TRANSCRIPCIÓN 🔗
WEBVTT
1
00:00:00.000 --> 00:00:14.540
[upbeat music] Hi, everyone. I'm Jess. I'm Tom. Welcome to the third episode of the Fast Track by Fastly.
2
00:00:14.780 --> 00:00:23.640
As you know, October is the Cybersecurity Awareness Month, and this episode is all about security and prevention. It's an important topic, I would argue.
3
00:00:23.700 --> 00:00:31.680
I mean, everybody nowadays is talking about AI, and this, I would say the second most important topic in every single conference everywhere that I go to is actually cybersecurity.
4
00:00:31.960 --> 00:00:38.680
So I think I'm just gonna start and ask you some questions that I'm pretty sure everyone is really interested to know about.
5
00:00:39.240 --> 00:00:50.780
For the listeners who hear zero trust everywhere, what does it actually mean in the practice, and how does it apply beyond network access? I mean, you mentioned a couple of concepts there that are important.
6
00:00:50.820 --> 00:00:58.420
I mean, zero trust, by definition, means trust no one, so that's just a literal translation of what, what kind of the idea was behind it.
7
00:00:58.440 --> 00:01:02.960
And but to be, to be completely aware of the term, we need to go a little bit backwards.
8
00:01:03.020 --> 00:01:15.620
I mean, for a lot of the people who are going to be listening to us are actually born in the cloud, so they don't know much about the world as it was before the cloud computing was a standard, as it is nowadays, or as it probably will be in the future.
9
00:01:15.660 --> 00:01:26.260
We'll see about that. The role was revolving around, around local area network that were part of absolutely every corporation, every organization. So kind of the security was always around the fortress approach.
10
00:01:26.380 --> 00:01:37.940
So anything that's outside of the fortress cannot get in, and we need to make sure that no one can get in. But all the people that are inside are to be trusted because they are inside of our organization.
11
00:01:37.980 --> 00:01:45.800
They are inside of our domain of control. So that actually means that it's... it should be kind of like easier approach towards those devices.
12
00:01:45.980 --> 00:01:56.700
Usually, that was, that was meant for the devices connected to the physical corporate network that should have a little easier access to the services than when you are approaching from the outside in.
13
00:01:57.100 --> 00:02:05.800
So basically, we would be blocking all the traffic that could enter our little fortress, and we would enable only the users that are inside to access the services.
14
00:02:06.340 --> 00:02:14.640
If, for some reason, somebody outside of the fortress needed to access our services, that would require us to set up a special way of connecting.
15
00:02:14.940 --> 00:02:21.740
That's how the concept of virtual private networking actually came to be or the VPN that everybody's familiar nowadays with the VPN term.
16
00:02:22.100 --> 00:02:32.860
That's what
Load full transcript
we started to use to enable people from the outside world that are not part of our internal network to actually access certain services. So at that point in time, even email was protected like that.
17
00:02:32.880 --> 00:02:40.080
You couldn't have an email on your phone or on your personal device. The only way to access the email would be when you are at work from your laptop.
18
00:02:40.420 --> 00:02:50.380
That would be a thing nowadays, you know, no access to email outside of working hours. Obviously, at some point, we needed to, we needed to do a little bit different approach, so that's we- when we created VPN.
19
00:02:50.480 --> 00:02:53.780
But this w- that was still the old approach, so the fortress approach.
20
00:02:54.300 --> 00:03:14.280
Now, due to the fact that there is no longer our fortress because a lot of the services that we are using are no longer under our control and are practically outside, so we needed to open that door and actually allow some of the software as a service applications like, I don't know, Gmail or Outlook or you name it, your Google Drive or Dropbox.
21
00:03:14.360 --> 00:03:22.980
Whichever of the services you're using, you needed to enable your employees to actually access those services. And due to that fact, you were starting to open your door slowly but surely.
22
00:03:23.560 --> 00:03:33.860
So then your fortress was no longer a fortress but was completely open space where there was wireless access that in- that was introduced in that at, at some point in time, so no longer just wired connection.
23
00:03:34.140 --> 00:03:44.060
We have wireless too. So wireless is something that you cannot protect like you can your local network. Local network is a physical cable, so somebody needs to be physically present there to connect to it.
24
00:03:44.380 --> 00:03:54.240
Wireless goes a little bit beyond your space, and the more you, the more, the more, uh, employees you have, the more different wireless devices you have, you're going to have stronger wireless systems. So they're,
25
00:03:55.200 --> 00:04:04.860
they're spreading around. So all of these things came at, not I wouldn't say simultaneously. They were building through the, through the decades of, uh, of internet access.
26
00:04:05.200 --> 00:04:20.540
And at certain point in time, we figured out that the only way that we could approach the security is no longer talking about firewalls on the, on the edges and trusting the internal users because you can never know who is anymore an internal or an external user.
27
00:04:20.940 --> 00:04:27.320
You need to configure. You need to approach your security from the point of trust no one, which basically means check everything.
28
00:04:27.720 --> 00:04:37.900
Whether it's a corporate device inside of the network or it's an actual mobile phone connected to some coffee shop wireless somewhere outside of your country maybe, doesn't matter.
29
00:04:38.080 --> 00:04:41.900
You still need to check everyone and every time and every single way.
30
00:04:42.280 --> 00:04:54.940
Everybody needs to accommodate the certain levels of security that are actually estimated or that are ac- that are, that are set by the administrators for the particular use case, for the particular application, by the particular user.
31
00:04:55.460 --> 00:04:59.160
That's what zero trans- zero trust essentially means. That's really interesting.
32
00:04:59.660 --> 00:05:13.920
Um, and when we think about it, uh, why is the traditional internal perimeter no longer a reliable security boundary in cloud, hybrid work, and SaaS environment? Because in SaaS envi- environment, what is internal?
33
00:05:14.400 --> 00:05:18.120
You're always accessing somebody- That's true... else's network. So there's no internal way.
34
00:05:18.220 --> 00:05:30.700
So you could technically say that your IP range, for example, if you have a local network for your company, and you have the dedicated IP range for that, then you could technically say, "My IP range is my local network, so that should be trusted."
35
00:05:30.760 --> 00:05:36.140
But why would it be trusted? Because at the end of the day, you are most, most of the time, you're actually not in that network.
36
00:05:36.200 --> 00:05:41.600
You might be even connected through the VPN and then having some kind of IP, but you are somewhere outside of that.
37
00:05:42.020 --> 00:05:50.280
So there was, there's actually no reason anymore to approach it from the point of, okay, I'm gonna trust just this network. Because why that network and not some other network?
38
00:05:50.880 --> 00:05:58.980
You are accessing some external system by definition. So that's one thing. Even though there are software and service solutions that are respecting still that kind of a point.
39
00:05:59.280 --> 00:06:07.160
So for example, you can have different type of authentication, authorization mechanism, authentication mechanism predominantly based on the network you're connecting from.
40
00:06:07.240 --> 00:06:12.060
So if I'm connecting from the network that's trusted, I would just need to provide, for example, username and password.
41
00:06:12.340 --> 00:06:21.288
While the, if I'm connecting outside of that network, I should also do some kind of MFA, you know, approve something in my authenticator or Receive a text message or something like that.
42
00:06:21.328 --> 00:06:30.408
So there are th- those approaches that are kind of somewhere in the middle. But technically speaking, for the software as a service, that their own network is the only local network for them.
43
00:06:30.468 --> 00:06:42.328
Everything else is the, is the external network. So that's one thing. Second thing is the attackers and the way that, that the viruses or all those potential malware are spreading,
44
00:06:43.248 --> 00:06:55.448
there is absolutely no way anymore that you can un- that you can completely safely say, "My device is clean. My device is safe. My
device completely complies to everything, and everything is perfectly fine."
45
00:06:55.908 --> 00:07:15.188
Because of the, of all of the different zero days that we just cannot protect with, with, for, for which patches do not exist, and the amount of malware that's produced, the amount of attackers, and now especially with the AI, this is just something that can no longer be controlled in a way that you can say, "As long as it's our device under our control, it's safe."
46
00:07:15.288 --> 00:07:28.768
No, it doesn't mean that it's safe. That makes total sense. So thinking about it, what is fundamentally broken about legacy vulnerability scanning models based on periodic scans and manual workflows?
47
00:07:30.148 --> 00:07:40.428
So I mentioned AI already, and technically speaking, before AI, if you look at it, the only attackers we are, we were fighting were the attackers that have the power like we do.
48
00:07:40.848 --> 00:07:52.248
Okay, they might be a little bit better with the technology, but that's not an order of a magnitude difference. Nowadays, we are no longer fighting those people. We are fighting the A... those people who are using AI.
49
00:07:52.308 --> 00:07:54.908
We are fighting the armies- Yes... of agents.
50
00:07:55.588 --> 00:08:15.768
So every single thing that you and I can do by using some kind of publicly available model like Gemini or, I don't know, uh, or GPT models, either by OpenAI or residing in Copilot or using Claude or using any of those tools, all of those tools are equally available to absolutely anyone who wants to be malicious.
51
00:08:16.368 --> 00:08:25.988
So most of the, the, a little, I would say, light in that whole story is that most of those models are actually controlled by corporations which put guardrails.
52
00:08:26.268 --> 00:08:30.928
So you cannot just ask a model like Gemini to generate you, uh, some kind of malicious code.
53
00:08:31.048 --> 00:08:41.168
That's never gonna work because those guardrails are in place, and they're under control of the, of the actual providers of the models. That's what they need to do. That's basically the, uh, responsible AI approach.
54
00:08:41.508 --> 00:08:52.788
However, what we wanted and what we always said that it's unimportant because of the whole free software movement is a lot of the open-weighted, open source models that are out there.
55
00:08:53.308 --> 00:09:07.448
So that's the model that both you and I, if we knew how to do it, could take, and based on somebody else's research, we could easily create a model that's without the guardrails and that can be as powerful as any of the commercials model out there.
56
00:09:08.008 --> 00:09:21.208
And we can then create an army of agents which are using models that are not guardrailed in any way, shape, or form. So they are becoming mashi- malicious actors. We're no longer fighting the army of hackers.
57
00:09:21.648 --> 00:09:28.128
We are f- maybe fighting one single hacker with the army
of agents. So that's an order of magnitude difference.
58
00:09:28.428 --> 00:09:38.768
What I always like to say, you cannot hire your way out of this one because you will never be able to hire as many people as there are AI agents that are attacking you. And you are in a- Yes...
59
00:09:38.788 --> 00:09:50.408
bad position because you need to protect your every hole while attacker needs to find only one. So that's really, really difficult position to be. Now, you mentioned vulnerability- Absolutely...
60
00:09:50.508 --> 00:09:51.828
scanning and all of those things.
61
00:09:52.248 --> 00:10:05.688
So they were fine previously because the way that we were changing the code was on a weekly basis or a biweekly basis, and then weekly code scans were fine due to the fact that they would be finding the vulnerabilities in the code.
62
00:10:06.148 --> 00:10:17.048
Two things, two things in our reality have changed. First of all, we are writing so much more code, and we are committing so much more code, so our applications are evolving multiple times a day. So that's one problem.
63
00:10:17.348 --> 00:10:33.408
That means weekly scans are no longer sufficient. Second problem is that vulnerabilities are being discovered much quicker, which is good, but they are being exploited. The exploits are found on an even quicker basis.
64
00:10:33.988 --> 00:10:42.628
So if you look at some of the data, I mean, every single percentage out there is probably thought of during the actual, [chuckles] during the actual talk, but we don't have to talk about just the percentages.
65
00:10:43.048 --> 00:10:54.368
If you look at exploit windows, so this is the statistics. Before, five years ago, six years ago, time to exploit newly found vulnerability was one point three years.
66
00:10:54.868 --> 00:11:05.568
So some vulnerabilities discovered in a operating system or, I don't know, some widely used application, it took roughly one point three years for it to be exploited first.
67
00:11:05.728 --> 00:11:17.748
Now, it takes one point six days, and this is just the statistics. So the window of opportunity which we had to fix our systems, which was years or maybe in some case, at least months,
68
00:11:18.688 --> 00:11:28.088
now we are talking about days and hours. This is something that the traditional approach to scanning just cannot follow. We need to change something about it in order to make it happen.
69
00:11:28.448 --> 00:11:46.548
So that's some of the, some of the approaches that, uh, most recent, uh, I would say fast lane engagements and most recent education that we did with, uh, with engineers from different companies actually have a lot to do about to actually figure out how we can use AI to protect ourselves against the hackers who are using AI.
70
00:11:47.468 --> 00:11:56.088
AI. That's really interesting, and that's exactly what everyone is talking about according to the latest news, and everyone,
71
00:11:57.048 --> 00:12:02.448
you know, is scared
about what's going on in this universe, how to defend ourselves, how to defend the company.
72
00:12:02.888 --> 00:12:08.088
Uh, so it's, that's really interesting for us to clarify what's actually happening in this real world, right?
73
00:12:08.848 --> 00:12:23.408
When we talk about, you know, continue with security, how should security leaders think about AI-generated code as a productivity opportunity and a new security risk or both? Both, of course. I would put it like this.
74
00:12:23.448 --> 00:12:31.048
I mean, AI-generated code is nothing more, nothing less than any other kind of AI gen-generated material.
75
00:12:31.536 --> 00:12:42.616
So whether it's a document that you are generating, whether it's a picture that you are generating, some clip, video clip or something like that, some sound, music, whatever you're generating, it's going to be AI-generated content.
76
00:12:43.096 --> 00:12:55.496
By definition, AI-generated content is generated by non-deterministic models, which are probabilistic machines that are generating purely which word is coming after a certain set of words.
77
00:12:56.736 --> 00:13:00.836
That might be oversimplifying it, but at the end of the day, in essence, that's what they do.
78
00:13:01.696 --> 00:13:14.436
If it's just the statistics, if it's just a probabilistic machine, that just being in quotation marks, that actually means that statistics can be wrong, and what model generates is often wrong.
79
00:13:14.976 --> 00:13:18.516
Might not be completely wrong, but it's wrong in the context that it's being used.
80
00:13:18.896 --> 00:13:26.016
So just like you and I can read certain article, for example, you are a marketing specialist, so you can read certain marketing message and figure out, "But this is good.
81
00:13:26.076 --> 00:13:35.256
However, this is not exactly describing the product that I'm trying to actually market nowadays." So we'll need to change it a little bit. So you can imagine that that same thing is happening with the code.
82
00:13:35.976 --> 00:13:42.096
The problem is that the way that models learned about the code was based on poorly written code.
83
00:13:42.676 --> 00:13:51.556
Because if you look at it on the public internet, there's more people looking for the solutions than there are people providing the solutions to the problems.
84
00:13:51.936 --> 00:14:03.596
So statistically speaking, more bad code is written on the public internet than good code, and due to that fact, models learned on the bad code, and especially security vulnerable code.
85
00:14:04.136 --> 00:14:10.416
And I would argue that there are some purposely put unsecure examples out there by those same people that we already talked about.
86
00:14:10.736 --> 00:14:21.096
And that's all going into the model because we can clear out some of the content, but usually we are also using LLMs to help us clear out the content. So that's not the perfect approach.
87
00:14:21.356 --> 00:14:31.636
Not every single
information that is entered into the training data is absolutely correct. A lot of them are, but not all of them. So just always like to give this one example.
88
00:14:31.696 --> 00:14:43.776
If you look at the Stack Overflow, which is typical database that developers are using, you ask a question. There's gonna be three wrong ones and one correct one, because that correct one is going to be accepted.
89
00:14:44.136 --> 00:14:55.336
No longer is anybody going to post another correct answer. So that means, statistically speaking, that model is entering three wrong and one correct into its training data. So that's a problem.
90
00:14:55.956 --> 00:15:08.776
And that's all of that put together, you can combine with a lot of human errors, because a lot of the code wasn't protec-protected as much due to the fact that those vulnerabilities were hidden by different mechanisms.
91
00:15:09.316 --> 00:15:29.236
Nowadays, when those mechanisms are no longer applicable because it takes roughly a couple of hours to find the first exploit for the zero-day vulnerability, now we are in problem, now we are in trouble because these are just, again, some statistics that I found, 45% of AI-generated code contains vulnerabilities, some kind of security vulnerability.
92
00:15:29.796 --> 00:15:38.716
So that's a huge number if you look at it. That doesn't mean that every single AI-generated application has a vulnerable code.
93
00:15:38.956 --> 00:15:44.656
That doesn't mean it, because we are also generating different kinds of and different elements of the app.
94
00:15:45.036 --> 00:15:51.276
So it doesn't mean that every single app, every single developer that's using AI-generated code has 45% of the code that's vulnerable.
95
00:15:51.696 --> 00:16:02.816
But in general, in an essence, it might be, because there are certain things that we thought we got rid of. For example, SQL injection, that's a technical term for somebody writing poor code to access the database.
96
00:16:03.456 --> 00:16:14.436
We thought we got rid of all of that with all the different ORMs out there, all the different technologies. Actually, only 83% of the AI-generated code passes those tests.
97
00:16:14.796 --> 00:16:23.316
So there's still 17% of the code that's actually vulnerable to one of the most known issues in the world of data access.
98
00:16:23.936 --> 00:16:36.056
On the other hand, cross-site scripting, uh, log injection, those things are really, really overflowing, overflowing the AI-generated code. So that means that we cannot just trust AI when it generates the code.
99
00:16:36.096 --> 00:16:49.775
We need to check it out. Problem is that we are getting into something that's called review fatigue. There is a lot more code that is generated than we can review, and that's why that code is going to...
100
00:16:50.215 --> 00:17:03.576
if we don't do something about it, no matter how moral your engineers are, they just cannot fight the review fatigue, and that code is going
to enter the central repository one way or the other.
101
00:17:04.136 --> 00:17:15.036
So we need something else. We will need something else to protect ourselves against this. And how would be this, I wouldn't say perfect, but, like, best practice approach for this problem?
102
00:17:15.296 --> 00:17:20.056
I don't think that anybody has a best practice yet. There are things that people are thinking about.
103
00:17:20.216 --> 00:17:37.976
Again, as part of our security sessions, this is something that we tell the customers, and we work with the engineers because especially on our hackathons, which we use some, to somewhat extent to approach the best practices, to basically teach them the best practices, but also to learn from them and figure out how we can position this together with the customer.
104
00:17:38.076 --> 00:17:41.056
So that's what we sometimes do. And there's basically...
105
00:17:41.176 --> 00:17:49.716
there are terminologies there that, I mean, maybe you heard about it, about the shift left approach, and then there's shift right approach, and that's just, you know...
106
00:17:50.116 --> 00:17:59.616
I love, I love when marketing actually enters into the technology world and everybody's full of buzzwords, and then we don't- Yeah... we need to understand what is actually, what is actually happening out there.
107
00:17:59.816 --> 00:18:10.636
Technically speaking, shift left is something that has been with us for quite some time. That actually means if you look at the pipeline, so how, how are people looking at the pipeline? There's a development process.
108
00:18:11.076 --> 00:18:22.176
There's the operations process. Traditionally, the security le- was living in this part of this pipeline. So w- developers are developing, developing, developing.
109
00:18:22.316 --> 00:18:32.776
That goes into the, some kind of testing and then is entering the production, and then it's production's care to figure out how to deal with protection of the actual environment.
110
00:18:33.216 --> 00:18:42.372
So when we said shift left, that actually means shift left. Let's Take the security measures into the software development lifecycle process. And you mentioned them.
111
00:18:42.452 --> 00:18:45.332
Those are vulnerability scannings that we were doing on our code.
112
00:18:45.752 --> 00:19:00.932
But that's no longer enough because just scanning a code in some cadence, whether that cadence is going to be daily, twice a week, three times a week, bi-weekly, it really doesn't change a lot due to the fact that you are still going to have certain periods of time that you are not scanning.
113
00:19:01.032 --> 00:19:09.812
So scanning it is not going to be enough. And what about zero-day vulnerabilities? How do you scan for those? They are just discovered. You don't have a scan yet. So that's just something that...
114
00:19:09.851 --> 00:19:21.712
I mean, I'm not against scans. Scans need to be there. You need to scan. That's absolute. But there are certain other tools that can actually be available for you, and those tools have AI power in them.
115
00:19:22.152 --> 00:19:34.372
So they are going to go with the AI through your code, not just scan for those known vulnerabilities, but also scan for some potential irregularities that might be affecting your code.
116
00:19:34.752 --> 00:19:40.512
That's something that LLMs can really understand. They can actually check. There are specialized also LLMs that...
117
00:19:40.812 --> 00:20:00.212
I mean, I don't think that any of them are yet publicly available, but they will be in the coming months for sure, coming from Google, coming from Microsoft, coming from, uh, Anthropic, different vendors, that are going to be specialized in understanding unsecure code and figuring it out that it's unsecure even before there was a published article, there will be a published article on the particular vulnerability.
118
00:20:00.692 --> 00:20:10.152
So that code scanning, not just code scanning, but code analysis, which is basically combining static and dynamic code analysis. Static means you just look at a code.
119
00:20:10.172 --> 00:20:20.212
Dynamic means you run the code and figure out how it works. So, uh, basically, it's combining those two approaches to create a set of alerts, set of potential vulnerabilities.
120
00:20:20.752 --> 00:20:27.512
Now, we always had the tools that were able to kind of detect potential vulnerabilities. The problem is, again, the review fatigue.
121
00:20:28.192 --> 00:20:35.792
If you just give all of those alerts to the developer, they are going to start to ignore it due to the fact that there was a boy who yelled wolf.
122
00:20:36.412 --> 00:20:45.432
And after you figure out that out of 10 vulnerabilities, nine of them are not, probably the tenth one is also not. [clears throat] And that tenth one might- Might ruin your career at the end of the day.
123
00:20:45.632 --> 00:21:00.712
So the problem is that we are having to the, the, the problem is, the approach is to have the tool that can not just find a vulnerability, but due to the fact that it's an agent which is AI-powered, try to exploit that vulnerability in the safe sandbox.
124
00:21:01.712 --> 00:21:12.912
And only if it's able to exploit it, it's actually going to alert the developer that there is an actual problem. If it was not able to exploit it, that means it's a false positive, so it can actually disregard it.
125
00:21:13.292 --> 00:21:24.092
So that way, developers are going to stay in their own environment and will figure out if agent says that this is a vulnerability, it actually means that it found a way to exploit this. Let me look at it.
126
00:21:24.332 --> 00:21:33.572
Let me fix my code. Agent can obviously also recommend the fixes. Then you would just as a developer double-check what it changed, maybe do run some, a couple of tests, and you can commit that.
127
00:21:33.992 --> 00:21:41.592
So that's one part of the approach that
we could take to actually have better code that's written. That's really interesting.
128
00:21:41.652 --> 00:21:52.832
Imagine what you can learn from an actual class or a hackathon when you can put it in practice, right? That's incredible. So when you talk about shift left security, how it actually...
129
00:21:52.892 --> 00:21:57.532
like, how does it actually look like without becoming a blocker for developers?
130
00:21:57.772 --> 00:22:13.072
As I mentioned, the important part of shift left is it-- And we, we're gonna talk in a second about a shift right also because that's, that's now the part of the whole, of the whole security package, which is kind of, you know, that infinite loop that, that always, that always goes beyond both of the organizations.
131
00:22:13.312 --> 00:22:25.592
For shift left is secru- it is crucial that any kind of security checks, any kind of security advanced tools are part of the software development life cycle in a way that they meet developers where they are.
132
00:22:26.132 --> 00:22:38.252
If your developers are working in integrated development environment called Vis- Visual Studio or JetBrains or some of the other ones, NetBeans, I don't know, that's where your security tool that developers are using must live.
133
00:22:38.772 --> 00:22:51.632
Because if you are going to shift, switch developers from their main tool into certain portal or let them make, make them run certain kinds of, uh, applications or certain kinds of command line interfaces, you're going to lose them.
134
00:22:51.712 --> 00:22:58.132
They... Every single tool that's touching on the shift left side of the developer, it needs to live where the developer actually works.
135
00:22:58.532 --> 00:23:09.292
So all the tools need to be able to be integrated directly with the integrated development environment, indirectly with the CI/CD tools like GitHub or GitLab or some of the other ones that people might be using.
136
00:23:09.652 --> 00:23:21.652
So meet the developer where they are, and they are going to trust the security people again. So I would say that's, uh, an important part of that whole, of that whole story. Awesome. And you mentioned the shift right.
137
00:23:21.932 --> 00:23:24.132
So can you explain a little bit for,
138
00:23:25.052 --> 00:23:38.392
not only for the engineers but for, because they probably know it, but for the end user, what is the main difference between left and right, and how we could be literally use it in day-to-day? So you mean politically or?
139
00:23:38.412 --> 00:23:44.992
[laughs] I'm just joking. Le- let's not talk about it nowadays. Everything is so all over the place. No, no, no.
140
00:23:45.392 --> 00:23:54.672
I mean, technically, uh, shift left, I mean, again, if you look at the pipeline, shift left actually meant, all right, let's move the security to the left side of the pipeline- Okay... where it never existed.
141
00:23:55.112 --> 00:24:02.172
Now, shift right is a little bit different because security was always living on the right side. So that actually mean what? We're gonna shift it even more right? No, it's not about that.
142
00:24:02.232 --> 00:24:11.772
It actually means shift the right side. Let's now change, make some changes in the right side, and maybe even move it along the way so that we can feed it back into the development life cycle.
143
00:24:12.212 --> 00:24:19.972
So the differences that are happening there are, to somewhat extent, there are a couple of things that are important, I would say. First of all, there's AI scanning.
144
00:24:20.092 --> 00:24:31.140
Now, it's not just collecting the signals, it's AI correlating those signals. So that's, that's what we got already with the tools that are part of this Extended detected response, XDR.
145
00:24:31.600 --> 00:24:39.780
So XDR is che-checking the behavior and trying to correlate the different signals in the production against the potential attack.
146
00:24:40.200 --> 00:24:52.960
If they- if it figures out by the machine learning capabilities that there is a potential attack, it's going to do certain things to remediate it, and it can't do that due to the fact that we have agents that are running inside of those agents, how do you say?
147
00:24:53.340 --> 00:25:02.120
Agents or applications or system services, doesn't really matter. They're not that autonomous by definition. XDR stops at the autonomy. There are things that goes beyond, that go beyond.
148
00:25:02.540 --> 00:25:10.980
But the fact is that XDR systems are collecting information at the source, and by definition, they understand what that information is.
149
00:25:11.380 --> 00:25:16.420
If you look at a couple of years before, we had SIEM, security databases by definition.
150
00:25:16.720 --> 00:25:34.640
So what they were doing, they were just collecting, and then it would be up to the actual security professional, some kind of, uh, SecOps professional or some kind of, you know, engineer at the end of the day, to correlate those signals and figure out whether something is potentially dangerous, and then figure out a way how to remediate it.
151
00:25:35.140 --> 00:25:43.880
Now that we have si- SIEM, SIEM plus SOAR systems, that SOAR actually is the response part, which would run some kind of automated responses.
152
00:25:44.280 --> 00:25:57.300
In both situations, we needed to predict to the certain level what attack can look like, correlate it, and then what the response should look like. So that's perfectly fine when you are fighting against the known attack.
153
00:25:58.100 --> 00:26:03.960
What if it's a completely unknown situation? And we said that the vulnerabilities are discovered and immediately getting exploited.
154
00:26:04.360 --> 00:26:16.100
Due to that fact, the modern tools need to take something that you would call a baseline operations, so a typical way that users are using the systems, and typical signals that those systems are generating.
155
00:26:16.480 --> 00:26:23.180
That becomes the baseline of your traditional production behavior. And then you are looking for abnormalities.
156
00:26:23.420 --> 00:26:34.400
You're doing anomaly detection, and if certain anomalies are happening, you're trying to correlate that information with some of the known, some of the potentially known attacks.
157
00:26:34.580 --> 00:26:41.840
And they are no longer coming, they're no... That information is no longer coming just from the databases, because databases now lag the actual reality.
158
00:26:42.200 --> 00:26:48.100
They are actually trying to use generative AI to figure out what could that set of signals potentially mean.
159
00:26:48.660 --> 00:27:00.860
Does that actually mean that even though there is no vulnerability that was discovered at that particular phase, I don't have anything in the database, but I can clearly see that the attack is happening right on that attack vector on that particular dot of our system.
160
00:27:01.200 --> 00:27:14.600
That actually means that I should alert someone, I should remediate the attack, I should create a patch, maybe even automatically generate a code and prepare that set of code, that script particularly, for an engineer to just approve it.
161
00:27:15.060 --> 00:27:25.820
So security operations professional would just look at what AI prepared, was alerted, "Cool, that looks good. Stop the attack. Let's apply this particular patch." So that's one part of the shift right approach.
162
00:27:25.860 --> 00:27:29.080
I wouldn't say that we are shifting it right, we are just shifting what's happening on the right.
163
00:27:29.640 --> 00:27:45.640
Second thing is to have all that info that's collected from the vulnerability scanning, from the anomaly detection, and feed that back into the development loop so that that's the information that's coming as a context for those agents that are running the scans on the development sides.
164
00:27:46.020 --> 00:27:59.220
And that extended context now can better tell developer, "Okay," because it might be you have a vulnerability, but it's not important because your system is not exposed to the outside internet, so that vulnerability will never get exploited.
165
00:27:59.780 --> 00:28:08.920
So due to that fact, that's a false positive. Or it might not be a false positive, it's still a good s- good, good thing to fix, but it shouldn't be on top of your agenda.
166
00:28:09.120 --> 00:28:19.460
There are more vulnerable things that you need to do. Total sense. So where does micro-segmentation fit into the architecture, and how does it reduce the blast of a compromised system?
167
00:28:19.720 --> 00:28:27.000
I mean, micro-segmentation as a term, it's particular... I mean, it's okay to use, so we can use it here. Let me, let me give you an example.
168
00:28:27.620 --> 00:28:41.860
So if we have two networks that are not connected together, so we have network A, three PCs here, network B, three PCs there. They're not, none of those PCs are connected to the public internet.
169
00:28:42.360 --> 00:28:51.820
They are just connected between each other, and we share some files here, and we share some files there. There's no way for a PC from the network B to talk to the network A.
170
00:28:52.340 --> 00:29:00.260
If there is no way for them to talk, there's no way for virus to transfer from one place to the other. So that's a segmentation by definition.
171
00:29:00.480 --> 00:29:13.000
Now, the question is, obviously, in order to, for us to make different systems communicate, this cannot be completely separate because they will never be able to communicate because they have a physical boundary of communication.
172
00:29:13.560 --> 00:29:16.080
This is how we started approaching the network security.
173
00:29:16.200 --> 00:29:29.400
So we have one large actual physical network, but due to the fact that, that not everybody was supposed to be talking to each other, we created something that we call virtual LANs or, uh, not, that doesn't have to be wireless.
174
00:29:29.500 --> 00:29:32.820
That doesn't mean Wi-Fi, wireless LANs. It means virtual LANs.
175
00:29:33.180 --> 00:29:44.500
So we kind of s- entered in the, into the way that we are signing the different packages that are going from one point, or from point A to point B, we also introduced a little component of the VLAN ID.
176
00:29:45.020 --> 00:29:56.040
So that's, that being said, there was no way that, I mean, there is a computer on this network that can communicate with this network, but only that computer. The other ones cannot communicate.
177
00:29:56.440 --> 00:30:04.820
So that's how we separated those things. So there are certain things that this one can decide to send over, but the other ones cannot do that without that particular machine.
178
00:30:05.160 --> 00:30:09.720
So this is how we approach that particular topic, and that's what was the network segmentation.
179
00:30:10.240 --> 00:30:17.740
Now, micro-segmentation would actually mean if there is a certain aspect, and especially nowadays when we are talking about the containers, this is really, really important.
180
00:30:18.180 --> 00:30:27.280
So if there's a certain aspect of the system that shouldn't communicate with the other systems, we should block that communication. So I can tell you, give you an example.
181
00:30:27.520 --> 00:30:41.320
Let's say that we have some kind of social network application like, you know, it's maybe for charities And we have a microservice controlling the list of charities, so you can use it to update your charity inserts, uh, something or whatever.
182
00:30:41.760 --> 00:30:46.680
We have a microservice controlling the posts, so all the announcements that that charity does.
183
00:30:47.060 --> 00:30:56.180
We have a microservice controlling the media aspects of videos, audio files, images, and we have the one controlling the users. So there are four microservices.
184
00:30:57.040 --> 00:31:07.300
In an essence, there is a web application that needs to talk to all of them. However, web application needs to talk to me... to, to charity, to posts, to users.
185
00:31:08.740 --> 00:31:21.800
Posts need to talk to media, and all of them need to talk to their own database. There is no reason why posts should talk to users or why media should talk to users. There's no reason.
186
00:31:21.900 --> 00:31:28.940
There's no logical reason to do that. Now, on the level of the security inside of the container as ecosystem, you're protecting that communication.
187
00:31:29.240 --> 00:31:37.800
You're not allowing those, those microservices to talk to anybody but the ones that they need to talk to. So what you are... What are you doing with that?
188
00:31:37.959 --> 00:31:54.900
You are providing, by definition, the security that if one of your pods, as we call them in a Kubernetes or containers in general, if one of your containers is compromised, your virus cannot jump to the other one, and jump to the other one and just spread its way around.
189
00:31:55.880 --> 00:32:05.880
So that's basically why, why this part is important because that's protecting us by definition, because we are protecting the communication in its own, in its essence. Awesome.
190
00:32:06.300 --> 00:32:17.160
So how should leaders prioritize fixing security issues based on exploitability and business impact? Well, just like any other situation in the company, this is a leader's problem.
191
00:32:17.700 --> 00:32:27.760
So basically, if you're looking at how security was treated like fifteen years ago, it was an IT problem because IT was just touch basing with some systems.
192
00:32:27.920 --> 00:32:38.080
Most of the companies were running without that much reliance on the IT. I mean, depending on the industry, again, obviously in bank- banking sector and finance sector in general, it was more important even back then.
193
00:32:38.500 --> 00:32:45.940
But in some of the other ones, it was maybe not as important. Suddenly, IT security is much more important than even the perimeter security.
194
00:32:46.060 --> 00:32:56.260
You know, if you're having a production unit, it might be attacked without anyone even approaching your actual facility. So that's what becomes the problem.
195
00:32:56.300 --> 00:33:11.500
And when something like that becomes the problem, it becomes on top of mind of every single CEO, which is then talking with its, uh, with its board member, uh, called CISO, Chief Information Security Officer, usually reporting to Chief Security Officer or working together with Chief Security Officer.
196
00:33:11.540 --> 00:33:18.980
That is, again, part of how the organizations are organized at the end of the day. So when you're looking at the business impact, it really depends what your company is doing.
197
00:33:19.400 --> 00:33:35.300
But I would argue that nowadays that every single exploit can have a potentially, a potentially huge business impact on your actual business continuity first, and then even data loss and some of the inability, inability to do business in the future.
198
00:33:35.780 --> 00:33:44.660
So business continuity is an impor-important fact. And I mean, we approach part of the business continuity from the backup point of view, so that's... I think that we are covered there.
199
00:33:44.720 --> 00:33:53.540
Everybody's thinking about backups and all of that. But I would, I would argue that IT information security is even more important or at least equally important as backups and all of that.
200
00:33:53.740 --> 00:34:04.760
Because your machines are constantly running and your whole system is constantly online, they are no longer... there's no longer switching off the IT. That used to be back in the day, but that's no longer a thing.
201
00:34:05.080 --> 00:34:12.640
So IT is running constantly, IT is constantly vulnerable, and if you don't fix it in time, it's going to impact your business in one way or the other.
202
00:34:13.120 --> 00:34:22.480
So technically speaking, if you are just starting as a leader, if you are just starting to apply those new practices of zero trust, I would again say start small.
203
00:34:22.920 --> 00:34:32.780
Because if you just lock everything down, your employees are going to riot, you're going to have trouble. At the end of the day, you're not gonna be able to do your business. So that's not a solution.
204
00:34:33.140 --> 00:34:36.960
Security should never be blocking the business. It should be a business enabler.
205
00:34:37.500 --> 00:34:51.159
So from that point of view, I think that every single IT leader should, first of all, educate themselves to learn more how these approaches are actually affecting them, what actually matters, and what we need to control immediately.
206
00:34:51.400 --> 00:35:03.480
And then it's just like any other change management. Let's fix things that we can do quickly first because that's logical. But again, in those things, let's prioritize based on the impact.
207
00:35:03.580 --> 00:35:13.380
What's gonna have the most impact? Where am I protecting the most? I would argue that one thing that each and every one of us can do today is switch on the multi-factor authentication.
208
00:35:13.940 --> 00:35:22.360
If you don't have it on your business accounts or on your personal accounts, go ahead, switch it on today because that's, that's going to stop...
209
00:35:22.900 --> 00:35:27.500
I wouldn't say, again, some kind of percentage because it's gonna be thought of, but a lot.
210
00:35:27.580 --> 00:35:36.900
I would, I would even argue that it's more than eighty percent of the attacks that can and will be done by exploiting your identity. So ident-- protecting the identity is crucial.
211
00:35:37.320 --> 00:35:47.480
That's the first thing, and it's so easy fix. Then I know that some people might argue, "Oh, then I have so many of those authenticators." But you as a business
control the identity.
212
00:35:47.900 --> 00:36:00.780
You as a business should think about the identity-based solution that's gonna give you the most convenient way of authentication, depending on whether you are Microsoft shop or Google shop or Cisco shop or any other shop out there.
213
00:36:01.080 --> 00:36:13.340
You should think about what is best for you, not what's best of breed, because all of those solutions were good. They're tested. They're run by serious corporations who are aiming to be the best corporations ever.
214
00:36:13.480 --> 00:36:22.060
So they're thinking about themselves, and they're thinking about those rep-- that their repu-reputation. So all of those solutions are good. The question is, which one fits the best into your environment?
215
00:36:22.760 --> 00:36:32.840
I always like to say, if you are a Microsoft environment, if you have M365 and all of that, leverage Entra ID as much as you can because there's everything there that you need. You just need to configure it correctly.
216
00:36:33.100 --> 00:36:44.020
There's no need for, you know, implementing some third-party system into, into that mix. Because again, with another system come more vulnerabilities, potential vulnerabilities and potential exploits.
217
00:36:44.540 --> 00:36:51.380
So you need to think or figure out your way. But I would say protect the identity is gonna be your first step In the right direction.
218
00:36:51.780 --> 00:37:03.820
And then from that, you can take different directions working with your teams, but it's going to require time, it's going to require effort, and it's going to require change management in order for you to be successful with that particular transition.
219
00:37:04.120 --> 00:37:18.920
If an organization is starting today, I want you to give a message to them saying, I don't know, three of the most practical actions that they can make on the next ninety days, for example. How they can start. Yeah.
220
00:37:18.980 --> 00:37:29.960
I think we started already by talking a little bit about identity now. I wouldn't say that there's an organization in the whole wide world that's starting with the security right now. I don't think that's a fact.
221
00:37:30.140 --> 00:37:38.780
I think that we are... Everybody's aware of that problem, and I mean, we are in a cybersecurity awareness month, and it's good that we are talking- Yes...
222
00:37:38.860 --> 00:37:45.960
about this topic and making people even more aware that it's even more dangerous than they might think about. But I don't think that anybody's beginning.
223
00:37:46.180 --> 00:37:55.320
So there are a ton of investments that companies already made into security. What they might not be aware of is what those investments are. Because they might...
224
00:37:55.540 --> 00:38:05.840
I-- and even though this is a podcast, I will still be completely open and honest with you, Jess. There are a lot of people out there who are buying software just because somebody
told them to buy it.
225
00:38:06.320 --> 00:38:10.700
That doesn't mean- Yes... that that software is implemented. That doesn't mean that that software is correctly implemented.
226
00:38:10.740 --> 00:38:22.220
So I would s-- I would argue that the first step anybody can do is the assessment of their cybersecurity reality. So that actually involves which software do you have? Which software do you use? Who is using it?
227
00:38:22.280 --> 00:38:34.680
Who is owning it? How is it maintained? And then we can also talk about the reality of your IT infrastructure, whether it's in the cloud or it's on-prem. How often do you patch it? How often do you scan it?
228
00:38:34.780 --> 00:38:47.500
What is your approach? So you need to be aware of what you have in order for you to make any kind of action that's just something that is practically normal. So I would say that assessment is an important part.
229
00:38:47.600 --> 00:38:57.600
Both software assets management from the point of what kind of software are you using or hardware or whatever you have in your protection system, and how you're using it and who's controlling it and who is the owners.
230
00:38:57.820 --> 00:39:08.000
Second thing is make sure that every single employee in your organization is aware of the things that we talked about now. It can be a fifteen-minute session with everyone.
231
00:39:08.200 --> 00:39:15.860
It can be an email with a set of URLs, set of links on the public web that people can actually just read about.
232
00:39:16.000 --> 00:39:23.680
It can be something done on the organizational level, or it can be something done on the team level, where you might have a cybersecurity champ inside of every organization.
233
00:39:24.040 --> 00:39:29.960
Really is totally up to the org to figure out what approach is, but we need to start having these, these conversations.
234
00:39:30.060 --> 00:39:52.880
It's really, really important to talk about it, because once everybody's aware that tomorrow you might receive an email from the CEO asking you to put some money in some account, and it might look like it's coming from your CEO, it might be for the project that just you and your CEO discussed yesterday in your email, it still might be phishing.
235
00:39:53.740 --> 00:40:01.780
So be careful about it, learn about it. That's the second thing that is important. And I would say that the third thing is the one that we mentioned, the first.
236
00:40:02.380 --> 00:40:06.840
If you are able to switch on the multi-factor authentication today, do it now.
237
00:40:07.000 --> 00:40:14.760
An important part is that we need to understand that there's a ton of things that you can do for yourselves and for your engineers and for your end users.
238
00:40:14.820 --> 00:40:26.120
And I mean, that's something that, I mean, I always, always keep that part for Fastl-- for what we do in Fastlane, but I'm really proud by all the work that we do, so I sometimes really want to intervene it into the whole content, because that has...
239
00:40:26.280 --> 00:40:34.720
there's a lot of education out there that engineers and end users can actually leverage. There are stuff coming from Cisco. Cisco is a traditional security vendor.
240
00:40:35.040 --> 00:40:45.540
Some of the other ones like Fortinet, like Palo Alto, that's all the trainings that we do, and they have its component in how to configure, how to configure those advanced systems that are protecting us.
241
00:40:45.860 --> 00:40:57.340
Also stuff coming from Microsoft, coming from Google, coming from Amazon, so that those are more cloud-oriented. I would say Cisco, Fortinet, and Palo Alto are also protecting your actual security in the on-prem world.
242
00:40:57.880 --> 00:41:01.680
On the other hand, there are general security things like CompTIA security certificates.
243
00:41:01.720 --> 00:41:08.920
For example, if you as a pro-- as a professional marketeer want to learn more a-about the security, that might be something that you're gonna look into.
244
00:41:09.199 --> 00:41:16.360
To figure out how can you protect your way into the s- into the public internet, how can you protect your way into the digital, so that you
245
00:41:17.320 --> 00:41:21.800
get yourself familiar with the typical practices that are important on the end user side.
246
00:41:21.880 --> 00:41:32.780
So just like with any other thing in the IT world, there's engineering education and there's an end user education, and I think that there's a ton of stuff that we can help our customers with just by providing them those kinds of services.
247
00:41:33.000 --> 00:41:43.320
Awesome. That's all for today's episode. I wanna say a big thank you, Tom, to sharing practical insights on building resilience against AI-powered threats, right?
248
00:41:43.660 --> 00:41:54.700
It's really important for us to keep the awareness not only on the Cybersecurity Awareness Month, but every day. Uh, even if you're an organization or if you're an end user, it's really important to, to consider.
249
00:41:55.640 --> 00:42:07.960
To keep building cybersecurity expertise, please explore Fastlane security training, and thank you for listening. Just remember, in cybersecurity, trust is never assumed. It's continuously verified. Thank you.
250
00:42:08.000 --> 00:42:09.280
Thanks, everyone. Bye.