Latest / Expert Insights Podcast / #97. AI Is Rewriting the Attacker Playbook: Juraj Janosik on Phishing, Automation, and Shadow AI
Transcript
- 0:00[music]
- 0:05SPEAKER_00: Hello and welcome to the Expert Insights Podcast.
- 0:08SPEAKER_00: I'm Caitlin Harris,
- 0:09SPEAKER_00: Head of Content at Expert Insights and your host for this episode,
- 0:13SPEAKER_00: where we're talking about how AI is and isn't changing the stakes for cybersecurity.
- 0:18SPEAKER_00: Here with me today is Urai Janashik, VP of AI at ESET.
- 0:23SPEAKER_00: Urai, thanks so much for joining me today.
- 0:26SPEAKER_00: I want to jump right into this topic because we have quite a lot of ground to cover today.
- 0:31SPEAKER_00: The NCSC says that AI will almost certainly increase both the volume and the impact of attacks.
- 0:37SPEAKER_00: Where are you seeing AI make an attack genuinely more damaging rather than just more frequent?
- 0:43SPEAKER_01: Yeah, so definitely I can confirm this.
- 0:46SPEAKER_01: We see the same in our data, in our telemetry.
- 0:50SPEAKER_01: So what's maybe interesting is that, yeah,
- 0:53SPEAKER_01: we are in the era of AI,
- 0:56SPEAKER_01: but generally still the most typical infection vector or the way how
- 1:01SPEAKER_01: the attackers made the way in is still the email.
- 1:04SPEAKER_01: And thanks to the AI,
- 1:05SPEAKER_01: what actually happened is that the whole quality of the business email compromise,
- 1:10SPEAKER_01: the fraud, the phishing, the scam, got better.
- 1:13SPEAKER_01: So now you don't see the clumsy translations or clumsy English,
- 1:18SPEAKER_01: but it's even in the quality of an English professor.
- 1:23SPEAKER_01: But it's not only in English,
- 1:25SPEAKER_01: but the change is significant in the local languages,
- 1:28SPEAKER_01: which was back in the days kind of like a problem.
- 1:32SPEAKER_01: We see really high quality of language localized scams and frauds.
- 1:39SPEAKER_01: And of course,
- 1:40SPEAKER_01: connected with that,
- 1:41SPEAKER_01: deepfakes,
- 1:41SPEAKER_01: which are trying to lure people from social media or messaging platforms
- 1:50SPEAKER_01: to actually landing page where they're trying to get some personal information
- 1:54SPEAKER_01: from them or basically get the money.
- 1:59SPEAKER_01: I would say this is definitely the most significant thing where we see
- 2:04SPEAKER_01: the AI actually improving the quality and the volume of the attacks.
- 2:09SPEAKER_01: It's important to say that thanks to the AI tooling,
- 2:13SPEAKER_01: which is more affordable,
- 2:15SPEAKER_01: more powerful,
- 2:16SPEAKER_01: the barrier of entry-level attackers,
- 2:20SPEAKER_01: it's basically now shifted more to techniques that we before see only
- 2:27SPEAKER_01: with like APT groups, so the most advanced attacker state sponsors.
- 2:31SPEAKER_01: So this will be definitely a problem for future, and some of it is already manifesting.
- 2:39SPEAKER_00: Absolutely, absolutely.
- 2:41SPEAKER_00: So I do want to talk about that barrier for entry a little bit more in a minute.
- 2:45SPEAKER_00: But just before we do, let's touch a bit more on that social engineering aspect.
- 2:51SPEAKER_00: If these attacks are becoming so much more difficult to detect,
- 2:55SPEAKER_00: what does a convincing AI generated social engineering attack or
- 3:01SPEAKER_00: a voice clone attack look like?
- 3:03SPEAKER_00: And are there any signals that still give it away when they are so convincing?
- 3:07SPEAKER_00: Is there anything that people can still look out for?
- 3:10SPEAKER_01: Yeah, definitely.
- 3:11SPEAKER_01: So it's not only in this case,
- 3:15SPEAKER_01: many times it's not only about the technology,
- 3:17SPEAKER_01: but it's more about like the training and education to bring the critical thinking
- 3:23SPEAKER_01: to the equation and look at what actually is shown to me if it's somehow correlated
- 3:29SPEAKER_01: with the reality.
- 3:30SPEAKER_01: If the person on the other side could be actually saying that type of thing.
- 3:35SPEAKER_01: We had several questions.
- 3:36SPEAKER_01: Several investment frauds using deepfakes in the recent months.
- 3:43SPEAKER_01: Probably what he also knew was deepfake with Elon Musk.
- 3:47SPEAKER_01: But there were many, many others which kind of...
- 3:53SPEAKER_01: But the thing is that you need to look into it if this will be something that
- 3:57SPEAKER_01: the person will be saying.
- 3:59SPEAKER_01: So kind of like look into from this perspective,
- 4:04SPEAKER_01: but also many times it ends on And the landing page
- 4:09SPEAKER_01: where it actually brings you this type of message is many type like looking
- 4:15SPEAKER_01: that it's pushing like in a sense of urgency that it's trying to like force you
- 4:20SPEAKER_01: that something needs to be like done right now.
- 4:22SPEAKER_01: Otherwise, you won't earn money or something bad actually happen.
- 4:27SPEAKER_01: So immediately when there is like a sense of urgency,
- 4:30SPEAKER_01: I recommend to actually like stop and wait for Couple of minutes and reffing
- 4:34SPEAKER_01: if it's kind of live real.
- 4:37SPEAKER_01: And precisely the technologies of the generating of the videos and photos, it's getting better.
- 4:47SPEAKER_01: You need less time and less recording of the person to actually create
- 4:53SPEAKER_01: a really good looking deep fake.
- 4:56SPEAKER_01: A photo is definitely easier.
- 4:59SPEAKER_01: For video, yeah, it took much more time.
- 5:01SPEAKER_01: But this actually got better in how these technologies are improving.
- 5:07SPEAKER_01: So there won't be things to look into the video or photo that you can actually catch
- 5:14SPEAKER_01: and see that This is fake.
- 5:17SPEAKER_01: So rather focusing on the content.
- 5:19SPEAKER_01: And for us,
- 5:21SPEAKER_01: we're not trying to,
- 5:22SPEAKER_01: with another AI,
- 5:24SPEAKER_01: identify whether something was generated with AI,
- 5:27SPEAKER_01: but rather we are looking on those landing pages or the messages itself as
- 5:33SPEAKER_01: a security vendor and trying to evaluate them
- 5:37SPEAKER_01: if there is not something like malicious or phishing going on and then block it for
- 5:42SPEAKER_01: the users.
- 5:43SPEAKER_00: Absolutely.
- 5:44SPEAKER_00: So it's really interesting that although the attacks are becoming
- 5:48SPEAKER_00: so much more realistic, a lot of those principles are still the same.
- 5:52SPEAKER_00: Analyze the context.
- 5:54SPEAKER_00: Could the person,
- 5:55SPEAKER_00: does it make sense for them to be asking that or if you just take the time to think
- 5:59SPEAKER_00: about it?
- 6:00SPEAKER_01: I would say that generally the thing is that the attackers are using
- 6:05SPEAKER_01: the same playbook they were using like five or 10 years ago.
- 6:09SPEAKER_01: It's basically the aspects are still the same,
- 6:13SPEAKER_01: but thanks to AI, it got accelerated and automated.
- 6:16SPEAKER_00: Absolutely.
- 6:17SPEAKER_00: Absolutely.
- 6:18SPEAKER_00: And because of that and something that you touched on a little bit earlier,
- 6:22SPEAKER_00: that barrier to entry is lowering for attackers.
- 6:27SPEAKER_00: But would you say that the bigger risk there is
- 6:29SPEAKER_00: that unskilled attackers are becoming more capable of doing these attacks or that
- 6:35SPEAKER_00: the groups that are more skilled are becoming more efficient and able
- 6:38SPEAKER_00: to carry out more attacks more quickly?
- 6:40SPEAKER_00: Absolutely.
- 6:41SPEAKER_01: Yeah, definitely.
- 6:42SPEAKER_01: Definitely it's both.
- 6:43SPEAKER_01: I think that's precisely they are trying to accelerate the attacks,
- 6:49SPEAKER_01: rise the volume of the attacks and automate as much as possible.
- 6:54SPEAKER_01: Same as what we are doing on the other side.
- 6:56SPEAKER_01: On the side of the defenders,
- 6:57SPEAKER_01: yes, we are automatizing sub-operations, processing of the samples, and so on.
- 7:03SPEAKER_01: The same way, they are basically doing the same thing.
- 7:08SPEAKER_01: And that's, of course, lowering the barrier to perform the attack faster.
- 7:16SPEAKER_01: When we look, for example, on the ransomware attacks, something which is...
- 7:22SPEAKER_01: Typically during the investigation of some ransomware attack count, it is like the dwell time.
- 7:27SPEAKER_01: So it's basically the time when the attacker got inside and
- 7:31SPEAKER_01: the data was either exfiltrated or encrypted.
- 7:34SPEAKER_01: And we see that in many cases right now, Now,
- 7:38SPEAKER_01: the time got really shrink to just a couple of hours or many times less
- 7:44SPEAKER_01: than two hours.
- 7:45SPEAKER_01: And that's mainly thanks to the automation, which is supported by the AI.
- 7:52SPEAKER_01: We see the scenarios.
- 7:54SPEAKER_01: Many times it's shown also on the dark forums that you can really quickly right now,
- 8:00SPEAKER_01: thanks to the availability of the tools,
- 8:04SPEAKER_01: even like a script kiddie,
- 8:06SPEAKER_01: basically somebody who doesn't know what he's actually doing,
- 8:10SPEAKER_01: if he asks the right question,
- 8:12SPEAKER_01: the AI actually learns him how to perform the attack, build the tools for him.
- 8:17SPEAKER_01: So this is the reality where we are living right now.
- 8:21SPEAKER_00: So what does that mean for the defenders?
- 8:24SPEAKER_00: Is there any hope in the face of all of these really sophisticated attacks coming
- 8:30SPEAKER_00: at us so quickly?
- 8:31SPEAKER_01: Yeah, so we still have like many, many opportunities.
- 8:36SPEAKER_01: It's not like that right now the attackers has tool in their hands
- 8:41SPEAKER_01: that we couldn't actually make any difference against.
- 8:44SPEAKER_01: It's not that good.
- 8:46SPEAKER_01: So typically what needs to be done is that you not just rely on solutions
- 8:52SPEAKER_01: that are AI powered and only AI powered,
- 8:55SPEAKER_01: but you need to have solutions which are like multilayers.
- 8:59SPEAKER_01: And with proper security telemetry,
- 9:01SPEAKER_01: observability,
- 9:02SPEAKER_01: visibility tools, remediation inside, like proper XDR, but with a human in the loop.
- 9:10SPEAKER_01: That's the important thing.
- 9:12SPEAKER_01: And now,
- 9:12SPEAKER_01: of course,
- 9:13SPEAKER_01: AI engines or tools which are looking for the misconfigurations in the network and
- 9:20SPEAKER_01: so on.
- 9:20SPEAKER_01: So basically,
- 9:21SPEAKER_01: the thing is that Many of those addressing of those attacks could be done with tools
- 9:31SPEAKER_01: that we already have in our hands.
- 9:35SPEAKER_01: And maybe one of the things is that Thanks to things like mythos
- 9:40SPEAKER_01: and these more powerful models,
- 9:43SPEAKER_01: the discussion about vulnerabilities again got in the center.
- 9:48SPEAKER_01: And yeah, those models are actually not giving us a way how to solve.
- 9:52SPEAKER_01: And in many attacks, it's not about the zero days, actually.
- 9:56SPEAKER_01: It's about leveraging the known unpatched vulnerabilities.
- 10:22SPEAKER_00: Absolutely.
- 10:23SPEAKER_00: Absolutely.
- 10:23SPEAKER_00: A lot of it comes down to those sort of core fundamentals.
- 10:26SPEAKER_00: The attackers are going to keep targeting them as long as people keep forgetting to patch them.
- 10:32SPEAKER_00: So a lot of the AI risk conversation focuses on attackers and the kind of attacks
- 10:37SPEAKER_00: that we've been talking about so far.
- 10:39SPEAKER_00: But I also want to talk about the risk that comes from internal AI use.
- 10:44SPEAKER_00: So employees pasting sensitive data into chatbots or using unsanctioned tools
- 10:49SPEAKER_00: and Agents with standing access, that kind of thing.
- 10:53SPEAKER_00: How dangerous are those internal risks?
- 10:56SPEAKER_01: Yeah, so in this era of many,
- 10:59SPEAKER_01: many companies trying to adopt AI as fast as possible to make themselves on
- 11:06SPEAKER_01: the cutting edge and being more effective and so on, the security is many times...
- 11:12SPEAKER_01: Not in this equation.
- 11:14SPEAKER_01: And people are many times like using the things that they know.
- 11:20SPEAKER_01: And in case that the company doesn't have like proper AI adoption or they're like
- 11:25SPEAKER_01: in Slow pace of adopting tools from renowned vendors,
- 11:32SPEAKER_01: they try to use whatever they have in the hands to be more effective.
- 11:36SPEAKER_01: And precisely then we see that private data are leaked to online chatbots and these platforms.
- 11:45SPEAKER_01: So, yes,
- 11:46SPEAKER_01: we are looking also as a company to this shadow AI,
- 11:51SPEAKER_01: is how it's called,
- 11:52SPEAKER_01: like using of the tools,
- 11:54SPEAKER_01: which are many times like sanctions,
- 11:56SPEAKER_01: sanctioned within the company or the employees shouldn't be using it
- 12:00SPEAKER_01: because there is a potential of data leakage and it's not properly controlled
- 12:06SPEAKER_01: and audited.
- 12:06SPEAKER_01: The only way to that, again, is like employee training.
- 12:10SPEAKER_01: And having the visibility and auditing tools in place.
- 12:15SPEAKER_00: Absolutely.
- 12:16SPEAKER_00: Absolutely.
- 12:17SPEAKER_00: Is there also perhaps an element there of having a conversation with your employees?
- 12:23SPEAKER_00: What are you using these tools for and providing a safe alternative, I suppose?
- 12:28SPEAKER_01: Yeah.
- 12:28SPEAKER_01: So every aspect of AI adoption, right now the pace is really fantastic.
- 12:34SPEAKER_01: There is a gazillion of tools we just discussed on social media.
- 12:40SPEAKER_01: But if you are a bigger company,
- 12:43SPEAKER_01: my recommendation is rather look for proper tools from serious vendors.
- 12:49SPEAKER_01: Yeah, it's not like let's not use something that was produced by a startup.
- 12:54SPEAKER_01: I'm not saying that,
- 12:55SPEAKER_01: but let's be reasonable about the security and look at how the tools are built,
- 13:02SPEAKER_01: who is actually building them,
- 13:05SPEAKER_01: what are the terms of use,
- 13:07SPEAKER_01: terms of service,
- 13:08SPEAKER_01: and where actually the tool is storing the data or how much data are processed.
- 13:13SPEAKER_01: So these are my recommendations how to look into that.
- 13:17SPEAKER_00: Yeah, that's some really great advice.
- 13:19SPEAKER_00: I think that's something that a lot of people struggle with is which questions
- 13:23SPEAKER_00: to ask your AI tool vendor to make sure that they are actually reputable.
- 13:27SPEAKER_00: So yeah, that's really, really helpful.
- 13:30SPEAKER_00: We've also seen recently a few headlines where AI agents that can take actions
- 13:35SPEAKER_00: and not just generate text.
- 13:37SPEAKER_00: Have been manipulated or have behaved unpredictably.
- 13:41SPEAKER_00: What can businesses do to protect themselves against that kind of behavior?
- 13:47SPEAKER_01: It's very similar to this kind of tooling.
- 13:53SPEAKER_01: So you need to have the visibility and way how to audit the decisions made by the AI.
- 14:00SPEAKER_01: And the second thing is that it's also about identity and access management.
- 14:06SPEAKER_01: Because the AI agent is just another user,
- 14:10SPEAKER_01: another identity person that you have in the company.
- 14:13SPEAKER_01: So imagine you have a thousand people within the company.
- 14:17SPEAKER_00: Everybody has the user accounts as a right.
- 14:19SPEAKER_01: You can audit the logs and see where was any of those employees,
- 14:26SPEAKER_01: which data has been accessed or where he was actually trying to connect.
- 14:31SPEAKER_01: Same things needs to be done with the AI engine.
- 14:33SPEAKER_01: So you need to treat or handle basically AI engines in the same way
- 14:38SPEAKER_01: as you handle like standard user accounts.
- 14:42SPEAKER_01: So kind of off the back of that,
- 14:44SPEAKER_00: according to IBM,
- 14:45SPEAKER_00: 63% of organizations reporting
- 14:48SPEAKER_00: an AI-related security incident didn't have any AI governance policy in place.
- 14:54SPEAKER_00: So why do you think AI governance is lagging so far behind that adoption curve?
- 15:00SPEAKER_00: And where is the biggest gap between what regulation is currently asking for and
- 15:05SPEAKER_00: where the actual AI risk sets?
- 15:08SPEAKER_01: So with every technology which is in such an evolutionary pace as AI
- 15:16SPEAKER_01: and transformative, the security is always lagging a little bit behind.
- 15:22SPEAKER_01: So the number of the tools and the pace of changes and new models,
- 15:28SPEAKER_01: more powerful,
- 15:30SPEAKER_01: We are getting in this situation that we are trying to fix the problem afterwards,
- 15:35SPEAKER_01: which is, of course, not ideal.
- 15:37SPEAKER_01: Right now, yeah,
- 15:38SPEAKER_01: we see that there are many voices about slowing down it a little bit because, yeah,
- 15:44SPEAKER_01: and it's like,
- 15:45SPEAKER_01: A good opinion or good situation that something like this is happening,
- 15:49SPEAKER_01: that finally we can stop a little bit and look into what's actually the problems
- 15:57SPEAKER_01: and how to address them.
- 15:59SPEAKER_01: When we look at what actually happened recently,
- 16:02SPEAKER_01: that many of those AI agents broke out and basically performed
- 16:07SPEAKER_01: a cybersecurity attack,
- 16:09SPEAKER_01: so this is a situation where in If it happened to any other company
- 16:15SPEAKER_01: or somebody else will do it, this will be considered a felony, right?
- 16:18SPEAKER_01: So we need to treat these incidents like really serious because just saying
- 16:25SPEAKER_01: that this was because of AI, that's There's not a valid argument, right?
- 16:30SPEAKER_01: There is somebody responsible for that.
- 16:32SPEAKER_01: If you are testing your agents and you say that,
- 16:36SPEAKER_01: well, it just got out, that means that there were no proper handling around.
- 16:42SPEAKER_01: And it can be done.
- 16:43SPEAKER_01: But if the pace of development is really fast,
- 16:46SPEAKER_01: typically things like guardrails and protections are lagging behind.
- 16:51SPEAKER_01: So yeah, I think that regulation will be important in this.
- 16:57SPEAKER_01: We need to be really talking.
- 16:58SPEAKER_01: I don't think that actually this is a problem.
- 17:13SPEAKER_01: This is a hard thing to say.
- 17:16SPEAKER_01: There is always somebody who is responsible.
- 17:20SPEAKER_01: Let me give you an example.
- 17:23SPEAKER_01: If you were running some cybersecurity product,
- 17:26SPEAKER_01: if there is a false positive or Or maybe a miss in the detection.
- 17:30SPEAKER_01: You turn yourself to the vendor because he is responsible.
- 17:33SPEAKER_01: And whether the vendor is using some AI tools to do that, no matter what, he is responsible.
- 17:39SPEAKER_01: And we need to be treating the AI engines in the same way.
- 17:43SPEAKER_01: There is always somebody somewhere who needs to be taking the responsibility for
- 17:49SPEAKER_01: the actions that have been taken.
- 17:51SPEAKER_01: Or there needs to be guardrails and systems to evaluate and audit.
- 17:58SPEAKER_01: Absolutely.
- 18:01SPEAKER_01: Absolutely.
- 18:02SPEAKER_00: So just in your opinion,
- 18:04SPEAKER_00: do you think that those guardrails need to be imposed at sort of
- 18:08SPEAKER_00: a federal government level or should they be down to the industry to come up
- 18:13SPEAKER_00: with those and agree to them?
- 18:15SPEAKER_01: I would say there needs to be definitely some regulations from government level,
- 18:22SPEAKER_01: but with the industry in the talk.
- 18:26SPEAKER_01: In the same way that cybersecurity is focused,
- 18:30SPEAKER_01: Four years working together with law enforcement and government bodies.
- 18:36SPEAKER_01: So this needs to be treated in the same way.
- 18:40SPEAKER_01: We can't be looking at the companies just because now there is a huge AI hype
- 18:45SPEAKER_01: that they are kind of like a special.
- 18:47SPEAKER_01: This is like a technology that needs a proper treatment.
- 18:51SPEAKER_01: But of course,
- 18:51SPEAKER_01: it needs to be in the hands of the industry to actually come with the guardrails
- 18:57SPEAKER_01: and proper security.
- 18:59SPEAKER_01: So if we are really talking about making democratization of AI and the availability
- 19:04SPEAKER_01: to almost everybody,
- 19:06SPEAKER_01: this needs to be also open to cybersecurity community to actually bring them in
- 19:12SPEAKER_01: and help with securing of it.
- 19:14SPEAKER_01: I would say let's don't make the same mistake as with social networks.
- 19:19SPEAKER_01: Yeah, absolutely.
- 19:20SPEAKER_00: Absolutely.
- 19:21SPEAKER_00: So in order to address some of the challenges that we've been discussing today,
- 19:25SPEAKER_00: ESET has been making major news this year.
- 19:28SPEAKER_00: Could you break down some of your priorities for AI innovation at ESET
- 19:32SPEAKER_00: and what is setting ESET apart in this space?
- 19:35SPEAKER_01: Yeah, so we announced 14 million euro investment in AI security.
- 19:42SPEAKER_01: So yeah,
- 19:43SPEAKER_01: somebody might say like 40 million compared to billions that are pouring in
- 19:49SPEAKER_01: to other like AI The thing is that we are a security company.
- 19:54SPEAKER_01: So we are not building a product that will be used by millions of people to chat
- 20:02SPEAKER_01: with the AI or build some agents or code.
- 20:06SPEAKER_01: So this is something different.
- 20:08SPEAKER_01: We are working on technologies, how to actually make usage of the AI secure.
- 20:14SPEAKER_01: So this, our activity, and it's not like starting right now.
- 20:21SPEAKER_01: We have been using AI in our products for a lot of years.
- 20:24SPEAKER_01: We were actually like one of the first companies back in 1997
- 20:28SPEAKER_01: that has been incorporating neural networks in the product for protection
- 20:32SPEAKER_01: against macroviruses.
- 20:33SPEAKER_01: And this is from...
- 20:35SPEAKER_01: From that time on,
- 20:36SPEAKER_01: we are still looking at the technologies that could help us as defenders
- 20:42SPEAKER_01: to protect our users against the threats,
- 20:46SPEAKER_01: but also right now with the generative AI,
- 20:50SPEAKER_01: the A new perimeter that needs to be also protected has emerged.
- 20:55SPEAKER_01: The AI is a new perimeter and we are looking and we want to bring technologies
- 21:00SPEAKER_01: to protect particular issues that are emerging right now.
- 21:05SPEAKER_01: The conversation,
- 21:06SPEAKER_01: the interactions between the user and the model to have security there in place.
- 21:12SPEAKER_01: Also, in the agentic platforms, we already released several extensions in our products.
- 21:20SPEAKER_01: To deliver that.
- 21:22SPEAKER_01: And on the side of the more enterprise is to build,
- 21:28SPEAKER_01: we are building a proper,
- 21:29SPEAKER_01: let's say, automated security operations on top of our XDR products.
- 21:35SPEAKER_01: Together with intelligence.
- 21:38SPEAKER_01: So threat intelligence that we actually collect and process for a lot of years,
- 21:43SPEAKER_01: but make it more,
- 21:45SPEAKER_01: let's say,
- 21:45SPEAKER_01: AI powered in a way that it could be plugged in the novelty systems
- 21:50SPEAKER_01: that are built right now.
- 21:51SPEAKER_00: Absolutely.
- 21:52SPEAKER_00: So it's kind of...
- 21:53SPEAKER_00: Combining a lot of those different methods that we have talked about today,
- 21:59SPEAKER_00: the human part,
- 22:00SPEAKER_00: knowing that context and the threat intelligence,
- 22:03SPEAKER_00: but also having that technology to actually identify and get to the root
- 22:07SPEAKER_00: of these attacks.
- 22:08SPEAKER_01: I want to emphasize to still keep the human in the loop in all of this.
- 22:13SPEAKER_00: Absolutely, absolutely.
- 22:15SPEAKER_00: I think that's probably quite reassuring for a lot of people.
- 22:18SPEAKER_00: It does feel sometimes like the technology is taking over everything.
- 22:22SPEAKER_00: So it might be quite nice for people to hear that, yes,
- 22:25SPEAKER_00: the human in the loop is still important and still important to be a part of
- 22:30SPEAKER_00: that conversation.
- 22:32SPEAKER_00: So I've just got one last question for you today.
- 22:35SPEAKER_00: And I would like to get your final words of wisdom for small businesses
- 22:40SPEAKER_00: that are worried about AI.
- 22:42SPEAKER_00: Are there any parts of the AI threat picture that you think
- 22:45SPEAKER_00: the industry might be overstating a little bit?
- 22:49SPEAKER_00: And where should they start when it comes to protecting themselves against both some
- 22:53SPEAKER_00: of those internal and external threats?
- 22:55SPEAKER_01: Yeah, I will say that what is definitely overestimated kind of or overstating
- 23:01SPEAKER_01: that we think about AI in a way that a lot of people put a lot of trust
- 23:08SPEAKER_01: in this system.
- 23:09SPEAKER_01: So to To be honest here,
- 23:12SPEAKER_01: the generative AI of right now that we have,
- 23:15SPEAKER_01: when we look deep inside,
- 23:17SPEAKER_01: the technology on top of which it is built is not like 100% accurate.
- 23:23SPEAKER_01: That needs to be like every time.
- 23:25SPEAKER_01: When we do anything with generative AI tools or processes,
- 23:31SPEAKER_01: needs to be kept in mind that these systems are not 100% accurate.
- 23:37SPEAKER_01: And for smaller companies, let's not be...
- 23:40SPEAKER_01: I'm really worried about AI threats because this is really something
- 23:45SPEAKER_01: that is currently really niche.
- 23:47SPEAKER_01: What is really happening is that the old attacks are again getting new,
- 23:52SPEAKER_01: but just really accelerated and made, let's say, in better quality with AI.
- 24:01SPEAKER_01: So every measure that...
- 24:04SPEAKER_01: That we have and every gap that we know,
- 24:07SPEAKER_01: every misconfigurations that we have or any lacking of the backup that is there,
- 24:14SPEAKER_01: if this will be fixed,
- 24:16SPEAKER_01: then it's definitely a step forward generally in the security in this AI era.
- 24:22SPEAKER_00: Absolutely.
- 24:24SPEAKER_00: I think that's some great advice to really focus on those security fundamentals.
- 24:29SPEAKER_00: Take a step back.
- 24:30SPEAKER_00: Think, okay, yeah.
- 24:32SPEAKER_00: Thank you for joining me on the podcast today.
- 24:34SPEAKER_00: I've really, enjoyed speaking with you.
- 24:36SPEAKER_01: Thank you.