The Dark Side of AI: Hacking with AI and Exploiting AI Security Flaws
![The Dark Side of AI: Hacking with AI and Exploiting AI Security Flaws The Dark Side of AI: Hacking with AI and Exploiting AI Security Flaws](/pic/the_dark_side_of_ai_hacking_with_ai_and_exploiting_ai_security_flaws/WVpxaVdGeXEtT0U_.jpeg)
are you seeing a rise of AI used in attacks what's happening in the real world not like what's in the movies but you know what are you seeing and is AI actually being used for offensive for hacking etc? The data is out there I I've seen recently with slack like using customer data even if you're paying to you know train the AI uh they didn't even ask for permission they just used it so from a privacy point of view AI bit of wild west it feels like from a cyber security point of view it seems really worrying any hope? There's hope there's absolutely hope whenever it comes to security a lot of security-minded organizations are trying to figure out what is going to be the guidance for not only today but 3 months from now 6 months from now a year from now and so [Music] Hey everyone it's David Bombal coming to you from Cisco live with a very very special guest Omar welcome. Thank you so much for having me here. Great to have you here you're very very well known in the industry I believe 27 books at the last count right? Yes sir I used them to raise my monitors so they're perfect for that. I don't believe that for a moment you've written books on all kinds of topics perhaps you can give us like a quick overview of some of the books that you've really enjoyed writing or that are very popular that people have given you feedback on but I mean 27 and you're writing new books as well I believe. Yes sir it's it's it's not overnight I'm sure all right it's been been at Cisco for 25 years and I started to contribute to some books before that it's a wide spectrum from certification books to books that are now used in University and a few related to how emerging Technologies like AI is impacting pretty much everything cyber security, networking, programming and so on. So I know some of the books that you've written like
uh PenTest+ you've also done Cyber Ops for Cisco and you've done some CCIE security books as well right? Yes sir the uh CCIE the exam that you're required to pass in order to get your CCNP yeah and CCIE yeah. And more recently another AI cyber security book is that correct? Yes sir actually I just published two uh about the same time, one was with a very good colleague of mine from the Oxford University his Miss Pitar yeah and um that one is around securing AI implementations a little bit more from a high level as a m fact I'm working on one that is a little bit more in depth right and my joke in those books is that by the time that I press the first key they become obsolete that technology is changing but it gives you a blueprint yeah of how to look into uh how AI is impacting the Technologies, what are the tools that are necessary for you to make sure that you're securing AI correctly and also how to use AI for security right so you have the two dimensions of using AI for security but also securing AI. What I do for Cisco mostly is a little bit of both but also looking into the security of the AI implementations across the whole portfolio. I need to a because I think a lot of people will be interested in this are you seeing a rise of AI used in attacks because I mean the one that we always know is deep fakes or you know videos with uh Tom Cruz or you know things like that but perhaps you can show you know tell us what's happening in the real world not like what's in the movies but you know what are you seeing and is AI actually being used for offensive for hacking etc? The short answer is yes, yep the longer answer is is beyond the fakes is beyond the traditional social engineering attacks right that's what I call more of the attacks of the 2023, 2022 interesting um the new waves that people are seeing nowadays is different officiation techniques okay and where attackers and if if you think about AI is abstracting a lot for people, so the way that you program the way that you interact with the computer is changing. So nowadays you have non-technical
people that can use you know libraries like LangChain, Llama index and lot of obstruction into the things that you don't have to have a PhD in data science to then use AI for something that something can be to create a new phone, creating your exploit, create a new occlusion technique look for vulnerabilities the perfect example of that that is beyond research last year to to catch the eye of my organization I did a workshop and I cloned the 12,000 security advisories at the time in GitHub and then what is in the security advisory well a link potentially to a GitHub pull request yeah what is in a pull request? A commit. What is in a commit? A diff. What is in a diff the code that was vulnerable yeah and the code that of course you know it has the fix if I put that into a model and say hey tell me what are the best ways for me to create an exploit of course if you ask ChatGPT is going to block you yeah there's a whole bunch of on sensor models that you can send that out so of course for my demonstration I uh I was showing you know how in some cases you know will hallucinate it will give you some very generic cross-site scripting you know always just script text but in some other ones you know opens your eyes and then you can ask of course what are the CWEs the Common Weakness Enumerators the root cause of the vulnerability Y and explain it to a developer so they can do not commit the mistake or you can actually do that to another model so make it better which is a different conversation but you know I digitize my voice so you know it will be a little bit real so the de fake was there fast forward a couple of months now I'm seeing you know very early indicators of and I will not say early indicators we're definitely seeing some um evidence that attackers actually using these techniques of course to create exploits yeah and especially in third party software in open source yeah that's the reason of you know the big push for supply fly chain security and you know the whole Asom uh you know movement and and so on so so yeah you know it's it's a combination of many different areas and where attackers are leveraging the ease of use of the technology and also combining the different techniques for another example is reconnaissance I was going to ask you about that yeah open source intelligence right forget about active reconnaissance just open source intelligence getting information from certificate transparency or from H shown or from you know social media and so on if I vectorize that information so in other words you know create beddings on different things that I can gather from uh different places put into a vector database like like ChromaDB, Pinecone PGVector the whole bunch of them right and then create you know agents with you know prom templates and agents with line chain or Llama index and so on now I have an army of people doing reconnaissance for me and profiling in an individual or a company while I sleep yeah that's a significant difference of what we're seeing nowadays of not only the potential but you know of course the the capabilities of of these newer technologies to accelerate attacks. So that's is that escalating you seeing more and more of those kind of attacks. It will continue to escalate yeah it's unfortunately that that's the we also have to think about how we reinvent literally reinvent incd response yeah for some of these a lot of the the techniques will be the same right but a lot of them will be different and I'll give you a couple of examples that be great one is remediation yeah what is remediation of a model is it pull in the plug yeah right uh you don't put the plug in BGP right you bring down the internet but as you become um reliant and and you rely on this technology the remediation will be a little bit different yeah but then the other one is for Y in an AI model forensics will be a little bit different right and there's a lot of metrics for accuracy for bias and and and so on for security not a lot of them exist so if I'm an attacker and I manipulate weights and so on there may be things that we're missing from a forensic capability so that's an area of research the other one is a really big one is reproducibility yeah if I get a prompt and I give it to a model to your rag implementation and so on it may be that with the same prompt with the same context of the data you may get different results yeah right so whenever somebody comes to you and say hey investigate this new vulnerability cross ey request fery service ey request fery n or SQL injection you typically have a set of instructions of click here put this payload you know basically the recipe of an exploit for this new implementation even if you give that prescriptive thing and the same prompt you may get different results and if an attacker is manipulating things behind the scenes it can do a lot of ausc on the actual attacks against the AI system itself yeah are you finding that attackers are using their own open source AIS they're also using like ones that they can just pay for that are available H there's an assumption that it's going to be used in both right yep um I cannot sugarcoat it right I mean that just the Dual nature of Technology yeah if I use uh CLA or GPT y or Mistral an attacker is probably using it faster better than you know cheaper um it is it is reality yeah and then also the definition of an attacker right criminals to scam people versus nation state yeah against the critical infrastructure yeah the other thing is AI comparing the models inference and and so on and all these systems that we're putting now together they're becoming critical infrastructure yeah it's like um you know back in the day we we looked at the security of routing and networking and and we still do right I mean that that's you have to worry about that plus plus yeah so um you looked at you know bgb for example we just talking about it right so how to secure these implementations and make sure that there's no route manipulation attacks now you have to think about the critical infrastructures inference coming from a specific model whether it's chat GPT your you so the attacks against that infrastructure while they use you know the same technology to attack you y that's something that we also have to figure out you know as a industry right but going back to your question related to them using a specific models yes if you ask CH GPT to create an exploit they have really good guard rails right to prevent that you can bypass that with prompt injection you know and even indirect prompt injection attacks you know embedding some prompts in a PDF for example a lot of them White Rabbit Neo is a perfect example and I know there's a company now but it's an open source model is uncensor and there's plenty of them out there that people are fine-tuning and then taking you know those guard rails out or not putting any guard rails to then be able to create an exploit to them be able to actually do you know x y and z y if that is somebody like me right she or ethical hackers imagine the very res sourceful attackers you know the ones that actually a lot of money and infrastru so for sure you know that that's that's an area of concern yes so we've got AI being used to attack but obviously AI is being used to protect but I'm just thinking about opportunities for everyone who's watching we definitely don't want any of the people watching to become black hat we want to steer them on the right path so there are a lot of opportunities right to help companies secure AI absolutely so can perhaps you can talk about that there are tons of opportunities in there so one thing that I always recomend commend is go beyond the model okay it's not securing just the model and even regulation I'm not going to go into the regulation technical but in the in the technical aspect of it a lot of people are focusing in the models you know uh model theft attacks which we have to pay attention to but now as we try to build systems that then are using multiple models yeah we're introducing a lot of security you know deficiencies that's a good way to put it it is yes that are security 101 yeah for example if you're familiar with retrieval augmented generation everybody's talking about this right you use your data your sensitive documents you use an embedding model you know something like C here or open eyes you know embedding models and there's plenty of them out there you're converting your data your text your you know PDFs Etc into numbers basically put them into a vector database that embedding model in some cases they're licking information at of the corporation exactly and then people are putting guard rails just into the inference model right the other one is as every company every sector is asking their people how can we use AI y to accelerate the things that we have yeah whether it's security collaboration Etc and a lot of people are experimenting yeah and as I mentioned these things are getting abstracted in the way that even non-technical people are playing with that that they have never thought about access control yep list privilege yep and so on there are other areas of focus on where now that we're moving to an agent based era y right so using more agents instead of chatbots and co-pilots how can we determine what is the required privilege to an agent because pretty much everything is running as a roote nowadays yeah and um that is pair to a human or to an application yeah because these are now Emeral entities that are going to spun up you know based on specific criteria and it's not only about rag prod using a tool getting data from an API and so on and then they will disappear right and you have the your personas so that's another area that we should focus on how do we have the appropriate Access Control mechanisms how we have the the basic oneone L privilege applied to these systems plus the more sophisticated attacks you know how can we detect if a weights in a model has been manipulated and the bigger one is supply chain security right if you look at huging face if you're not familiar with huging face tell us biggest platform think about like the GitHub y of models yeah and hog face is the one that is is being mentioned the most yeah there are others right tensor flow has one and so on but definitely huging phas if you look at huging phas early this year I'm not talking about you know you know the way back machine that you can go back so if you go back and use the way back machine and look at huging face in the beginning of the year was probably something about 300,000 models I haven't checked today but definitely last week as I was giving a presentation about it it was 650,000 plus right wow so if you look at it you know amazing I love it technolog is moving so fast and and so on in some cases it's scary but uh always uh optimistic of Technology a lot of those models actually the majority of them have used very insecure framework like pickle right and you saw you know couple of months ago 100 models being taken down from hugging face because real life attackers manipulate them so now imagine if you're a software developer I found out about you know a specific model because of a research paper oh I want to try this it's in hogging pH yeah oh there's a new version because this person says that it's you know better examples Etc I grab it I put that into my infrastructure Y and what I can actually do so that supply chain security is really tough of mind right and even even if you look at inventory that's a bigger one most companies don't have a good way to do inventory of their implementations yeah and U even so I sit on the board of Oasis is a standards organization I'm the chair of this thing called the common security advisory framework U we use machine rual of security advisories and try to modernize the way that we do you know vulnerability management so on and um a lot of the conversations in there is how can we secure this supply chain right with ass bombs you know software build materials get a good inventory if you look at AI models luckily now there's a lot of progress in there but a few months ago I'm not talking about years months ago we didn't even have a good way to even capture the inventory of a model and people may think well it's the same as an sbom because it's software uh you have you know ESP bones for quite some time spdx is a um a standard for that another one is Cyclone DX from OAS so you have a good way to inventory uh the list of ingredients that make up software in an AI model it's a little bit different it's beyond that you have to worry about the sbom but you also have to worry about the training environment that the model was uh trained on yep also if that is manipulated or it's not approved by a company that's another thing that you want to worry about the one is what is the data AI is not AI without data that you use to train or fine tune or rag you know there's always the three elements you train from scratch super expensive you know only a few companies you know like Cisco and you know and the the rest you know are doing that fine-tuning a little bit more accessible still requires expertise and and infrastructure and raging yeah and um so looking not only the model but the whole system again what is the data what is the purpose of the implementation what is the training environment and then all the auxiliary libraries like L chain and so on that's part of the traditional esom so spdx in the Linux Foundation they just released they're about to release 3.0 that includes and profile for uh AI bombs right and um I work with a company called manifest yep that uh we we looked at you know they they they're pretty well known in the s bomb and we looked at you know what are the minimum requirements that you should have in inventory and then we also fed that back to spdx and Linux foundation and also OAS to Cyclone DX so now um within weeks you're actually going to see these uh standards now supporting at least a good way to have an inventory of U of AI implementations yeah sounds like a bit of a wild west right it is Wild West yeah it is it's it's worrying because I think it's you mean we you we've been in this game for a while the um it's this everyone wants to go to the cloud now everyone wants Ai and everyone's jumping on this and that's a that's a really good observation many many many people have asked me is this just Cloud again yeah right uh or even is the cloud dead because it's expensive you know I'm not going to go into that but but is this another Cloud do I have to worry about from security from another Cloud the your answer is no okay what one of the big differences of this technology is that everybody is now becoming accessible to everybody yeah like you said non-technical people non-technical people I I'll give you an example this is kind of behind the scenes Cisco and so I sit in an organization called security and Trust we are pretty large organization so you have the product security inent response team I Bel I used to belong to that team CER the traditional inent response team within the company yeah then we have The Insider threat team a very big offensive security team that we're tasking to find zero day vulnerabilities in Cisco products and yeah and services and other uh entities and this is not Talos this is be Beyond Talos yeah and then um and a whole bunch of other you know forens six teams Etc it's a pretty big organization about close to a th000 people Y and um within there kind of like my carrot was I want to do an ation challenge this is last year and everybody has to play with AI y whether you're at pm at intern you know and so on I made a little competition right and even though we have a function that is dedicated you know for AI security and everything but I want it for everybody to truly understand and think about how this technology can impact everything that we do yeah from doing forensics analysis into a machine to do an evaluation evaluation we call it evaluation but like a pen test or red team you know and how we augment that to programming how we can create better tools and so on a lot of people they open my eyes I was joking that if I have three teams it would be amazing and we had 27 teams you know 70 people participating Etc and what I was seeing is that non-technical people they were not afraid y to using these Technologies watching a video follow through putting l chain you know in a way and ask asking the questions that we all should have been asking 10 years ago what embeding model should I use it's like oh wow we don't have a strategy for that let me actually you know assign people to that most Vector databases don't even support encryption right crazy so it's my fact I was even talking here about how to secure Vector databases I just published a paper about it but again Vector databases is a newer type of databas is not new uh basically whenever you convert texts or images into numbers you put them a store that you can do similarity searches yeah many methods on that and then you have retrieval you know that's does the word retrieval augmented Generation Y retrial method reranking techniques Etc that you get the data provide some context and you reduce the likelihood of hallucinations yeah very attractive for many people yeah if I vectorize again from certificate transparency information in ENT to my security policies examples of vulnerabilities and so on I can then accelerate provide better support to an engineer that is either programming or fixing a vulnerability yeah these are other examples you know or potentially even do a pull request automatically plenty of use cases and a lot of these non technical people were putting into the asking the questions and the main value at and the benefit was for the security people you know for us yeah to and again these are also security folks but you know non technical roles asking these questions that again should have been asked 10 years ago in the industry y not at Cisco y right the whole industry um so it's very fascinating on how this of course is impacting you know many different areas around technology and um you know Healthcare and so on and how we should look at securing those implementations as well I mean to me it looks really Bleak like wild waste the data is out there I I've seen recently with slack like using customer data even if you're paying to you know train the AI uh they they didn't even ask for permission they just used it so from a privacy point of view AI bit of wild west it feels like from a cybercity point of view it seems really woring any hope there's hope there's absolutely hope um there are a lot of organizations across the industry that are asking the same questions good whenever I've been blessed in my career that if I'm fixing a vulnerability again using the bgp example because top of my head for some reason sometimes I have been on the phone more with juniper yeah than even Cisco trying to figure out the the how to solve it yeah and then give the information to Cisco and then do a coordination of vulnerability you know disclosure and so on you know for many years now I'm seeing the same whenever it comes to security a lot of security-minded organizations are trying to figure out what is going to be the guidance for not only today but three months from now six months from now a year from now and so on a lot of that guidance unfortunately has been more around regulations and so on but we're moving outside of that right that we're trying to see how these new best practices or the assisting best practices apply to these Technologies and creating Frameworks about it so um I'm blessed to be part of the board of Oasis and uh with anthropic Microsoft Nvidia Intel IBM of course Cisco uh mic oft we're all creating a Consortium for securing AI right and of course we're collaborating with nist with uh other agencies like cisa the NSA and so on and we're we're putting a lot of guidance in the industry but how to operationalize that guidance and create sample reference architectures and so on one of the original contributions is Google's the open siif it's a open secure AI framework change the name for the coalition because it got uh donated to Oasis or the IPR got transitioned and that's the foundation we kind of rebooting that right modernizing even though it was created a few months ago right and you may think that oh my god modernizing of something that yeah but is is creating a framework on how fast we can create this guidance as technology evolves yeah right and then examples of that give you the the the perfect example right now what is an AI vulnerability yeah all right what is an AI security vulnerability yeah in many cases you're saying okay so if a prompt injection is one for sure you know easy yeah H but prompt injection if you think about me chatting with a chat bot and saying that Omar is ugly does this blah blah blah who's the victim of that the person that is doing yeah the so is that a true vulnerability or not yeah but what if I have a far World co-pilot yep or you know an HR system that I can give a resume you know that's a typical example you give a rume with an indirect prompt injection so a prompt within the PDF with white funds that the human cannot read but the machine can read and then do some action or a firewall co-pilot that then I can inject some type of configuration or ofus skate a configuration and then manipulate the system right that absolutely is a an AI vulnerability but there's a lot of talks about an AI vulnerability or security vulnerability versus bias ethics and and we're also trying to get those taxonomies out of the way uh so everybody will be somewhat in the same page Y and then truly understand how we should address all of them right we definitely should address them but what security teams should be prioritizing what your AI governance and legal teams should be prioritizing and working together really more closely than ever and it's not because of regulations and everything else but it's because of the whole AI governance within the organization I'm talking about inventory right really knowing what your people are using where it goes you know what technology goes into products into the applications that you have whether again you're a Cisco of the world or um anthropics of the world or Health Care System right so so that's the the scope of the charter of that organization that's great from an industry point of view right I'm really glad to hear that but let's bring it back to the individual yeah if I want to ride this wave I always recommend everyone rides the new wave and AI is that Wave It's the hype it feels like in many cases at the moment what's your advice if I'm a young person or I want to get into this field what would you advise people to do yeah so one thing even outside of AI whenever people ask me hey kind how can I get started in ethical hacking always my guidance was to space yourself technolog is moving fast yeah fast forward until now technolog is moving a lot faster y don't get overwhelmed right uh think about what are again the non technical people asking questions that the really hardcore technical people should have asked 10 years ago y right so look how this is affecting the work that you do what are the things yes the way that we can communicate with computers and the we use computers will change the way that we program computers now it's been augmented it will also will change but how that technology you know you you have to always have some fundamental knowledge of how this works yeah the human in the loop is crucial nowadays yeah and yes there's some technical works that will evolve right I'll give you a good example many people whenever they start in cyber security they start in a security Operation Center in a sock yeah right the immediate or the the question that I get all the time immediately whenever I talk to the young folks and so on should I even go to that yeah am I going to get replaced if I go into that and the short answer is no right hey it is evolving so tier one analyst and again I'm going to give you a real life example not by the book tier one analyst right now in my organization is working with a Jupiter no and of course this plong of the world and everything else but you know you're going to get augmented and we have been doing this before the CHP era and so on with traditional machine learning and and uh what you're seeing is that the level of a tier one is becoming a tier two yeah because uh also the way that you learn is accelerating faster yeah the way that I learn now is is significantly different than before yeah and also has to adopt augment you know having your tutors Etc but looking into how can you de develop a program in place that you don't get burned out right again Technologies you know will continue to change but it's somewhat actionable to look at the preium right on how AI is solving some of those problems a lot of people actually trying to get AI and find the problem to in of looking at the we do need a lot of people technical people that really can spot the fod versus reality yeah that can also be augmented you know with AI start uh go beyond the model go beyond chpt try to understand you know how this technology can interact with other systems what are vector databases where how dimensionality of data uh works you know so for an AI model to actually do something right um and then of course you know um if you're going to ethical hacking how somebody can manipulate this yeah if you're going more into instant response how can I use the technology or an attacker can use that to ofus skate you know these Technologies and then how how do I do forensics better faster and and so on and unfortunately it's going to be using AI right yeah this augmenting using AI especially in this concept I don't know if you have heard of constitutional AI right tesc so constitutional AI was a concept anthropic you know had probably two and a half years ago and he basically using AI to monitor AI right uh after they introduced that um a lot of people thought about okay this is good of looking at a model observing a model can I use AI to observe the rest of the system right and uh the answer is yes and of course you know Cisco has technology and that like modific and and so on but looking like if you're getting started looking on how the techniques of doing instant response monitoring how um things that AI potentially have gaps that will never ever ever you will always have to have some type of human in the loop yeah you know as people are developing programs I will I will definitely you know encourage you your newcomers because it's very very overwhelming to look at those areas right and then of course you still have to have the foundations networking was not going to go away you know as a matter of fact it's going to evolve but the technology will continue to to start with the fundamentals in there even programming is a very hot topic of is it going to get replaced and so on if you are not familiar with the underlying technology on how even AI may be constructing something who's going to defend who's going to be able to do forensics who's going to be able to actually do truly instant responds after something is manipulated right so I don't think that you know security and programmers and everything is that is definitely change and look into the way that how technology changes and how those roles but the fundamentals always are going to be there right so do you is it still the same as in the past go and do uh CCNA or network plus can you perhaps give us like just a sort of like a a road map if you've got like just I'm starting at zero like what would I do if I started today cuz I want to jump in this wave right I don't want to like everyone I don't want to like spend 10 years getting to a certain level I want to try and do the quick way like the I would say the hack hack hack the way so have you got like sort of a a road map that I can follow yeah yeah absolutely so typically it all depends on the role right is so broad let's but let's focus on red teaming perhaps or like hacking I like that I like that so if you are thinking about ethical hacking definitely you have to have a foundation of networking yep do you have to pass a CCNA to actually go in there probably not yeah but CCNA will definitely give a blueprint yeah and for me what certifications do uh in many cases is is create a blueprint and push me that I have some achievement that I can say okay you know this is something that I'm now can I apply yeah most companies you are evolving into like requiring a specific certification and so on but at least that gives you a a really good blueprint yeah so of course you know in the case of networking CCNA you know is is the de facto uh other ones you know related to ethical hacking uh a lot of people try to jump from like C8 to OSP yeah you know pays yourself probably something like pentest plus yeah to get you the the whole methodology of hacking yeah and in some cases even the non-technical aspects of the ethical hacking in real world you have to deal with scope and uh you know the things that will keep you out of jail and in business right so those certifications will actually provide you that and then you know from there you pace yourself to the ocps of the world and there so many of them as a matter of fact um I have a repository we can make it available that we have tons and tons of certifications and and what I um always say is that look at the blueprints of the certification they're free you don't have to actually pay you know if a blueprint is asking the same concept two or three times that's something that you have to prioritize yeah for sure yeah right so that's a fundamental outside of networking even though programing is going to change you have to have some of a foundation on how applications actually are created yeah even if it's augmented by a python the typical you know even for data science python how I got started and this is all a little bit of a personal story great I didn't get started with the AI thing because of the CH GP er um a few years ago a good friend of mine is no longer a Cisco we wanted to uh do automated trading like literally we wanted you know but here we are I'm still here so I was not that successful in doing right which I'm glad about myself so but I I I went to a program you know two-year program with Harvard but it was a mini competition we looked at the blueprints of things you know of course there was no certification just of machine learning at the time but we're looking at what are the fundamentals that I need yeah at the time was R everybody was programming in R and looking at to how we can do back preparation and and how we can do a little bit more again all predictive you know models and um and then apis with like interactive brokers Etc right so that pushed me to then look at the fundamentals not of how you know machine learning was actually revolutionizing things what are the things that I need to know from a programming perspective right that is applied to that use case right in that case it was automated trading again that was many years ago a lot of that still applies here right whether is generative Ai and and so on and now of course python libraries make it so much easier but I would say go and try to figure out what are those libraries doing behind the scenes yeah and especially if you want to go into security especially if you want to go into ethical hacker and especially if you want to mimic what a real attacker is actually doing y right looking at the limitations of the technology and also the power of the technology yeah right uh but I I think that that will be my guidance to people that are getting started is there a AI cyber security type SE or is it do I read your book that that you've published what cuz that's the piece that seems to be missing right CU we talk about the basics but how do I do the AI cyber piece you will see a flow of you know uh training and uh certifications being as matter of fact Cisco just announced an AI uh certification uh for security currently as of today again may change tomorrow they're not that many right but what you're seeing is that assisting certifications and assisting programs whether it's for inant response for non-technical things like Risk um management and governance and so on they're starting to incorporate AI into it so how to use um even a GPT a very basic me GPT for accelerating uh the threat hunting and then getting information from the M attack framework for example and what you're seeing the certifications evolving the some of the assistant certifications evolving to include those topics so it is inevitable that they will become domains yeah within a certification and task for somebody to then probably be augmented by Ai and then knowing what you know lanen can do or or the library of the 2025 right yeah because it's going to continue to evolve so you're going to see yes new certifications current certifications evolving and even the way that we do training right even though we're talking about books and everything else in the way that I learn I use books for blueprints yeah and then since information is changing so fast I create my own scripts to actually get papers get of course AI to summarize the papers if it's related to cyber security I get a you know very basic alert you should read this on the Saturday because it's very overwhelm and this changes so fast and then you know you create that type of framework on okay yes this is new research but the areas of security are the same exactly what we were just talking about earlier a lot of security oneone things that are being omitted because people are experimenting so um yeah cyber security I I will not say that it's going to get obsolete it's going to get more complicated it's probably going to be more demand for cyber security than than the opposite Omar unfortunately they're chasing us we got to end I could keep you here for hours thanks so much absolutely really appreciate it so for everyone who's watching please connect I'll put Omar's links below if you've got any questions put them in the comments below hopefully we'll get you back for more more interviews absolutely thanks so much thank you [Music]
2024-07-23 15:16