MJ Casado | Cybersecurity Risks of AI

Yes. Okay. Um, hello, my name is MJ. Uh, first of all, thank you very much for having me. And I'm sorry this is nap time. Very bad timing for this solid uh topic to discuss, but well, I hope you enjoy. Um, let me introduce you to this today. I know that back in 2019, I think it was software as a service, the word of the moment. Then back in 2022 it was uh blockchain the word at the moment. Okay, welcome to 20 20,000 when it is um artificial intelligence. So let's take a look at the impacts of uh artificial intelligence on cyber security. Um I wanted to sorry um sorry now I wanted to introduce

myself just want you to know that I'm into threat intelligence. I'm not a red teamer, not an incident responsor. I mo most of my time into the intelligence thing and apart from of cyber security, I'm passionate about flying the world, playing sports and New York City. If you have any question about that, I know more about New York City than artificial intelligence and int and and um threat intelligence. Well, uh in a nutshell, right, let's introduce the topic. what what is the gist of what really interests us from a cyber security perspective regarding artificial intelligence? I don't want a very solid uh introduction or description or um let's say um uh definition. Okay. Okay. The most

important things are three things from my perspective. Um artificial intelligence simulate human behavior. It replicates the structure of the human brain and it also uh learns from data. Um I'm sure that you all are familiar with artificial artificial intelligence and all of you use them on a daily basis. I do myself. Those are the my four four favorite ones most favorite ones. Okay. What are the implications for uh of the use of artificial intelligence from a cyber security perspective? In the one hand it's automation capacity is massive massive the automation capacity this technology has. It also has a very strong computing capacity and finally it learns as we said. So in a nutshell artificial intelligence made cyber security and

cyber threats bigger than ever uh because of threat actors obviously exploiting artificial intelligence and artificial intelligence tools. Um they enhance conventional cyber attacks. You can think of many examples. I will I'll show you some of them later on. Uh but also threat actors can manipulate artificial intelligence models making them even more dangerous. And of course there is some unwilling exposure affecting and impacting cyber security from the use of uh artificial intelligence like for example in terms of data privacy uh well sensitive information can be used for training and can be leaked and also the there's another point which is bias and well this is not the the gist of our presentation today but that um could um

well uh let let's say that could be another presentation but anyway What is the challenge for us uh of threat actors using artificial intelligence? Well, that means a huge massive capability uplift. What they were able to do before in terms of the scale, sophistication and speed, it is massive now. Um well, let's see some examples. Of course, impersonation, uh, social engineering, spur fishing, well, it bolstered scope and efficiency, lowered, uh, the barrier of technical skills for non non- very sophisticated threat actors. We'll see that. And also reconnaissance. This is my favorite because this is what I do um as for my as my daily tasks, so to speak, uh, doing um digital footprint analysis. And well threat actors can

look for a better targets and also look in a better way for vulnerabilities to exploit. Also discovering and exploiting zero days of course identificating them and also creating custom uh malware to exploit. We'll look into that later on. and also compromising the very artificial intelligence systems uh manipulating them degrading performance reliability and also security related consequences like for compliance or for financial loss um regarding um degrading performance and re reliability uh well I think that's an issue maybe it might not be today but it might be in a few years when we have all incorporated uh artificial intelligence in our daily basis if we are not doing that. So, but let's let's focus for a minute in

compromising artificial intelligence systems. Let me present you with uh Google security artificial intelligence framework. It's called safe well S Aif risks. Uh Google has identified or ruled like uh the possibility of compromising artificial intelligence systems in two levels or two phases. when the model is being created and when the model is being used. Uh in terms of creation of course uh well there are two main factors. On the one hand is uh manipulating data and on the other hand is manipulating infrastructure and well what most what is the most important part for me which is model application um in in at this model usage sorry uh the use of application and well Google has some controls and some theoretical

controls uh on this how to um mitigate them but I think um today the topic is anal analyzing all of those uh all all of this framework how it impacts cyber security and what can we do as cyber security defenders. So um let me show you a little bit in detail these those risk identified by Google uh but um I would like to focus on the ones that really impact cyber security from a direct direct standpoint from my perspective. Um, of course at the at the timing of creating artificial intelligence models. Well, definitely uh there are some data poisoning and things and stuff, but I think it's the the key word here is using I think because uh

well, let's think of the of the idea. Of course we can uh get uh poisoned artificial intelligence into use but I think the the most um likely situ situation is that we use some exist existing models like the ones that we have previously seen of general general use even if all all the models can be applied to all different tasks and well as I said I expect in the future to um artificial intelligence models being there worldwide and well tampering them at the creative phase will be an issue but let's focus on today Okay. Um I think there are some hybrid risks risks at the moment of creation and consuming but also some that are exclusively of at

the consuming time from Google perspective and I think they are worth um pointing out um yes in terms of uh the med creation even if I didn't say I said I I I wasn't going to focus on them I want you to take a look at them well data poisoning unauthorized training data uh model source tampering and excess excessive data handling. I can think of two examples at the moment of some of them uh regarding unauuthorized training data. I remember back in 2023 when Spotify removed some content because um there were there were some songs that were being created learning from the training data on on from illicit uh training data. So which was funny and regarding excessive data

handling I think I can think of uh a good example but it's uh I think it's almost everywhere. It was um not too long ago uh I don't remember the exact date but I'm thinking of Samsung when uh Samsung banned the use of artificial intelligence in the whole company because of exa excessive data handling uh all the processing all the retaining the data or the database was massive. Uh well, but let's move on to these hybrid risks that Google describes and um um th those are the um uh this this phase of model creation and consuming uh of course model Xfiltration when the model is being exfiltrated. Uh I think there is a good a good example of this

one as well. I remember I think it was I don't know in English but in it's Lam I think is the the artificial intelligence geni of meter well it was uh leaked in a uh in in the underground markets like 24 hours or something uh after it was released and well it was obviously unwillingly but let's focus on model deploy deployment tempering which is one of uh special interest from a cyber security perspective uh well that means the modification of components as you can see used for deploying uh a model deployment basically um prompt injections this is my favorite this is causing a model to execute commands injected inside a prompt we'll see an example of that later on model evasation

of course and uh sensitive data disclosure this is I think one of the the most serious nowadays uh I don't know if I've told you that uh I I work for a company that is into the yaching industry and maritime industry. Well, uh I remember once uh doing like some plane with CH chat JPT and finding out uh an actual the data of an actual contract of uh a yaching uh management company in the yaching industry of one of our customers. Well, there you go. you know they were using the model just um uh unconsciously and also of course inferred sensitive data when the model is capable of inferring the data and uh this infer in inference is actually true

finally uh well those are the ones that interest uh us I think or cyber security defenders uh the most and uh yes and let's move on of the third category of risks identified by Google which are the ones on the consumer consumer side on the consumer end. Uh denial of service of course you just have to translate the concept of denial of service to artificial intelligence reducing reliability and stuff. Uh reverse engineering um insecure integrated components. There are some times when all the models have some vulnerabilities within the software libraries or whatever it uses. Um and an insecure model output of course uh well the output is not appropriately validated and you know results can be really

really messy and get some risks and and of course rogue actions but uh well that is uh more of physical um yeah this is uh the Gemini and Google approach let us move on a little bit to the miter atlas uh the miter atlas is basically the database uh created by MIT um well with a knowledge base and let me tell you something of course they have introduced in all the techniques and the moments all the all the particularities of uh um how artificial intelligence can exploit the or can use the miter um framework and of course my purpose is not describing you all of them I just want you to to to be aware of this the

existence of this and and and to check it because it's really really curious to understand how artificial intelligence really works because let me tell you that I have been working on this for one year and a half trying to investigate and and and dig into the risks to translate it into my general my daily practice and you know the lack of transparity transparency is massive. So this can get you a very good insight. Um uh well I also invite you to take a look at the MITER Atlas navigator and and you can see here in green you can see the specific uh functionalities that can be enhanced by the use of artificial intelligence. I think it's it's very

curious and it's very interesting. Oh, I wanted to provide you with an example. H let me before going into the example, a quick one by the way. Um I don't know but uh I'm sorry everybody's talking about how artificial intelligence is empower empowering threat actors and all attacks and everything but you know what most of the examples I can find they are research in a lab by some companies and together with univers universities this is a very good example of this and most of them by the way they are registered in the miter Atlas case studies database which is a good one but some of them are date back date back to 2019 2020 very old ones for

me it was hard to find one some recently well this was a a well it's a it's a worm it's a worm that uh was developed by developers in a lab and with a zero click well it was able to replicate from one model to another without doing anything and then tampering results in a nutshell you can have u here is the all the detail but anyway we we we don't have uh so much time and I want to tell you about some of the things uh well let's go back to reality okay we have seen all the some of the research well um some cases that were publicly published and that we have seen Android

malware iOS malware steals faces to defeat biometrics with artificial intelligent swaps Well, uh I don't know. My mom, she has a twin sister and you know when my my my my mom's sister, she can unlock my mom's face uh my my my mom's mobile and she uses no artificial intelligence. But anyway, this is something Yeah. Well, but this is something that um you know it's it's going to happen more and more I guess. Uh this is a tricky one or a good one. I will talk about this at the very very end of the presentation. uh how they uh attackers can scam people by getting them getting emotional because they are think that they are listening to the

faces of their loved ones. Ah this is another good one I think I'm sure you have heard of it that was last year I think. Yes it was February 2024 when somebody thought that he was he was getting a a team's meeting call uh from the CFO of the company and and and ordered a bank transfer for 25 million. Well, anyway, using uh deep fake and well, I'm I I really like cars. So, this one uh strike me when um I heard that the Ferrari CEO was impersonated, but well, I think it u eventually it was disclosed, but I think that no, I don't remember, but I think nothing happened, but um they didn't lose money, but it

was a good try. Um okay, going again, trying to go into real life cases. uh I don't know if you have seen this report by uh chat JPT well u the company that uh uh has created that model uh about the use of malicious actor of chat JPT well there is some information on some information on uh information operation the use of it for um influence for uh hybrid work for deep um deep news fake news sorry and stuff. But well to be honest, the only real report or or or reports that I have found with information on how threat actors are using artificial intelligence is the one of Gemini. Uh they it's called Gemini

report on adversarial misuse of generative intelligence. And let me share with you their conclusions. Uh boom, there are handful of cases. the majority of loweffort experimentation uh just to jailbreak prompts and trying to bypass safety controls but that is the bulk of the user. Um, there are some malicious actors trying to use Promp Gemini for guidance on how to abuse Google products, which is which is fun. And well, there are some governmentbacked attackers that were trying to do some coding, some scripting, gathering some information for reconnaissance, uh, researching on known vulnerabilities and well enabling post compromised activities like defense evasion. again again uh let let me go back just for a moment why are we focusing on what

uh Google is saying of the use of artificial intelligence because as cyber security defenders and practitioners I mean it's very difficult to find real examples apart from the ones from the news well um I think during the uh during lunch uh I was discussing with some some of you uh about uh APS and nation states and well that is something I I'm really interested in uh for personal reasons and well GE this Gemini report dedicates some uh sections to talk about how uh threat actors real threat actors the big ones are using artificial intelligence and their artificial intelligence and when well it is surprising it's a it's a a curiosity is the fact that Iran a

threat actors are like uh the heaviest users and they use it for a wide range of purposes including research on defense organizations and vulnerability research. Uh well of course crafting fishing and conducting reconnaissance. Um well they are interested in in the US and all the defense uh organizations in the western countries and well also generating content on cyber security. Uh this is particularly AP 42. I I think uh Google or the the report says that in in Iran I think there are like 22 AP registered that use uh uh Gemini as artificial intelligence uh geni model. Uh well, Chinese of course uh well they basically use this for reconnaissance, scripting, development, troubleshooting, code and and um well on a research on how to get

deeper access to targets and well topics they are interested in as you can see lateral movements, privilege escalation, data exfiltration, data exfiltration and detection aversion. North Korea uh interesting one as well. Uh well, they have used it to to well support several uh parts of the attack life cycle including research uh reconnaissance, payload developments, uh assistance with scripting of course, evasion techniques and research on topics of interest for the for for North Korea like cryptocurrency and the Korean and some military South Korean military things and also US uh military uh um infrastructure and and something that is it's uh funny to me as well which is also curious is the fact that um they have a very strong policy of

infiltrating western uh cooperations. So you know they're trying to to to to embed North Korean workers into western countries uh companies. So well they have used artificial intelligence to enhance their power to be uh um hired And uh yes that is um that that is curious from my from my perspective. Finally Russia uh with Russian there is something curious as well. Remember that we are talking about Gemini only. Okay, we we have no re reliable data on chatter JPT apart from the fake news creating and some other things of that like but um they are not using Gemini at all or if not at all uh they are very they're very uh are using that very

limitedly. We don't know if it is because they don't want to disclose information as they the other ones are doing on what are their interests on the model or I don't know any other reason or maybe because they don't find it interesting or useful we don't know I think it might from my perspective probably it is the former but anyway uh still they are the the only use they have done it has been um focused on changing converting uh artificial intelligence model and uh and developments into different um developing languages. Well, and adding encry adding encryption uh functions as well. Uh well uh just to give you just a big picture but uh or a short picture

short picture of um how they're doing them for the different phases of the m the life cycle of the attack. Well, for the face of reconnaissance, Iran is doing reconnaissance on experts, international defense, topic related to Iran and Israeli proxy conflicts. North Korea is doing research on companies across multiple sectors and geographies. Uh the US military and operations in South Korea like we said and on free hosting providers, free hosting vendors. This is curious as well. China, US military, US uh based IT service providers. uh understanding a public database of US intelligence personnel. Let me tell you that um there are some public evidence as well that Chinese has a similar policy like the one we have

seen in North Korea u trying to embed Chinese workers into um different corporation worldwide to get information. And just uh let me get a little bit sidetracked just for a second. Um and going back to my threat intelligence backgrounds, well um you all will be very aware that uh most of the uh China Chinese uh well not most but a huge amount of uh Chinese activity in the in the cyber space is related to cyber espionage and their interest on that. But anyway and then they're also doing some research on target networks. Okay. and determining um domain names of targets. Next phase weaponization. You can see what they are doing here. Here we don't have much

information on who sorry on who is doing what but well basically developing and conversing rewriting and well uh things related to encryption as well. uh regarding delivery, what they're doing, better understanding, advanced vision techniques, um generating content again uh generating uh uh yes on defense organizations and on and well generating content um like like like we like we saw uh with artificial intelligence themes uh in terms of exploitation well reverse engineering endpoint detection um access Microsoft Exchange using password hashes is uh research vulnerabilities in win rm protocol understand publicly reported vulnerabilities anyway regarding installation well signing an outlook visual studio tools for office plug in and deploy it silently to computers add a self-signed certificate anyway you can

read them all there I just wanted you to have some bullets some ideas of how uh how it is uh the practice of how it is used for the different uh like I said stages of the um attack life cycle common and control. Well, establishing communication channels with the compromised systems. Well, here you have um some some examples of what they're doing and and action and objectives uh achieving the the goal. Okay. Um for example, data theft or disruption or or whichever it is. Um what are my overall conclusions after reading this report and doing some research trying to combine it? There is another good very very good report I will me mention later on but it's related to this one and it's

been published by the um UK national cyber security center it's it's it has some some some some inputs as well my overall conclusions are that uh like I said nowadays most of the current discourse and misuse uh um about artificial intelligence is confined to theoretical research Uh yes there is a huge potential but uh they don't really reflect nowadays the reality of the threat landscape we face. Um we would expect a disruptive change that is not taking place but um it depends on the profile of the actor. Skilled actors take uh well take advantage and harness artificial intelligence in quite an efficient way while less skilled actors they well they learn they use it as a productivity

tool. They also uh have experienced a um decrease of the technical skills required for cyber attacking. But anyway, this is u most uh of it and also um two conclusions I have is how artificial intelligence and LLA large language models are you can be used by um threat actors um to um accelerate their campaigns speed. again uh remember that I called it a an uplift capability uplift and also well uh it can be used uh to instruct um artificial intelligence agents to take uh malicious actions. Um well you know I work for a company that has a sock. So I wanted to present you well two cases that we have seen uh within or soak uh incident

response and monitoring activities like the practice. I presented you Gemini and I wanted to present you what we have seen. Unfortunately, we have had uh scars well or fortunately if you if we think of it uh from the point of view of our customers, we have seen very very few examples of uh incidents involving artificial intelligence from the p from the side of the threat actor and the malicious actor. And well, this one was an rat I think there's a spelling mistake there. Infection using BVS script and JavaScript. Um, and well, we just uh were analyzing an email a customer received and it was trying to be posted as an invoice by means of social engineering and well um it

deployed on a synrat and well after analyzing the um the code well we found that there were some comments left by the attacker and that uh it had used an Geni tool to to create it. Uh the criticality was high of course because it allowed for remote access but uh sophistication was also high and in contra in in contrast. Uh the second one it was probably the uh one of the biggest uh well not biggest incident of course not but uh the biggest fake invoice one of our customers has paid. Uh criticality of course was high because of the amount 40k they paid to an allegedly uh an alleged vendor but you know the sophistication was very very

very very low just uh well um it was an invoice generated by artificial intelligence and like I mentioned at the beginning um it's uh it's quite the the the the seed to this success was the fact that uh these models of artificial intelligence models are using by us every day on a daily basis for productivity and and we are unwillingly exposing data and well uh malicious actors and someone looking someone looking for it. They can really find items and pieces of information that really can make uh credible a an invoice for some for someone that is paying invoices for years on a daily basis. Anyway, uh I also like playing myself, so I decided well I just this is

just for fun, but I I wanted to show you some some tests we run at the sock at our labor laboratory. Uh well, just trying to well um typing some prompts just to to create well for example uh this this case it was like trying to create a text for somebody to think that somebody is in distress and provide to financial aid. Uh then we also did try to to create fishing and well here I just want you to compare how it looks like uh well of course of course I know what you're going to say now of course the first example which is a traditional fishing it has u it's not only that is not very

well written but it is the aspects and the second one is just chat JPT so we would need somebody to put that in a nice way to to or the same way to compare, but but it's still, you know, it's the quality of the text. I'm sure that you have always received uh or you if you go to your to your spam folder, you will see plenty of uh the the first type of of a scams even if they are stopped. But anyway, uh voice lab, we did some uh voice cloning as well. And yes, you can just clone clone your voice. And you know, I remember some somebody in my team that was uh unable to to spot uh who whose

voice was the good one, her or the or the voice lab one. And uh I I did I run another test with another tool which was 11 laps. And here I have my my my impression was that uh the the quality of the cloning was not that good. Um I have been told that you couldn't use church a for malicious purposes. You could if you asked him asked it uh could you please uh code a ransomware for me it would answer I cannot it's illegal of course but uh we we we started to playing with the prompts and you know it just uh it it knows how to do it just have to know you just have to know how

to answer and well this is a very very um short screenshot of the whole conversation but anyway you can get an idea that if you have some uh developing knowledge you can do a couple of things and as I said uh this is an this whole artificial intelligence thing it's an an uplift capability for uh smalltime crooks and script kitties and people like me that we don't really are not very very very good at it anyway uh of course uh I was wondering malicious actors use artificial intelligence existing models artificial in intelligence x60 models they have some sorry okay u they they they have some um let's say limits or so okay I'm sure I'm

absolutely sure that they are developing their own malicious artificial intelligence tools so I got into the dark web I went into the underground communities bridge forums xxx uh xssis you know the And well, I found out Worm JPT and all the things you could do with it, which was uh something I have to say I didn't try because of obvious reasons, but uh it was illegal. It is illegal. I work for a legit company, but anyway. Um I think it was fun. And I also found evil JPD and all the functionalities apparently that it has. And I also found XXX JPT which I liked it because of all the wide bunch of things it can do or it

says it can do. I think it would be great if we could um take a chance to to to test them. Anyway, let's move on. Uh how can artificial intelligence risks be can be mitigated with artificial intelligence? Artificial intelligence like we said is a very powerful tool. It can enhance malicious act activities but it also can enhance uh defensive activities. It can be applied for different cyber security process like threat detection, flagging, suspicion, uh suspicious activities, telemetry in real time. As we mentioned at the beginning, the automation capacity uh the u volume of data it can process it may it turns into a very powerful thing also with vulnerability assessment. Okay. uh well of course pinpointing

exploitable weaknesses and in the in terms of incident response of course it can helps triage assessing alerts and well helping analysts dedicate their time not doing all the triage rather than um uh focusing on on on the actual uh alerts when they have been triaged properly. uh accuracy uh oh sorry the volume of data the because of the speed it has because of the accuracy it has some if it's well trained of course and anticipation capacity it is a very powerful tool but here's my question I don't know how many of you have been trying to assess the performance of artificial intelligence applied to cyber security defensive tools I've have loads of vendors that come to me and say my EDR

is the best. Um, my CM is the best. It incorporates artificial intelligence, but I cannot see how there is a huge lack of transparency. So, I don't know. I think we'll have to to dig a little bit more on this. How can we demonstrate that uh the the models they are being that they are being used in the first place uh that they're being applied to that technology and the this how how they they perform. I was just trying to investigate on this and this and this and I found a curious example not curious but one that I thought it was not worth noting. Magical. This is uh Google again. This is a file type

identification for malware detection. And the important thing here is this is that they for the first time uh and correct me if I'm wrong if you have some other examples. I'm open to them. I I'm I'm really really curious to see this uh delivery performance uh assessment um how it helps or it compares to other tools in terms of malware quality and compared to different solutions. Okay. Magica. Uh well let's give it that try if you have time uh and and check right let's move on what to expect in terms of cyber security and artificial intelligence and attackers using artificial intelligence um of course an increase in the use of artificial intelligence for many things

very like I said not rocket science very very rudimentary things very basic things like fishing vision and stuff Of course, all the deep fakes for identity fraud, voice cloning, cloning and so on and so forth. Even know your customers uh processes. Let's see. There are lots of applications an increase of course of the artificial intelligence used for vulnerability research code code development and reconnaissance since it's very powerful like I said for identifying the best targets to touch and to attack and for identifying the level of exposure of the targets also threat actors will be able to analyze exfiltrated data I think faster and more effectively um also the time uh is going to between the the release of security updates and

the time with the attackers exploiting those vulnerabilities is going to be uh faster faster is it's going to be uh uh a challenge. I think this is one of my the one that uh I think it's it's it's very important from a tech from a from a practical perspective. Okay. For cyber security uh practitioners and defenders um also more sophisticated uses of artificial intelligence operations. um are likely to be restricted to and this is key right to threat actors with the resources with the quality uh with with data training data that have good quality with time and with significant expertise. If you think of this, that leaves out loads of different threat actors apart from nation statebacked and

criminal groups, organized criminal groups. Uh I think those sophisticated use are going to be reduced. So basically it's going to be I I'll show you a chart in in in one minute and you will see my prognosis of what is how it's going going to do. So basically threat uh very powerful threat actors very skilled threat actors they will be the ones that are going to benefit from artificial intelligence geni models. Um also there is there is going to be an increase I think uh in underground forums in underground markets in the dark web and everywhere of LLM models for that are unlimited like the ones we saw like uh chat um fraud well fraud I

didn't mention but this is another one xxs JPD and and warm JPD and and everything and well uh here let me just add the last one which is not that relevant for us at the moment don't know if it will be in the future maybe but uh it's related with geopolitics. It's the fact that well uh nation statebacked actors will continue with their information operations because as I like to say you know the war nowadays it's it's it's used to be by sea by air by land and now it's also in the cyber space. So uh I think cyber cyber threats are are another weapon to be used in geopolitical uh confrontations between countries. So it's going to be

um an issue. Anyway, uh here yes, here is my my impact assessment for the next two years. Since everything goes so fast, I wanted to do like a fiveyear um strategic uh assessment, but uh I didn't dare. I just uh took these next two years time and I how are they going to benefit from artificial intelligence? Well um like I said high capable state threat actors they will really benefit in intent in capability social engineering for social engineering for exfiltration moderately for reconnaissance and for uh developing tools for malware or exploits and well from lateral movement I'm not that sure it's not that I have uh many data but I'm not that sure in the in in between

the high capable and the less skilled uh at threat actors we have like the capable some organized cyber crime groups needless to say that there are some organized cyber crime groups that should be considered in the first place or in the first category but anyway let's put it like this not to be so scary and I think they will also be able to benefit quite a lot but um a little bit less that high capable uh state threat actors and finally less skilled hackers uh well they like I said uh or IC hackers well they will benefit but not that much. They don't don't harness it so much. And um depending again on the threat actor profile what are the

implications these uh use from my perspective is going to have for cyber security. Well, highly capable state threat actors, they will be best placed to harness artificial intelligence potential. Oh, artificial intelligence tools and they will um use it for advanced operations. Uh the second capability well it would be a capability uplift like I says uh for reconnaissance, social engineering, expilation um well and uh artificial intelligence enabled tools like XXS deputy and less skilled uh hackers or hackers for hire or opportunistic cyber criminals activist for example. Oh well I think they well they will just get uh their barrier load to entry to effective um scalable access operations and well also volume don't forget about volume they will increase

their capacity to to do at this speed. So for me expertise equipment uh time and financial resourcing is going to be key on what to expect. Uh well this is just uh my final my wrap up to to this. So what can we do? How can we enhance cyber security in the uh in the light of artificial intelligence? I think we have to of course deal with the technical perspective. Um, of course, use cyber apply cyber security measures that can be efficiently deployed like enable security protocols, monitor and support capabilities. All of these artif

uh technologies that use artificial intelligence and try to assess and get vendors to show and to demonstrate how they enhance their cyber security capabilities. and also well for example for CMS you know and also a human perspective and this is something that I always recommend to whoever is uh my audience in any of the presentations I I have the chance to to have uh the human perspect perspective is always key even if we have the most sophisticated technology in the whole world like artificial intelligence we can always defeat it from take if we have taking or or the human's perspective have a a huge huge weight in this. Of course, awareness training, of course, having

always a suspicious mindset and uh controlling different factors. When there is something that is urging you to do something, be suspicious. If there is something that is getting making you getting emotionally uh like remember the example of these scammers uh faking the voice of one of her, um loved ones in distress. Well, if you're getting too too emotional, stop. There is a red flag there. And finally, my favorite factor, the smartest. How about when you have the chance to buy an iPhone 13, 16 Pro Max for €400? You're going to be very happy. You're going to say, "Listen, you know, I got these much much much cheaper." Another red flag, be careful because probably it's a it's something you

should uh doubt about. And I think uh that's a wrap for me guys. Uh I just hope you enjoy and thank you very much.

Test test volume. Oh, do we have any questions for MJ? No. Okay. Uh the next presentation