --> 00:14:03,200 condensate tank from a from an email of a friend. 204 00:14:03,200 --> 00:14:08,400 It's More likely that you will click or open the attachment. 205 00:14:08,600 --> 00:14:12,500 And the third one is high-performance AI, that is 206 00:14:12,500 --> 00:14:17,100 capable of building. A text that is believable, the 207 00:14:17,100 --> 00:14:21,800 text that is easy to believe in and adopt as a legitimate 208 00:14:21,800 --> 00:14:26,200 message and act accordingly and current high performance. 209 00:14:26,200 --> 00:14:30,600 AI is already by the large language models is the ideal 210 00:14:30,600 --> 00:14:35,200 tool for an explosion of the coordinate attacks and this is 211 00:14:35,400 --> 00:14:38,600 what I'm worried about. This is why I wake up every day 212 00:14:38,600 --> 00:14:43,400 and go to work because I want to contribute to protection against 213 00:14:43,500 --> 00:14:49,300 AI enabled coordinate attacks. I have been the subject of very 214 00:14:49,400 --> 00:14:53,300 targeted phishing attacks. Not have ever been successful to 215 00:14:53,300 --> 00:14:58,100 my knowledge but where people have bad actors have 216 00:14:58,100 --> 00:15:04,100 impersonated people that I know and in texts and Annie Emails. 217 00:15:04,700 --> 00:15:10,500 How can a, i and the tools you're developing protect me, 218 00:15:10,600 --> 00:15:14,300 these things are so believable. And I, I'm so used to being 219 00:15:14,300 --> 00:15:16,700 attacked that I research everyone and I know how to 220 00:15:16,700 --> 00:15:20,400 manually research it. But how can a I help in this on 221 00:15:20,400 --> 00:15:24,000 the contrary thinker? I think differently than others. 222 00:15:24,500 --> 00:15:27,900 And in the past the whole subdue, the consumer. 223 00:15:27,900 --> 00:15:32,700 Subsequently adopted this concept of Cyber security under 224 00:15:32,700 --> 00:15:36,100 the hood me as a user. I do not need to understand. 225 00:15:36,100 --> 00:15:38,800 I don't need to see. I just buy this product and I'm 226 00:15:38,800 --> 00:15:40,700 covered. It's gone. 227 00:15:40,700 --> 00:15:43,400 This is history. Now we live in different times. 228 00:15:43,600 --> 00:15:47,700 Currently we need to build and gauging sophisticated tools 229 00:15:47,800 --> 00:15:51,800 tools that will be there for people will be there with people 230 00:15:51,800 --> 00:15:55,100 and will be helping people to be more resilient against 231 00:15:55,100 --> 00:15:58,700 coordinate attacks. So, assuming that your I 232 00:15:58,700 --> 00:16:04,100 installed this thing, And as a result of this, I will never be 233 00:16:04,100 --> 00:16:08,600 attacked by fish. Or scam is just false 234 00:16:08,600 --> 00:16:10,200 assumption. This will never happen. 235 00:16:10,200 --> 00:16:16,000 So we as a as a Defender need to change the perspective and try 236 00:16:16,000 --> 00:16:20,900 to build a companion tools, that will be there with people will 237 00:16:20,900 --> 00:16:25,300 be gamifying cybersecurity for people will be rewarding people 238 00:16:25,300 --> 00:16:30,400 with more transparency and what is going on when they read and 239 00:16:30,900 --> 00:16:34,800 Receive a message. And when you ask me how a I can 240 00:16:34,800 --> 00:16:39,700 help the large language models that are now used for creating 241 00:16:39,900 --> 00:16:43,200 text. Can be also used successfully 242 00:16:43,200 --> 00:16:47,200 for being able to detect text. That is scam. 243 00:16:47,200 --> 00:16:53,600 That is fraud that this extortion, that is by many other 244 00:16:53,600 --> 00:16:56,900 means malicious. So, the capability of detecting 245 00:16:56,900 --> 00:17:01,800 and classifying text Stacks with malicious intent, and this is 246 00:17:01,800 --> 00:17:04,599 currently enabled by large animal models. 247 00:17:04,599 --> 00:17:10,800 And by AI, which reach, we investigate and study in jenin 248 00:17:11,099 --> 00:17:14,599 here are looking for the patterns among very large 249 00:17:14,599 --> 00:17:19,400 numbers of phishing attacks and obviously based on your work and 250 00:17:19,400 --> 00:17:23,400 your research, those patterns are there if you can only find 251 00:17:23,400 --> 00:17:25,300 them quickly enough, I assume. Yeah. 252 00:17:25,300 --> 00:17:27,800 And the old, we are lucky because we don't, we don't need 253 00:17:27,800 --> 00:17:30,700 to do this work or so this is what the AI does. 254 00:17:30,900 --> 00:17:34,800 Us, you know, we only need to provide a good quality training 255 00:17:34,800 --> 00:17:41,000 data and then let the Deep neural network to learn its 256 00:17:41,000 --> 00:17:44,000 classification. People are always asking me 257 00:17:44,000 --> 00:17:48,100 Miguel why I just can I use a GTP to do this for me? 258 00:17:48,500 --> 00:17:54,300 And you just just ask the question and the chat window and 259 00:17:54,300 --> 00:17:57,600 my response. Always is that cybersecurity is 260 00:17:57,600 --> 00:18:01,400 much more serious deed. We Zenda, lots of 261 00:18:01,400 --> 00:18:05,300 responsibilities to our users. So whenever we help you sir, we 262 00:18:05,300 --> 00:18:08,200 need to be Crystal. Clear, what is it that we are 263 00:18:08,200 --> 00:18:08,900 here? What is it? 264 00:18:08,900 --> 00:18:13,000 That we are trying to user and if we are advised not to click, 265 00:18:13,000 --> 00:18:16,500 we just need to be certain and current setup of the large 266 00:18:16,500 --> 00:18:20,400 large, Which models that are generating generally designed 267 00:18:20,400 --> 00:18:23,500 Angela trained just do not give you the certain it. 268 00:18:24,000 --> 00:18:30,500 So our our added value, and AI pipeline of systems is twofold. 269 00:18:30,600 --> 00:18:33,400 Heh. And we just want to make sure 270 00:18:33,600 --> 00:18:36,900 that the chat board. The large language model that we 271 00:18:36,900 --> 00:18:43,700 query is not giving some of the, your son don't senses answer and 272 00:18:43,700 --> 00:18:46,200 second, what we do is we do problem, think. 273 00:18:46,600 --> 00:18:50,700 So, we are prompting the language model with the data 274 00:18:51,000 --> 00:18:55,900 with like small amount of special sample data that are 275 00:18:55,900 --> 00:18:59,900 helping the rational model to do the classification. 276 00:18:59,900 --> 00:19:04,600 That is direct contextual to a situation in which the user 277 00:19:04,700 --> 00:19:09,800 receives the attack. So we have a question from our 278 00:19:09,800 --> 00:19:15,900 Salon Con on Twitter and he wants to know if AI is being 279 00:19:15,900 --> 00:19:19,400 used for cybersecurity. Does that mean that Boards of 280 00:19:19,400 --> 00:19:23,800 directors don't need to worry about cybersecurity? 281 00:19:23,800 --> 00:19:27,100 In other words is the AI just handling this problem and the 282 00:19:27,108 --> 00:19:30,700 problem is going to go away. So it's very similar to my own 283 00:19:31,200 --> 00:19:34,300 In the third of consumer service regretted, the fact that we are 284 00:19:34,300 --> 00:19:37,700 in a world where ordinary consumers need to worry. 285 00:19:38,600 --> 00:19:44,200 The same applies to the board members so you can get 286 00:19:44,300 --> 00:19:46,600 insurance. But it doesn't mean that the 287 00:19:46,600 --> 00:19:51,000 insurance will work always. So the difference between a 288 00:19:51,900 --> 00:19:57,200 board is that the bird do have a responsibility for business. 289 00:19:57,200 --> 00:20:00,900 So their budgets for an investing in cybersecurity She 290 00:20:00,900 --> 00:20:04,900 is a sort of a different order of magnitude while ordinary 291 00:20:04,900 --> 00:20:09,300 consumers who payday payday their daily bills. 292 00:20:09,300 --> 00:20:13,300 They're they're spotted their Spotify, their Netflix and 293 00:20:13,700 --> 00:20:15,100 whatever they need on the internet. 294 00:20:15,100 --> 00:20:17,000 For them. Can I think an extra for cyber 295 00:20:17,000 --> 00:20:19,100 security as a material part of the bill? 296 00:20:19,500 --> 00:20:24,100 So I would say the difference is with the the investment but 297 00:20:24,300 --> 00:20:28,100 nobody neither ordinary users. Not import members are currently 298 00:20:28,100 --> 00:20:33,500 relief from a Let's intubate the tension and to make the right 299 00:20:33,500 --> 00:20:36,200 decisions. These are tough problems. 300 00:20:36,200 --> 00:20:40,800 And I think for board members, it's in some cases, even more 301 00:20:40,800 --> 00:20:45,000 difficult because they don't have the technology background. 302 00:20:45,000 --> 00:20:47,000 That's necessary to really understand this. 303 00:20:47,000 --> 00:20:51,800 And so they have to therefore rely upon a group of Technology 304 00:20:51,800 --> 00:20:55,800 experts without without the transparency and explain ability 305 00:20:55,800 --> 00:20:58,400 that you described earlier. Yes, yes. 306 00:20:58,400 --> 00:21:00,300 I would say that. Currently, we have the fastest 307 00:21:00,300 --> 00:21:05,500 error, a change in the types of attacks from the Bad actors. 308 00:21:05,500 --> 00:21:08,700 So I think that there is a huge expectation for the Cyber 309 00:21:08,700 --> 00:21:12,900 Security Experts to really be up to the speed to really try to 310 00:21:12,908 --> 00:21:16,800 understand what are the new dangerous threats and also what 311 00:21:16,800 --> 00:21:20,100 are the new technologies that people use for attacking as much 312 00:21:20,100 --> 00:21:22,600 as for protecting? So there is a huge amount on the 313 00:21:22,600 --> 00:21:26,400 experts to really be there for people who need their advice. 314 00:21:26,400 --> 00:21:32,600 Many technology experts are Not sufficiently comfortable 315 00:21:32,600 --> 00:21:37,900 communicating with the boards, which presents a problem because 316 00:21:37,900 --> 00:21:42,100 the security officer, for example wants to explain but 317 00:21:42,100 --> 00:21:45,700 doesn't not know how to present it in terms that are 318 00:21:45,700 --> 00:21:49,100 straightforward enough for the board to understand. 319 00:21:49,100 --> 00:21:51,600 And, and that's a gap that causes problems. 320 00:21:51,600 --> 00:21:54,900 In some cases, I agree with you Michael and there's the other 321 00:21:54,900 --> 00:21:58,300 problem, which is that there's the current economical 322 00:21:58,300 --> 00:22:01,900 environment, the budgets are stretched, Act with with 323 00:22:01,900 --> 00:22:05,400 everybody. So and I guess in the past kind 324 00:22:05,400 --> 00:22:08,400 of big corporations were, if they didn't understand they, 325 00:22:09,300 --> 00:22:15,400 they were going to okay to pay an extra for like an extra tool 326 00:22:15,400 --> 00:22:20,000 that this under security guy, requested these days I think is 327 00:22:20,000 --> 00:22:22,600 going to be different, you know, they'll be budget fights for 328 00:22:22,600 --> 00:22:24,400 everything including cyber security. 329 00:22:24,400 --> 00:22:29,100 So this presents a accorded growth for cyber security 330 00:22:29,100 --> 00:22:32,400 personnel and Caesars. The big companies to be able to 331 00:22:32,800 --> 00:22:37,500 explain better to the board. What is it that they are buying 332 00:22:37,500 --> 00:22:40,000 and why they need to invest? So the time is changing for 333 00:22:40,000 --> 00:22:43,600 everybody, it is always extraordinary to me the number 334 00:22:43,600 --> 00:22:45,400 of companies even security company. 335 00:22:45,400 --> 00:22:51,200 As I mean, look at last pass that have breaches and after the 336 00:22:51,200 --> 00:22:54,200 breach they always say we're going to invest more. 337 00:22:54,200 --> 00:22:57,000 We're going to well, why didn't they do this before? 338 00:22:57,400 --> 00:23:02,100 The fact that the old are those centers on the internet that are 339 00:23:02,100 --> 00:23:07,700 worth breaching that store users private data and the Bad actors 340 00:23:07,700 --> 00:23:11,200 are interested in attacking. I think it's wrong design of the 341 00:23:11,200 --> 00:23:14,300 internet. We should have less and less of 342 00:23:14,300 --> 00:23:18,600 such places and more and more private data should reside with 343 00:23:18,600 --> 00:23:22,900 the users on their hands. Be much more difficult to kind 344 00:23:22,900 --> 00:23:28,800 of make a large curved breach, through which you steal hundreds 345 00:23:28,800 --> 00:23:34,500 of thousands of of By d's and passport numbers. 346 00:23:34,800 --> 00:23:37,900 So I really believe that internet needs to undergo a 347 00:23:37,900 --> 00:23:41,500 change, but there is much more opportunity for users to take 348 00:23:41,700 --> 00:23:47,200 responsibility for their own private data and not those who 349 00:23:47,200 --> 00:23:52,100 just kind of need to validate and verify to check in the 350 00:23:52,100 --> 00:23:54,900 Privacy pre-selling manner. What is it that I keep in my 351 00:23:54,900 --> 00:24:00,300 wallet and my bullet needs to be secure and modern and good 352 00:24:00,300 --> 00:24:03,000 quality. And powered by good compute so 353 00:24:03,000 --> 00:24:05,800 that whoever can and needs to check me check my wallet and 354 00:24:05,800 --> 00:24:11,300 doesn't need to contain the record of my personal data and 355 00:24:11,300 --> 00:24:17,600 the data base, which creates a danger for the vendor, who keeps 356 00:24:17,700 --> 00:24:21,200 keeps my data. Since you brought this topic up, 357 00:24:21,200 --> 00:24:25,000 can you just briefly give us advice the the the listeners for 358 00:24:25,000 --> 00:24:27,300 cxo? Talk are very smart, very bright 359 00:24:27,600 --> 00:24:33,500 and can you give us advice on? As individuals what we can do to 360 00:24:33,500 --> 00:24:35,800 protect our data. Just along the lines of what you 361 00:24:35,808 --> 00:24:38,100 were saying, just briefly very briefly. 362 00:24:38,100 --> 00:24:41,400 We going to need to have a good cyber security hygiene to kind 363 00:24:41,400 --> 00:24:45,400 of work with a password manager. Do know that do not kind of 364 00:24:45,400 --> 00:24:49,600 store or send passwords by a text message. 365 00:24:50,100 --> 00:24:54,200 Use do good tools for cybersecurity understand where 366 00:24:54,200 --> 00:24:57,500 you weren't, where you are sharing private data for what 367 00:24:57,500 --> 00:24:59,600 purpose, whether this is really necessary. 368 00:24:59,900 --> 00:25:03,600 Ask Then there's to delete data, because in many countries, there 369 00:25:03,600 --> 00:25:06,700 is registration of the vendor, is asked to delete the data. 370 00:25:07,300 --> 00:25:13,100 They are obliged to. So be cognizant of tracking. 371 00:25:13,500 --> 00:25:19,100 So we attract without tracking. There is no a personalized 372 00:25:19,100 --> 00:25:22,900 experience on the internet. So we gonna need tracking but 373 00:25:22,900 --> 00:25:27,600 gonna be focused on when. And why delete your cookies? 374 00:25:27,600 --> 00:25:32,400 Do not agree with every single cookie pop up that Is bothering 375 00:25:32,400 --> 00:25:35,200 you. So these are the basic advice 376 00:25:35,600 --> 00:25:40,800 but the truth is and you will actually the storage of private 377 00:25:40,800 --> 00:25:45,900 data is is very much connected with algorithmic manipulation. 378 00:25:46,800 --> 00:25:51,200 The more data the vendors know about myself, the better 379 00:25:51,900 --> 00:25:56,200 personalized digital experience, our receive but the better 380 00:25:56,200 --> 00:26:01,300 digital experience means that the the vendors are restricting 381 00:26:01,300 --> 00:26:05,400 my choices. When I search for an interesting 382 00:26:05,400 --> 00:26:10,000 article if the internet Knows all about me, it tries to 383 00:26:10,000 --> 00:26:13,000 second-guess. What is it that I want and 384 00:26:13,000 --> 00:26:16,000 serves the content that they assume and I need. 385 00:26:16,000 --> 00:26:22,700 So there's there's a piece of manipulation and I truly believe 386 00:26:23,000 --> 00:26:26,900 that people need better. People need better tools and 387 00:26:26,900 --> 00:26:30,600 Technologies for keeping their privacy in check. 388 00:26:31,000 --> 00:26:35,900 But also for understanding how the recommender algorithms 389 00:26:36,600 --> 00:26:39,100 That's on the internet. How they work? 390 00:26:39,100 --> 00:26:41,900 How they work for me? How they work for me in that 391 00:26:41,900 --> 00:26:43,700 situation. In the other situation? 392 00:26:44,000 --> 00:26:49,500 I need to understand when YouTube offers me those tiles. 393 00:26:49,800 --> 00:26:52,600 Why are there? What is it that I did that? 394 00:26:52,600 --> 00:26:56,900 I see this offer what is it that I watch with this added a didn't 395 00:26:56,900 --> 00:26:58,800 watch? What is it that I posted about 396 00:26:58,800 --> 00:27:02,500 why the recommender is acting in a way he's acting. 397 00:27:02,500 --> 00:27:04,600 Well, you know, there's no transparency. 398 00:27:04,600 --> 00:27:08,500 We don't know, we are Expect accepting the recommend 399 00:27:08,500 --> 00:27:11,300 recommendations from the internet as they are. 400 00:27:11,300 --> 00:27:14,100 And this is because this great technology. 401 00:27:14,100 --> 00:27:19,400 It's on the other side, the great technology is at the site 402 00:27:19,400 --> 00:27:21,300 of the vendors of the intent companies. 403 00:27:21,300 --> 00:27:25,700 The amount of technology that is with users in their wallets and 404 00:27:25,708 --> 00:27:28,200 their browsers and their phones is actually quite limited. 405 00:27:28,200 --> 00:27:35,600 So and as 30 years ago there were kind of first Bad actors 406 00:27:35,600 --> 00:27:39,200 and Attackers. And as a result, got going to 407 00:27:39,200 --> 00:27:43,700 big, massive cyber security industry, the game became 408 00:27:43,700 --> 00:27:47,700 reality, and this industry started to protect users, and as 409 00:27:47,700 --> 00:27:50,900 a result of this users are kind of safe. 410 00:27:51,700 --> 00:27:55,100 I believe that something similar must happen, in the field of 411 00:27:55,700 --> 00:27:59,900 privacy and algorithm on Appalachian that there is a more 412 00:27:59,900 --> 00:28:04,600 tank covering people's back when it comes to algorithm 413 00:28:04,600 --> 00:28:08,200 manipulation misinformation. Ian and privacy handling, we 414 00:28:08,200 --> 00:28:13,000 have a couple of questions now that are popping up on Twitter. 415 00:28:13,000 --> 00:28:17,700 So why don't we jump there? And first is from Chris 416 00:28:17,700 --> 00:28:23,100 Peterson, who says Ai and ml tool sound great for cyber 417 00:28:23,100 --> 00:28:28,800 defense is they're analogous research in penetration testing 418 00:28:28,800 --> 00:28:35,000 and Red Team Tools or better traps and honey pots for luring 419 00:28:35,000 --> 00:28:39,000 in threat. Hearse in my bread Labs, we have 420 00:28:39,000 --> 00:28:44,600 done research and we were able to demonstrate that the use of 421 00:28:45,300 --> 00:28:48,300 large language models. For generating malware is 422 00:28:48,300 --> 00:28:54,900 possible that you can generate more over by a lot but by GDP. 423 00:28:56,500 --> 00:29:01,100 The truth is that is it really necessary? 424 00:29:01,700 --> 00:29:07,100 If I look back into how kind of moderate campaigns are Created 425 00:29:07,100 --> 00:29:09,800 writing a piece of mother is only a small component. 426 00:29:10,100 --> 00:29:14,600 And there are kind of script algorithms for automated model 427 00:29:14,600 --> 00:29:19,200 Breitling available in the mower Community for the last 15 years. 428 00:29:19,900 --> 00:29:24,800 So is the contribution of a GDP generated Marlborough. 429 00:29:25,000 --> 00:29:30,800 So changing for a Bad actors. Honestly, I don't think so. 430 00:29:30,800 --> 00:29:36,500 I think that the added value is only limited, however, You talk 431 00:29:36,500 --> 00:29:41,300 about the attacks that are in the form of a coordinate, the 432 00:29:41,300 --> 00:29:46,400 tank, the form of scan and fishing and manipulation then 433 00:29:46,400 --> 00:29:49,800 the story is totally different. There are the role and added 434 00:29:49,800 --> 00:29:55,900 value of a large language models and modern, AI or Bad actors. 435 00:29:56,300 --> 00:30:01,300 It's just, it's just massive the rate through which they can 436 00:30:01,300 --> 00:30:06,700 create a believable unique content is just Amazing. 437 00:30:07,000 --> 00:30:13,100 And not only the quality of the content but also the capability 438 00:30:13,100 --> 00:30:17,100 to test to A/B test the effectiveness of the cognitive 439 00:30:17,100 --> 00:30:21,600 attack, right? You're currently their attackers 440 00:30:21,600 --> 00:30:26,000 who are kind of writing fiction? Email they collect some data 441 00:30:26,000 --> 00:30:29,000 from the internet to learn how affected they have been. 442 00:30:29,400 --> 00:30:31,900 They take some learnings they end up the strategy. 443 00:30:32,000 --> 00:30:35,900 They try something else through a method. 444 00:30:36,200 --> 00:30:41,100 Two reinforcement learning together with large language 445 00:30:41,100 --> 00:30:45,400 model based tax generation. This cycle can get automated. 446 00:30:46,600 --> 00:30:49,500 So this is my worry, you're in the Bad actors. 447 00:30:49,700 --> 00:30:55,200 Start to really automate to generate a unique dangerous 448 00:30:55,200 --> 00:30:59,000 content at the same time to be able to learn automatically how 449 00:30:59,000 --> 00:31:05,500 effective data has been and adapt the the text generation to 450 00:31:05,500 --> 00:31:07,400 me. This is very dangerous and this 451 00:31:07,400 --> 00:31:12,500 is this is an area where we as Defenders need to pay big 452 00:31:12,500 --> 00:31:15,000 attention. It's so interesting that you 453 00:31:15,000 --> 00:31:21,200 just Tribe this A/B Testing because among the attacks that 454 00:31:21,200 --> 00:31:25,600 have been against me because of who we interview, it's EXO too 455 00:31:25,600 --> 00:31:29,500 awkward. You know, we're a Target and so 456 00:31:29,800 --> 00:31:33,700 I get these requests very, very strange. 457 00:31:33,700 --> 00:31:39,300 I got these requests from what appear to be women on the 458 00:31:39,300 --> 00:31:43,600 internet, you know, accomplished women and what I've noticed and 459 00:31:43,600 --> 00:31:45,300 I'm married. I have no interest. 460 00:31:45,300 --> 00:31:48,600 Okay. But I've noticed that the 461 00:31:48,600 --> 00:31:51,800 attackers because they're obviously fake because I 462 00:31:51,808 --> 00:31:55,200 research everything. So I noticed that the attackers 463 00:31:55,200 --> 00:31:58,900 have been changing specific variables. 464 00:31:59,600 --> 00:32:03,400 So they'll change, for example, a little bit about the 465 00:32:03,400 --> 00:32:05,800 background, all the variables will be the same, the 466 00:32:05,800 --> 00:32:10,700 characteristics of the proposed connection, they'll change 467 00:32:10,800 --> 00:32:15,600 ethnicity, they'll change the tone a little bit but keep 468 00:32:15,700 --> 00:32:18,900 Everything else, the same. And so I've suspected very 469 00:32:18,900 --> 00:32:22,500 strongly that that this is a be testing going on. 470 00:32:22,700 --> 00:32:24,600 Yes, yes. And this would be testing can be 471 00:32:24,600 --> 00:32:28,700 automated if there's a Tracker in your email or tracker in your 472 00:32:29,100 --> 00:32:32,300 browser, that is hoping the attacker to can. 473 00:32:32,300 --> 00:32:35,700 I report to understand how effective it has been, is going 474 00:32:35,700 --> 00:32:38,400 to be automated. We have another question now 475 00:32:38,400 --> 00:32:44,100 from Twitter and this is again from our Salon Khan who says, 476 00:32:44,100 --> 00:32:51,500 can a I make recommendations about not collecting certain 477 00:32:51,500 --> 00:32:54,900 kinds of data since it's prone to be attacked. 478 00:32:54,900 --> 00:32:59,600 I think he's referring to recommendations to individuals 479 00:33:00,100 --> 00:33:03,300 about what kind of tracking to allow. 480 00:33:03,300 --> 00:33:06,700 For example or not based on recommendations from an AI. 481 00:33:07,100 --> 00:33:12,200 This is a big challenge or one of my teams we are trying to 482 00:33:12,200 --> 00:33:15,600 understand to which extent a I can help. 483 00:33:15,700 --> 00:33:18,000 People to make the right privacy decision. 484 00:33:18,800 --> 00:33:23,200 So we build good quality AI. That helps the user to assess 485 00:33:23,200 --> 00:33:30,300 and explain, what is the Privacy impact of using this or the 486 00:33:30,300 --> 00:33:35,400 other app or being on this or the other webpage. 487 00:33:35,500 --> 00:33:40,300 So kind of rebuild classifiers and detectors that help users to 488 00:33:40,300 --> 00:33:42,900 give them some basic information about that. 489 00:33:43,100 --> 00:33:49,900 The truth is, Currently users are not ready for and I used in 490 00:33:49,900 --> 00:33:54,400 this information so that they would be impacting their privacy 491 00:33:54,400 --> 00:33:59,200 Behavior. So you're we see that users by 492 00:33:59,200 --> 00:34:03,300 big big parts are taking a binary decision. 493 00:34:03,500 --> 00:34:07,000 Okay. So I don't care or I just got 494 00:34:07,000 --> 00:34:11,199 think Anita or I don't care, I use Google search or I care and 495 00:34:11,199 --> 00:34:12,400 I use DuckDuckGo. Right. 496 00:34:12,400 --> 00:34:15,400 There's actually couldn't nothing in between and I To the 497 00:34:15,400 --> 00:34:19,100 technologists and the technology firms need to build tools that 498 00:34:19,100 --> 00:34:25,300 would allow users to kind of set up their privacy approach in in 499 00:34:25,300 --> 00:34:29,699 very fine-tuned way in the context that the time ago. 500 00:34:29,699 --> 00:34:35,300 Based on what they do, where is with a search for base when they 501 00:34:35,300 --> 00:34:38,900 are physically and let the user to kind of fine, tune the 502 00:34:38,900 --> 00:34:43,900 preferences, and then the able to be with the user and adapt 503 00:34:43,900 --> 00:34:47,800 the privacy. Preferences based on the past 504 00:34:47,900 --> 00:34:50,600 Behavior models. So I actually think this is a 505 00:34:50,699 --> 00:34:54,800 like a missing piece that the users needs to get the idea. 506 00:34:54,900 --> 00:34:58,400 That's an optimization problem. To get the best of the 507 00:34:59,100 --> 00:35:03,200 personalized, dental experience on one hand and protect your 508 00:35:03,200 --> 00:35:05,100 privacy. On the other hand because these 509 00:35:05,100 --> 00:35:09,200 are gonna joint. John, bardos macaw. 510 00:35:09,200 --> 00:35:15,000 Can you describe to us some of the most significant challenges? 511 00:35:15,200 --> 00:35:20,200 Especially the technology challenges that you face in your 512 00:35:20,200 --> 00:35:21,700 work and your research right now. 513 00:35:22,100 --> 00:35:26,900 One of the really fascinating research challenge is explain 514 00:35:26,900 --> 00:35:29,700 ability in the field of cyber security. 515 00:35:30,000 --> 00:35:35,300 So Kara for the last 10 years, many areas and have been working 516 00:35:35,600 --> 00:35:39,800 in the fourth explainable at trying to deliver some good code 517 00:35:39,800 --> 00:35:43,400 explanations on AI verdict in cybersecurity. 518 00:35:43,500 --> 00:35:46,300 This is, this is Is this fascinating? 519 00:35:46,300 --> 00:35:50,800 Because if you're gonna dig deep into the mind of threat, 520 00:35:50,800 --> 00:35:55,400 researcher threat analysts, there is a combination of deep 521 00:35:55,400 --> 00:35:59,500 knowledge, great intellect, fantastic reasoning, and 522 00:35:59,500 --> 00:36:02,900 intuition. And the intuition is the piece 523 00:36:02,900 --> 00:36:04,500 which is very difficult to optimize. 524 00:36:05,100 --> 00:36:08,400 So whenever there are AI scientists actually experts, 525 00:36:08,900 --> 00:36:12,800 they just want to do statistic, they want to optimize your, give 526 00:36:12,800 --> 00:36:16,200 me optimization functional training across the fire. 527 00:36:16,300 --> 00:36:20,900 This is, this is, this is how I people work and to be able to 528 00:36:21,100 --> 00:36:28,800 marry this statistical approach to world with the intuition that 529 00:36:28,800 --> 00:36:31,200 is, inside of the security. That's fast rate. 530 00:36:31,800 --> 00:36:34,600 And actually, I think it's transferable it's transferable 531 00:36:34,600 --> 00:36:38,400 to to other domains. I would say that a medical 532 00:36:38,400 --> 00:36:42,800 doctors are very similar. I think that at a time when a I 533 00:36:42,800 --> 00:36:47,100 will be taking over in the healthcare, they would need to 534 00:36:47,100 --> 00:36:51,000 be able to resolve similar challenges, the challenges of 535 00:36:51,000 --> 00:36:56,700 explaining AI in such a way that we are getting a professionals 536 00:36:56,800 --> 00:37:01,100 that are not being replaced by AI, but their impact is being 537 00:37:01,300 --> 00:37:06,000 Then X multiplied by a proper use of high-performance Ai and 538 00:37:06,000 --> 00:37:09,500 I'm lucky to be in this field that I have a first-hand 539 00:37:09,500 --> 00:37:14,500 technical experience. So this this definitely is is is 540 00:37:14,600 --> 00:37:19,100 is one of the challenges and then secondly, oh I'm I'm really 541 00:37:19,300 --> 00:37:25,800 excited to be in this time of explosion of the large language 542 00:37:25,800 --> 00:37:31,600 models and to see the first time in my lifetime when AI is Being 543 00:37:31,600 --> 00:37:36,900 a true consumer proposition. Until now a, I actually was a 544 00:37:36,900 --> 00:37:42,100 preposition was a B2B Enterprises firms. 545 00:37:42,100 --> 00:37:45,100 Corporations were using good quality AI to deliver products 546 00:37:45,100 --> 00:37:50,000 to users, but this is the first time where users are exposed to 547 00:37:50,000 --> 00:37:54,000 Ai. And I would say that one of the 548 00:37:54,000 --> 00:37:58,500 big challenges for me and for my team's is to understand how this 549 00:37:58,500 --> 00:38:05,700 new AI Is threatening people. What are the dangers and limite 550 00:38:05,700 --> 00:38:08,700 an ethical limitations of this new age. 551 00:38:08,700 --> 00:38:14,500 AI that can hack the can have impact on our users and going to 552 00:38:14,500 --> 00:38:19,500 be there and to think about those dangers and to try to 553 00:38:19,500 --> 00:38:22,800 build a technology that is protecting users against not 554 00:38:22,900 --> 00:38:26,000 current, but future AI dangerous. 555 00:38:26,200 --> 00:38:28,800 It's very rewarding. We have a question from 556 00:38:28,800 --> 00:38:31,000 LinkedIn. This is from From Nash in blue 557 00:38:31,500 --> 00:38:35,300 and she is a, she runs an organization. 558 00:38:35,300 --> 00:38:37,700 That works with Chief Information officers. 559 00:38:37,900 --> 00:38:42,100 And she says this, she says there are quite a few, a I 560 00:38:42,100 --> 00:38:45,200 powered cyber security tools out there. 561 00:38:45,600 --> 00:38:49,700 What's your take on the level of maturity of these tools? 562 00:38:50,200 --> 00:38:55,300 What's the confidence level of organizations as far as adoption 563 00:38:55,600 --> 00:38:58,200 goes? Especially since the popularity 564 00:38:58,200 --> 00:39:01,800 of chat GPT? Should be explained ability. 565 00:39:02,100 --> 00:39:06,100 There are tools that people believe because they've been 566 00:39:06,700 --> 00:39:11,400 well explained and the users have trust in the tools and then 567 00:39:11,400 --> 00:39:14,000 there are going to New Kids on the Block that have hard times 568 00:39:14,000 --> 00:39:18,800 to prove their value. So um, Optimist, there are lots 569 00:39:18,800 --> 00:39:21,200 of great AI tools. That's understood. 570 00:39:21,200 --> 00:39:25,300 Experts are using, is non-trivial to be able to detect 571 00:39:25,300 --> 00:39:29,200 what is working for your use case and was not what advice do 572 00:39:29,200 --> 00:39:33,300 you have? For Enterprise Chief information 573 00:39:33,300 --> 00:39:36,200 security officers. I know it's a hard, it's a 574 00:39:36,207 --> 00:39:38,800 really hard question to ask. Just brought it Vice, but you 575 00:39:38,800 --> 00:39:43,600 have such an overview not to play catch-up, but think for the 576 00:39:43,600 --> 00:39:50,100 future, I know that cisos often are busy with solving the crisis 577 00:39:50,100 --> 00:39:54,000 and with extinguishing Fires at this is pretty much a work. 578 00:39:54,400 --> 00:39:58,200 Not try to find at least 50% of your tongue. 579 00:39:58,200 --> 00:40:02,400 Where you think about the future about What may come up as soon 580 00:40:02,400 --> 00:40:07,600 as the technology becomes better developed, and can more more 581 00:40:07,600 --> 00:40:11,600 powerful, do not us as a cyber Security. 582 00:40:11,600 --> 00:40:16,000 Experts stuck in the, in the current situation, but let us 583 00:40:16,000 --> 00:40:20,400 find time to focus on the future because this is the only path to 584 00:40:20,400 --> 00:40:25,300 resilience and any advice for Chief Information, officers 585 00:40:25,800 --> 00:40:30,700 Chief Information officers need to partner well with see Usos 586 00:40:30,700 --> 00:40:35,500 and cdos and to gonna be in a good company and try to take the 587 00:40:35,500 --> 00:40:40,300 advantage of what the Cyber Security Experts can do for them 588 00:40:40,400 --> 00:40:43,900 and what's the technologist can help them and driving their 589 00:40:43,900 --> 00:40:48,700 future decisions until I don't invite and view the same with 590 00:40:48,900 --> 00:40:52,200 investment in security, but will be under scrutiny. 591 00:40:52,400 --> 00:40:57,100 The same time, we will see investment in it and clothespin 592 00:40:57,100 --> 00:41:03,000 to undergo, big scrutiny in 2023 and R2 a year working on certain 593 00:41:03,000 --> 00:41:06,600 problems challenges in cybersecurity and AI. 594 00:41:07,300 --> 00:41:12,400 Where do you see the results ending up of your work over the 595 00:41:12,400 --> 00:41:16,600 next couple of years? The way how I see evolution of 596 00:41:16,800 --> 00:41:20,600 scan and fish the coordinate attacks on the Internet. 597 00:41:20,900 --> 00:41:27,400 It's from like short. Thanks to long-term persistent 598 00:41:27,400 --> 00:41:31,100 coordinate attacks from manipulation to misinformation. 599 00:41:31,500 --> 00:41:34,400 And I actually think that in the future as current, the 600 00:41:34,408 --> 00:41:38,600 cybersecurity is changing from attacking people devices to 601 00:41:38,600 --> 00:41:40,000 people. Cognition, I think in the 602 00:41:40,000 --> 00:41:43,000 future, the subjects are basically will be changing from 603 00:41:43,000 --> 00:41:47,600 a immediate kind of attacks, which is this one click towards 604 00:41:47,800 --> 00:41:50,600 manipulation. How am I changing changing 605 00:41:50,600 --> 00:41:53,500 People mind that? There would be less. 606 00:41:53,700 --> 00:41:58,000 Resilient and more susceptible to to an attack. 607 00:41:58,000 --> 00:42:02,300 How do I move people between a coach a, so that they would be 608 00:42:02,400 --> 00:42:04,900 more vulnerable? They will have more 609 00:42:05,000 --> 00:42:08,400 vulnerabilities for me as a bad bad actor. 610 00:42:08,600 --> 00:42:14,600 So I think that this combination of a scam and phishing attacks 611 00:42:15,000 --> 00:42:19,700 and misinformation and manipulation will be a big Topic 612 00:42:19,800 --> 00:42:22,800 in AI cybersecurity in the years to come. 613 00:42:23,200 --> 00:42:30,100 So your Saying that the shift takes place from AI in 614 00:42:30,200 --> 00:42:33,800 technology attacks. Meaning attacking firewall, for 615 00:42:33,800 --> 00:42:40,900 example, to cognitive attack, such as fishing to broader 616 00:42:40,900 --> 00:42:47,000 overtime attacks that manifests as misinformation disinformation 617 00:42:47,300 --> 00:42:50,000 and broader psychological manipulation. 618 00:42:50,500 --> 00:42:54,300 Yes, yes, exactly. And this is where I see the job 619 00:42:54,300 --> 00:42:59,200 of a, I substitute cybersecurity expert needs to be even more 620 00:42:59,200 --> 00:43:03,200 exciting and would require a wider scope of knowledge 621 00:43:03,600 --> 00:43:05,700 comparing to subsidy Experts of the past. 622 00:43:06,100 --> 00:43:09,600 It's fascinating, because essentially what you're saying 623 00:43:09,600 --> 00:43:13,400 is that the field of misinformation and 624 00:43:13,400 --> 00:43:17,700 disinformation, which right now we look at as a social media 625 00:43:17,700 --> 00:43:20,200 problem. In fact, it's a cybersecurity 626 00:43:20,200 --> 00:43:23,200 problem. It is, it is a security problem. 627 00:43:23,700 --> 00:43:28,400 Live on to today because the most effective weapon in this 628 00:43:29,000 --> 00:43:33,100 future cyber security, world is making people more resilient the 629 00:43:33,100 --> 00:43:38,300 helping people to be fit. Mentally third, to be 630 00:43:38,300 --> 00:43:43,300 inquisitive to be excited about checking sources and to be a 631 00:43:43,300 --> 00:43:48,500 resilient against future attacks, which is why those who 632 00:43:48,500 --> 00:43:52,800 can expect the technology firms to build a manipulation firewall 633 00:43:52,800 --> 00:43:55,300 that I put in. My browser and doesn't show any 634 00:43:55,300 --> 00:43:58,100 fake news at the wrong approach for one. 635 00:43:58,100 --> 00:44:03,300 It it cannot be done and for a second, it reduces my mental 636 00:44:03,300 --> 00:44:05,900 Fitness are going to need to be there still. 637 00:44:05,900 --> 00:44:09,800 I just want to be on the, I want to be independent autonomous and 638 00:44:09,800 --> 00:44:14,200 keep my life under control and to this, also belongs to my 639 00:44:14,200 --> 00:44:18,800 capability to distinguish, which news I believe and which use? 640 00:44:18,800 --> 00:44:21,300 I don't believe, I just don't want this right to be taken away 641 00:44:21,300 --> 00:44:23,900 from it. Seems like this is a A 642 00:44:23,900 --> 00:44:29,000 particularly difficult challenge because now what you're asking 643 00:44:29,000 --> 00:44:34,900 to take, place is silos of researchers to converge. 644 00:44:34,900 --> 00:44:37,400 Because right now, you have cyber security researchers. 645 00:44:37,900 --> 00:44:40,800 And then you have folks who are looking at news. 646 00:44:40,800 --> 00:44:45,400 Essentially media researchers, these are different groups of 647 00:44:45,400 --> 00:44:48,300 people is they are, but they'll be soon. 648 00:44:48,300 --> 00:44:51,900 Sharing the same objective. We make the internet a safer 649 00:44:51,900 --> 00:44:55,500 place for everybody. Well, Well, on that, I'm afraid, 650 00:44:55,500 --> 00:45:01,000 we are out of time so I shall. I just want to say a huge. 651 00:45:01,000 --> 00:45:04,100 Thank you for taking the time to be with us today. 652 00:45:04,500 --> 00:45:08,100 Thank you for having me Michael. Thank you so much to McCall. 653 00:45:08,100 --> 00:45:13,100 Piccu check the CTO of Jan. Now before you go, please 654 00:45:13,100 --> 00:45:15,700 subscribe to our YouTube channel. 655 00:45:16,200 --> 00:45:19,500 Hit the Subscribe button at the top of our website so we can 656 00:45:19,500 --> 00:45:21,800 send you our great newsletter.