תעתיק 🔗
WEBVTT
00:00:25.039 --> 00:00:27.600
Hello and welcome to Full Tech Ahead.
00:00:27.760 --> 00:00:31.280
I'm your host, Amanda Razzani, and with me today I have Ben Morrell.
00:00:31.359 --> 00:00:34.079
He is the VP of Security Strategy at Coro.
00:00:34.399 --> 00:00:35.359
How are you doing?
00:00:35.679 --> 00:00:37.280
I am doing absolutely fantastic.
00:00:37.439 --> 00:00:38.240
And yourself?
00:00:38.479 --> 00:00:39.759
I'm doing well.
00:00:40.000 --> 00:00:50.320
Well, can you start off by just sharing about your role there at Coro and then talk about being customer zero and what does that mean and what does that show you?
00:00:50.719 --> 00:00:51.359
Yeah, for sure.
00:00:51.679 --> 00:00:55.600
I am uh VP of Security Strategy, which means I play kind of a dual role in the organization.
00:00:55.679 --> 00:01:04.000
Uh I work with our product and research and development teams to help define what Coro does, what we're doing, why we're doing it, how we're doing it, who are we doing it for?
00:01:04.159 --> 00:01:07.840
And more importantly then, I can take that back internally and say, how does it work for us?
00:01:08.000 --> 00:01:09.680
Is this doing what we expect it to do?
00:01:09.760 --> 00:01:13.920
And if it's not, how can we expect that to function for someone else that we are providing it to?
00:01:14.079 --> 00:01:18.239
So that role allows me to sort of uh treat both sides as independent.
00:01:18.400 --> 00:01:21.280
I come in-house and say, is this working for us?
00:01:21.519 --> 00:01:23.439
Are we happy with the way that this is functioning?
00:01:23.519 --> 00:01:24.719
Have we hit the nail on the head?
00:01:24.799 --> 00:01:31.840
And if we haven't, I can then take that back to the other side of the team and say, the guy on the other side of the fence who is still me doesn't think this works that way.
00:01:32.079 --> 00:01:33.840
So it's an interesting sort of exercise.
00:01:34.000 --> 00:01:39.920
We are technically, while I have a lot of RD and dev teams, we are still like lean IT when it comes down to it.
00:01:40.079 --> 00:01:46.480
So being able to use our own infrastructure and our own talking points and our own methodology is a very important exercise for us.
00:01:46.879 --> 00:01:47.760
Wonderful.
00:01:48.079 --> 00:01:52.799
Well, AI is, of course, in security everywhere these days.
00:01:53.040 --> 00:01:56.959
So, from your experience, what are people getting right?
00:01:57.519 --> 00:01:59.120
Where are they getting hung up?
00:01:59.200 --> 00:02:05.200
And is it gonna be sort of a silver bullet for someone if they if they mess up?
00:02:05.840 --> 00:02:19.759
So uh AI, uh like LLMs for a long time, has done a fantastic job and still does crunching data, uh taking complexity out of comparisons and correlations and bringing things together and being able to say, here's what really matters out of all of that data that you had.
00:02:19.919 --> 00:02:28.400
So people are using it for that purpose
Load full transcript
fantastically, whether it's cybersecurity, which is predominantly what we're talking about, or even other industries that uh have to do a lot of data crunching.
00:02:28.639 --> 00:02:34.159
Where I see it get really overcomplicated is when it's applied to things that doesn't probably need to be applied to right now.
00:02:34.240 --> 00:02:38.080
It's like, how do we force AI into everything that can be used for?
00:02:38.159 --> 00:02:49.599
Uh and sometimes I've seen processes and methodologies that were simple and straightforward enough that just became probably more complicated by trying to work out how we can AI-ify it or even slam an LLM on top of it.
00:02:49.840 --> 00:02:56.800
Now, I'm sure just like me, yourself, mayor, you've seen a lot of people potentially claim silver bullets, and it's always a very risky exercise.
00:02:56.879 --> 00:03:00.879
And I think you've got to be very careful of anybody that ever comes along with that sort of scenario.
00:03:01.039 --> 00:03:03.759
I think AI is gonna be here to stay in some capacity.
00:03:03.919 --> 00:03:11.120
What that capacity looks like is I would have to have a crystal ball to know perfectly, and we would both be having a very different conversation.
00:03:11.360 --> 00:03:12.879
But it's gonna stay.
00:03:13.039 --> 00:03:15.840
And to that extent, we have to work out how we use it correctly.
00:03:16.000 --> 00:03:18.080
Is it just continuing to do data crunching?
00:03:18.159 --> 00:03:19.199
Can we get more value out of it?
00:03:19.360 --> 00:03:21.039
How do we stop complicating things?
00:03:21.199 --> 00:03:23.039
But I don't think there's a silver bullet there.
00:03:23.199 --> 00:03:33.439
Things like your user base, your risks, misconfigurations, these are all things that aren't just going to go away because now AI can see a hacker or an attacker trying to probe your network.
00:03:33.599 --> 00:03:36.639
The other issue there, obviously, is they're also using similar tooling.
00:03:36.719 --> 00:03:41.360
So it's sort of an arms race to the top or the bottom, depending on how you look at it.
00:03:41.759 --> 00:03:42.080
Okay.
00:03:42.639 --> 00:03:44.960
Well, we hear about shadow AI too.
00:03:45.120 --> 00:03:50.000
Shadow AI, it's often described as the next evolution of shadow IT.
00:03:50.319 --> 00:03:56.639
So share a little bit more about that with our audience and what does that look like for small and medium-sized businesses?
00:03:57.039 --> 00:03:57.439
For sure.
00:03:57.599 --> 00:04:07.599
So shadow IT, uh, as you brought up, right, was the methodology of everyone's got a favorite browser, but we require everyone to use browser apps, let's say Firefox in the organization.
00:04:07.680 --> 00:04:09.199
So we lock everything out.
00:04:09.360 --> 00:04:11.360
That way, everyone's forced to use Firefox.
00:04:11.439 --> 00:04:16.399
It makes sense, it's an easy concept to understand, and generally it's a fairly easily enforceable thing.
00:04:16.560 --> 00:04:20.319
Shadow AI is that similar methodology, but everyone's got a favorite AI.
00:04:20.399 --> 00:04:23.600
Uh, someone's using Gemini at home, someone's using something else on their phone.
00:04:23.839 --> 00:04:26.959
ChatGPT is very popular with you at your standard user base.
00:04:27.040 --> 00:04:33.920
But as an organization, perhaps we've settled on Claude, and that's what we're paying for, it's what we're rolling out, that's what we provide our security metrics for.
00:04:34.079 --> 00:04:39.519
But now I've got users that go and use ChatGPT through a web browser or through a plugin or through an agent, right?
00:04:39.600 --> 00:04:47.439
And this shadow AI is done potentially not maliciously, but just through the course of I'm comfortable with that, it makes me comfortable, that's what I want to use.
00:04:47.759 --> 00:04:50.879
The exercise there is, well, okay, let's do the same thing as shadow IT.
00:04:50.959 --> 00:04:51.759
Let's lock it down.
00:04:51.920 --> 00:04:53.680
You can't touch any of the others.
00:04:53.920 --> 00:04:56.560
The issue there is closing a browser was very easy.
00:04:56.639 --> 00:04:58.879
They're not going to shift somewhere to try and use another browser.
00:04:58.959 --> 00:05:00.000
It's too complicated.
00:05:00.240 --> 00:05:03.120
But people are very comfortable with what they're using when it comes to AI.
00:05:03.199 --> 00:05:08.319
They've, this is the one I know how to talk to, this is the one I know that I get my responses and how to get those prompts.
00:05:08.399 --> 00:05:17.680
Uh, and your average user will then go, well, okay, if I'm being asked to do more work or to do this with AI, but I don't know how to do it with the product that we've been given, I'll take it to my own system.
00:05:17.839 --> 00:05:19.839
I'll verbally communicate it to my phone.
00:05:20.079 --> 00:05:22.240
I will OCR read it with a camera.
00:05:22.399 --> 00:05:29.680
And so what happens is if you start to block it instead of perhaps trying to educate around it why we're doing this process, you drive it underground.
00:05:29.759 --> 00:05:35.439
And the moment you've driven it underground, you lose visibility of it, regardless of what tool you want to decide to start using in the future.
00:05:35.680 --> 00:05:48.959
So a big part of Koro's exercise in this Shadow AIX uh theme is detecting where and what's being used in your organization and then educating your users on hey, that's not what we're using, here's why.
00:05:49.040 --> 00:05:51.279
And more importantly, here's what we're using and how to use it.
00:05:51.360 --> 00:05:54.720
It's not just an exercise in security, it's an exercise in comfort.
00:05:54.800 --> 00:05:58.720
Um, so it's a much more difficult exercise to control than the old shadow IT.
00:05:58.879 --> 00:06:01.040
It's got bigger repercussions, I think, as well.
00:06:01.199 --> 00:06:08.879
But the fact that so many people are involved in it, most of your users may have
had a favorite browser, everyone's probably got a favorite AI at this point in time.
00:06:08.959 --> 00:06:12.240
So you're just gonna have to get used to trying to control that exercise.
00:06:12.560 --> 00:06:22.079
Who in a company, from your experience, who should own that conversation, the AI education and responsible use of the AI tools?
00:06:22.399 --> 00:06:22.879
For sure.
00:06:23.040 --> 00:06:25.519
So we we deal with a lot of what we call lean IT.
00:06:25.600 --> 00:06:36.000
Lean IT for us is organizations that may have an individual that wears multiple hats that does everything from IT to network to procurement of security services to security itself.
00:06:36.079 --> 00:06:39.360
Uh, and now AI, which is that new one added on top.
00:06:39.519 --> 00:06:50.639
This might also be a team who's specialized in different parts, uh, but what we also see is a lot of uh either an individual that isn't quite adapt at that, that is outsourcing that material to a managed service provider who we deal a lot with.
00:06:50.720 --> 00:06:52.560
We've got a very big managed service program.
00:06:52.639 --> 00:06:58.480
And so what we find is that conversation needs to involve them, but also involve the rest of the upper business.
00:06:58.639 --> 00:06:59.519
What are we using?
00:06:59.600 --> 00:07:00.560
Why are we using it?
00:07:00.639 --> 00:07:04.160
And then also remove that conversation down into the users.
00:07:04.399 --> 00:07:06.160
This is why we're using what we're using.
00:07:06.319 --> 00:07:25.600
So, yes, the the IT leader, the CISO, the security leader, the head of IT, whatever they end up being in that organization, the outsourced provider, uh, if that's the possibility, need to own that as an initial conversation, but need to work out how they involve the rest of the organization in understanding what we're doing, why we're doing it, what are the outcomes we're trying to get.
00:07:25.759 --> 00:07:29.120
Because at the end of the day, we're not here to make your life harder.
00:07:29.279 --> 00:07:35.439
We're here to protect the rest of the business and yourself and what we're doing so that everyone doesn't have a bad day.
00:07:35.839 --> 00:07:36.720
Absolutely.
00:07:36.879 --> 00:07:48.000
So when it comes to education, I know that AI has really thrown a wrench in the field with voice being able to mimic, voice being able to mimic images, video.
00:07:48.240 --> 00:07:56.720
So with these new phishing and AI scams, how do we make sure that um employees and staff are educated and stay on top of this?
00:07:57.279 --> 00:07:58.879
It's a lot of it's a lot of this.
00:07:59.040 --> 00:08:01.279
If you have tooling that can do that, absolutely.
00:08:01.439 --> 00:08:08.399
If you're in a position where you can hold these conversations and involve your users as a part of your defense strategy, you definitely should.
00:08:08.480 --> 00:08:15.600
Uh, a lot of your managed service providers may have tools or programs or
even in sessions, uh, lunch and learns, as we sometimes call them.
00:08:15.839 --> 00:08:23.920
We come in and explain here's what's happening, here's why you shouldn't take calls, here's why your CEOs are going to phone you on a Saturday about buying Amazon gift cards, right?
00:08:24.000 --> 00:08:25.680
It's all about at the education flow.
00:08:25.920 --> 00:08:36.960
For a long time, there was a small period where we had the ability to sort of run tooling, lockdown systems, control them and keep them as they are, and the end user just had to flow through the system as they were guided.
00:08:37.120 --> 00:08:39.440
Now phishing is absolutely rampant.
00:08:39.679 --> 00:08:49.840
AI makes that even easier to do with the ability to pull data, create personalized email, pull from LinkedIn to work out who owns what and does what and create those conversations.
00:08:50.000 --> 00:09:00.879
So turning your, you know, users uh and even your users further up the chain, uh, into your ELT teams, for example, into lines of defense is a pretty pretty big requirement.
00:09:00.960 --> 00:09:05.360
And it's going to be one of the big requirements when it comes to how do I protect against voice phishing?
00:09:05.440 --> 00:09:08.159
Because it's a quite difficult thing to get tooling in from.
00:09:08.399 --> 00:09:12.399
And I don't know if there's going to be a solution for that anytime soon that makes anyone happy.
00:09:12.480 --> 00:09:19.840
So it's better to potentially involve the team and work out how do we create that as a part of a defense strategy of turning ourselves into defense.
00:09:20.240 --> 00:09:20.559
Okay.
00:09:21.039 --> 00:09:26.559
And so now we have AI token cost attacks as an emerging vector.
00:09:26.720 --> 00:09:28.240
So explain that.
00:09:28.399 --> 00:09:30.720
What does that mean and what are they doing?
00:09:31.120 --> 00:09:38.960
Yeah, so uh in my time, I get to rub a lot of shoulders with people who do what I do and listen to a lot of people that I'm surprised I ever get the chance to speak to.
00:09:39.039 --> 00:09:41.039
They're quite quite intelligent in what they do.
00:09:41.200 --> 00:09:48.879
One of the things that they're starting to see is uh a couple of fields in being able to utilize a company's tokens for malicious purposes.
00:09:49.039 --> 00:09:59.759
Um, this either comes into the idea that, okay, uh an attacker, a group, uh, whatever it happens to be, uh doesn't have access to or doesn't have the cost associated to running so many tokens.
00:09:59.840 --> 00:10:10.720
So they'll almost bottom it up a whole bunch of uh different organizations, clauds or open AIs or whatever it happens to be in order to do what they need to do with them and utilize their token costs.
00:10:10.960 --> 00:10:27.759
You also see on the other side, it's a very easy way to potentially hit the hip pocket of an organization, in one way I've heard described as the old uh black printer ink attacks of the 90s, where you
just push black ink out of a printer once you can get onto it in order to just use up all their printer ink and cost them money.
00:10:28.000 --> 00:10:30.240
Same thing, use up all their tokens and cost them.
00:10:30.320 --> 00:10:33.360
The tokens are quite costly right now, and I don't think that's gonna go away.
00:10:33.440 --> 00:10:38.399
I think it's actually probably going to go up at some point, which means that you know they're gonna be something that needs to be held on to.
00:10:38.639 --> 00:10:47.440
A lot of organizations, especially in the lean IT, haven't done uh the full due diligence to go, here's how we lock our token costs down, and here's rush allowed to use what.
00:10:47.600 --> 00:11:01.759
Then you start mixing in people bringing in their own AIs to use and their own token costs, and everyone's just sort of a ticking time bomb, waiting to have their tokens stolen and used for nefarious deeds or potentially just black printer attacks in the same vein.
00:11:02.159 --> 00:11:04.879
Well, so small teams are flooded with alerts.
00:11:05.039 --> 00:11:10.080
How does correlation help separate a real targeted attack from all the routine noise?
00:11:10.399 --> 00:11:10.960
Absolutely.
00:11:11.120 --> 00:11:12.559
So there's a lot of alerts.
00:11:12.720 --> 00:11:16.879
In coro we call them tickets, so I'm gonna probably use those as interchangeable terms.
00:11:17.039 --> 00:11:20.879
Uh, and in that face, you know, you've got a lot of different parts of your business.
00:11:21.039 --> 00:11:32.879
Uh, email, uh, what's happening in your cloud infrastructure, what's happening on your endpoints, EDR, DLP type exercises, things happening in your network, whole host of places that are going to be throwing you alerts.
00:11:32.960 --> 00:11:40.559
Uh, and when you get those as an individual uh alert, you may look at it and go, I don't know what that means, or okay, got blocked, it's handled.
00:11:40.720 --> 00:11:44.240
But correlation allows you to identify is there a bigger peak meaning to this?
00:11:44.320 --> 00:11:46.559
Does it is it a part of a larger exercise?
00:11:46.720 --> 00:11:51.759
And is that larger exercise what I should be worried about rather than the fact that we blocked 300 phishing emails?
00:11:51.919 --> 00:12:02.320
Well, if 20 of those phishing emails are related to potential theft of credentials that are now being root forced through your 365, it speaks a lot more to where I should be focusing my efforts.
00:12:02.480 --> 00:12:04.639
And again, that's trying to get out of that drowning.
00:12:04.720 --> 00:12:18.559
And if we go back to what potentially AI can help with, it's trying to correlate that noise and surface what actually potentially matters rather than swimming in all of the stuff that lean IT absolutely get drowned in, especially when lean IT is wearing so many hats.
00:12:18.639 --> 00:12:25.759
And it comes down to if I haven't seen an actual attack and I am getting an alert that something was handled, what do I got to worry about?
00:12:25.840 --> 00:12:26.799
Let's let's move on.
00:12:26.960 --> 00:12:41.440
So a lot of that correlation helps drive better exercises in security, better outcomes, better methodologies of using your time or your outsources time uh to work out what actually needs to be focused on and what does it actually mean to your business?
00:12:41.919 --> 00:12:49.840
Well, as we come close to year's end, what's one thing that lean IT leaders could do right now?
00:12:50.639 --> 00:12:51.200
For sure.
00:12:51.360 --> 00:13:00.639
I think coming into year's end with a plan going into the next year is, and I've sort of touched on it, start to treat your users as a part of your defense.
00:13:00.799 --> 00:13:06.480
Um, this means educating them, but it also means involving them in what is happening security-wise within the organization.
00:13:06.559 --> 00:13:21.200
Uh, we do this exercise internally where our CEO maybe gets uh pretended to be someone else on LinkedIn after we have a new staff member join, and we make it clear to staff and new users that you know they're never going to reach out to you to ask for Amazon gift cards after you join.
00:13:21.360 --> 00:13:25.919
It's a very silly exercise, but people need to be empowered to understand this and to know this.
00:13:26.080 --> 00:13:36.639
You need to open up those lines of communication, not only between the IT and security teams and the users, but perhaps that user to the person who is asking them on you know LinkedIn for Amazon gift cards.
00:13:36.799 --> 00:13:40.639
It's very easy to just go, hey, why did you want Amazon gift cards for the barbecue on the weekend?
00:13:40.799 --> 00:13:41.200
I didn't.
00:13:41.279 --> 00:13:42.320
Cool, that's closed.
00:13:42.399 --> 00:13:44.320
Then we've now protected ourselves.
00:13:44.559 --> 00:13:52.799
Your users are always going to be a part of your attack surface, regardless of where AI goes, regardless of what potential silver bullets become, they're always going to be there.
00:13:52.879 --> 00:13:55.519
So you're better off trying to turn them into a part of your defense.
00:13:55.679 --> 00:14:09.360
This means educating them on how to better identify phishing, how to better identify those attacks, how to use AI correctly, how to use infrastructure correctly, where not to put data, how to avoid people calling you on WhatsApp and saying that they're your boss, right?
00:14:09.519 --> 00:14:13.840
Uh that way you can start to move towards, okay, we're now involving our users.
00:14:14.000 --> 00:14:18.720
It takes a lot of pressure off the lean IT team because they're not chasing every little nook and cranny.
00:14:18.799 --> 00:14:22.720
They now get to work on those larger, surfaced correlated events.
00:14:22.879 --> 00:14:31.759
So, yeah, towards the year's end, especially with seasonal attacks coming up on phishing, start to treat your users less like a weakness and more like a defense.
00:14:32.480 --> 00:14:32.879
All right.
00:14:33.039 --> 00:14:37.919
Well, if there was one key takeaway you could leave our audience with today, what would that be?
00:14:38.639 --> 00:14:39.279
Absolutely.
00:14:39.519 --> 00:14:42.879
Educate, educate and involve your users as a part of your defense.
00:14:43.120 --> 00:14:44.080
Tooling is fantastic.
00:14:44.159 --> 00:14:47.840
We will always produce tooling to help you generate and get to that point.
00:14:48.080 --> 00:14:50.080
But treat your users as a part of your defense.
00:14:50.159 --> 00:14:51.919
It's the best outcome you're gonna get.
00:14:52.159 --> 00:14:52.639
All right.
00:14:52.799 --> 00:14:54.960
Thank you so much for coming on the show today.
00:14:55.360 --> 00:14:56.720
Thank you so much for having me, Amanda.
00:14:57.200 --> 00:14:58.480
And thank you to our listeners.
00:14:58.639 --> 00:15:00.559
If you have any questions or comments, leave them below.
00:15:00.639 --> 00:15:02.960
I'll try to reply back as soon as possible.