TRANSKRIPT 🔗
WEBVTT
00:00:00.160 --> 00:00:06.320
Welcome to the Talos Takes podcast, where we discuss Talos' latest research and security news.
00:00:06.320 --> 00:00:10.720
This podcast is for everyone, from the C -suite to the front lines.
00:00:12.560 --> 00:00:18.559
Hello and welcome to this special Cybersecurity Awareness Month episode of Talos Takes.
00:00:18.559 --> 00:00:21.199
Hope you've been having a lovely October.
00:00:21.199 --> 00:00:27.920
Here at Talos, we are spending this month studying the fine art of frustrating the adversary.
00:00:27.920 --> 00:00:36.880
We often discuss kind of this race between the defenders and the attackers through the lens of total prevention.
00:00:36.880 --> 00:00:41.359
So trying to construct a wall that the adversary can't climb over.
00:00:41.359 --> 00:00:53.920
But let's explore why we should be creating friction, making their life on your network a living hell as painful, as slow, as expensive as humanly possible.
00:00:53.920 --> 00:01:00.240
So Martin Lee is back to talk about how we can implement some of these strategies.
00:01:00.240 --> 00:01:06.719
And more importantly, how you can do it practically without creating this mountain of extra maintenance for your team.
00:01:06.959 --> 00:01:10.000
For those in the know, Martin Lee is no longer with Talos.
00:01:10.000 --> 00:01:15.040
He's back in wider Cisco, but I managed to snag him for another episode.
00:01:15.040 --> 00:01:16.719
So welcome back, Martin.
00:01:17.040 --> 00:01:17.599
Thank you.
00:01:17.599 --> 00:01:18.959
Thank you so much.
00:01:19.280 --> 00:01:22.560
Well, there are so many ways to frustrate an adversary.
00:01:22.560 --> 00:01:25.599
You've talked about them multiple times before.
00:01:25.599 --> 00:01:34.879
Um, you know, from the simple canary tokens and honey credentials to more advanced things, AI honey pots, fake ID objects.
00:01:34.879 --> 00:01:42.239
Um when a team is kind of looking at all of these options in front of them, how do they decide where to start?
00:01:42.239 --> 00:01:45.200
It's kind of like you're trying to boil the ocean.
00:01:45.680 --> 00:01:46.159
Right.
00:01:46.159 --> 00:01:54.959
I the place to start is to think about what actually is a cyber attack.
00:01:54.959 --> 00:01:59.840
So I'm afraid we'll have to go into a theoretical metaphorical conversation.
00:01:59.840 --> 00:02:04.799
Um, you know, cybercrime is an economic crime.
00:02:04.799 --> 00:02:20.960
This is this is where we start that the attacker believes that they're going to get value from their target in excess of the the value of the resources that they have to expend to get in there.
00:02:20.960 --> 00:02:33.199
So um it's clearest in um in a cybercrime attack, if someone's going into your systems to steal data, it's because they think that they can get money from reselling that data.
00:02:33.199 --> 00:02:44.879
And in fact, they're going to have um an easier and more profitable life by working as a as a hacker, hacking into systems than than they would flipping burgers or something or something similar.
00:02:44.879 --> 00:02:47.199
Fundamentally, that's what we're doing.
00:02:47.599 --> 00:02:51.840
Same, uh very similar when it comes to a nation-state attack.
00:02:51.840 --> 00:03:08.800
The nation-state believes that the value that it can get from um getting inside a situation, uh, an organization or a system and achieving certain objectives is worth the cost that it takes of getting in there.
00:03:08.800 --> 00:03:17.120
So that involves, you know, hiring people, having infrastructure, investing time in developing malware or learning techniques.
00:03:17.120 --> 00:03:18.400
All of this costs.
00:03:18.879 --> 00:03:35.199
So if we want to frustrate the attacks, what we want to do is we want to make it as expensive as possible for the attacker to get inside um the organization and fulfill their objectives.
00:03:35.199 --> 00:03:43.919
So this for me is is where the uh honeypots and the honeypot approach really um applies.
00:03:43.919 --> 00:03:47.199
What we want to do is lay decoys.
00:03:47.199 --> 00:03:58.080
So decoy systems, things that that look like the type of juicy system that an attacker is going for, that in fact is is completely fake and is under your control.
00:03:58.080 --> 00:04:04.639
Um, these might be external to your environment or internal in your environment.
00:04:04.639 --> 00:04:13.520
But the important thing is they look and behave similar to what the attacker is expecting to find.
00:04:13.520 --> 00:04:32.240
Um, same with uh people, uh, create fake profiles, um, you know, create fake social media presence, create fake email addresses, um, or use expired domains um that you might have, or buy expired domains.
00:04:32.480 --> 00:04:41.519
But anything that looks like a nice juicy target for for the bad guy, firstly, it's going to chew up their time.
00:04:41.519 --> 00:04:44.800
It increases their their their expenses.
00:04:44.800 --> 00:04:46.959
This is this is our first benefit.
00:04:46.959 --> 00:04:55.279
Secondly, when the attacker interacts with your honeypot, they're telling something about themselves.
00:04:55.279 --> 00:05:00.879
At the very simplest, they're giving away the fact that they're actually conducting an attack.
00:05:00.879 --> 00:05:02.720
Um, and that you can see this.
00:05:02.720 --> 00:05:05.680
So already, this is this is an alert, this is threat intelligence.
00:05:05.680 --> 00:05:07.920
This is telling you something is happening here.
00:05:07.920 --> 00:05:16.079
How the attacker interacts with your honeypot also tells you um about how they are conducting the attack.
00:05:16.079 --> 00:05:28.639
So you get to know that there's an attack in progress, um, or at least being being prepared, you get to know how the attacker is um is attacking, and this you can use to boost your defenses.
00:05:28.879 --> 00:05:46.879
So um think about it in terms of slowing down the attacker, making
Load full transcript
life expensive for them in chewing up their time, and also in terms of generating threat intelligence for yourself that you can then use.
00:05:46.879 --> 00:05:52.079
And of course, don't forget threat intelligence is only any value if you actually put it to good use.
00:05:52.399 --> 00:05:52.879
Yes.
00:05:52.879 --> 00:06:03.920
And I feel like part of this um decision that people have to make on, you know, what route to go down is like the ability to maintain it, right?
00:06:03.920 --> 00:06:08.720
Everyone is already overspent as it is with their resources.
00:06:08.720 --> 00:06:13.360
No one wants to spend their entire day like managing fake objects or things like that.
00:06:13.360 --> 00:06:29.360
Um, so how how do you approach these various impediments so that they're like actually practical and low maintenance and don't create more work for the security team than they do, you know, for the attacker who's the person we're trying to make more work for?
00:06:29.600 --> 00:06:32.959
Well, the great thing about fake objects is they don't do anything.
00:06:32.959 --> 00:06:42.079
Um, you know, go create a fake, a fake executive, you know, put put put their profile on uh on LinkedIn, write a couple of blogs in their names.
00:06:42.079 --> 00:06:42.800
You know what?
00:06:42.800 --> 00:06:54.720
That fake executive doesn't have any expense reports, doesn't need travel uh organizing, isn't going to you know, create an HR incident or whatever.
00:06:54.720 --> 00:06:58.000
They don't do anything, they're just they're just fake, they just sit there.
00:06:58.160 --> 00:07:06.480
Um, so in terms of things, um, certainly um fake profiles and fake individuals, you really don't need to do anything.
00:07:06.480 --> 00:07:10.959
It's a very low value investment um in time.
00:07:10.959 --> 00:07:21.040
Create a social media profile, create um uh a bit of engagement on social media, perhaps a blog, put their email address somewhere, and then just forget about it.
00:07:21.040 --> 00:07:26.879
And maybe in a year's time you might do something to make them look like they they've they've done something.
00:07:26.879 --> 00:07:30.399
But you the the the great thing is you don't need to do a lot.
00:07:30.399 --> 00:07:39.439
Um same thing goes for fake email addresses that you're that you're creating that um attackers might be looking or spidering.
00:07:39.439 --> 00:07:49.759
Um stick a an email address somewhere, maybe on your website, maybe on a post, it doesn't really matter, just stick it somewhere, wait for a bad guy to find it.
00:07:49.759 --> 00:07:53.839
Even better if um it it starts with A.
00:07:53.839 --> 00:07:59.279
So Aaron Aardvark is a very, very good um uh persona.
00:07:59.279 --> 00:08:07.839
Um, because if attackers are going through email addresses from A to Z, which they often do, that's the one that's going to get the attack first.
00:08:07.839 --> 00:08:20.800
Um, you know, maybe not as obvious as Aaron Aardvark, but if you wanted to create a um uh a fake persona whose first name and surname begin begins with A.
00:08:20.800 --> 00:08:23.279
Um and put that somewhere.
00:08:23.279 --> 00:08:24.399
Yeah, have fun with it.
00:08:24.399 --> 00:08:30.399
And then wait, you know, just set up automatic uh monitoring of their of their email address.
00:08:30.399 --> 00:08:44.639
Um and any emails that come through, because you know that this is a fake employee that doesn't do anything, they can't subscribe to email lists, they can't email contacts, sales, partners, customers, whatever.
00:08:44.639 --> 00:08:49.519
Anything that comes through, you know, is spam or is malicious in some format.
00:08:49.519 --> 00:08:59.759
And if it's come through your your email filters, you know you need to update your email filters, you need you need to um to to to block something.
00:08:59.759 --> 00:09:16.720
Now, this can become a time-saving um process because if you're collecting the phishing attacks first and you're updating your spam filters to say, oh dear, this this shouldn't have come through.
00:09:16.720 --> 00:09:17.360
I'll block that.
00:09:17.360 --> 00:09:42.080
And tell you what, any of these emails that have actually gone through to an inbox, I need to remove, you're then reducing the possibility of your of your users clicking on that phishing attack, engaging it with it, leaking with credentials, which is then gonna require you to respond to an incident, change their change their their login credentials, kick them off, you know, log them out of their sessions, etc.
00:09:42.080 --> 00:09:44.080
etc., which is time consuming.
00:09:44.080 --> 00:09:50.320
Just a little tiny bit um of investment up front makes your life easier.
00:09:50.320 --> 00:09:59.679
It's not gonna attract every single phishing attack or everything which is coming through, but it'll attracts some and it gives you that heads up.
00:10:00.000 --> 00:10:03.600
Very, very similar um issue when it comes to fake systems.
00:10:03.600 --> 00:10:08.320
If you're setting them up, you know, create a fake document um server.
00:10:08.320 --> 00:10:10.399
Nobody uses it.
00:10:10.399 --> 00:10:11.519
This is the whole point.
00:10:11.519 --> 00:10:15.200
As soon as someone engages with it, you know that it's bad.
00:10:15.200 --> 00:10:22.639
Um, kick the device off the network, quarantine it because you know that something that shouldn't have happened has happened.
00:10:22.639 --> 00:10:29.440
So you're saving yourself time by detecting things early.
00:10:29.440 --> 00:10:39.279
Um, other things, you know, seed some API keys for a you know very low-value honey pot cloud environment.
00:10:39.279 --> 00:10:43.440
So if somebody steals those, again, you know that you've got an incursion.
00:10:43.440 --> 00:10:44.960
You've know that you've got a problem.
00:10:44.960 --> 00:10:55.200
And it's giving you a heads up and a head start on resolving um an incident and stopping that incident becoming worse.
00:10:55.200 --> 00:11:03.519
Because the the the later you discover an incident, the more mess there is, the more cleanup, the more timing that it's going to be.
00:11:03.840 --> 00:11:04.240
Yeah.
00:11:04.240 --> 00:11:09.440
Um, and you have specifically written about AI honeypots before.
00:11:09.440 --> 00:11:12.080
So talk to me a little bit about this.
00:11:12.080 --> 00:11:19.200
In in this time when so many adversaries are trying to use AI to speed up their own operations or to automate them.
00:11:19.200 --> 00:11:21.759
Um, how does it help us stay ahead?
00:11:22.159 --> 00:11:25.279
Well, we're using the attacker's tools against them.
00:11:25.279 --> 00:11:34.879
Um, you know, if an attacker is using AI to um crawl across your networks and systems and discover them, well, great.
00:11:34.879 --> 00:11:42.000
Let's just use AI to create fake systems for their fake aid, for their AI agent to discover and engage with.
00:11:42.000 --> 00:11:48.159
And we're just chewing up those resources um of the attacker.
00:11:48.159 --> 00:11:53.840
So there, for a long time, there's been all sorts of software projects about creating honeypots.
00:11:53.840 --> 00:12:09.679
So these are systems that that look like genuine systems to a bad guy, um, but in fact are used for collection of threat intelligence um by by defenders and also for chewing up the the time of the uh of the attacker.
00:12:09.679 --> 00:12:13.840
The difficulty is um uh honeypots aren't real.
00:12:13.840 --> 00:12:25.279
You know, if you know what you're doing and you engage with a honeypot, it doesn't take you enormously long time to work out that there's something not right here.
00:12:25.279 --> 00:12:37.200
It's kind of like engaging with uh with a fake AI customer service agent, you know, uh only a few minutes before you realizing, hang on, you're a robot, aren't you?
00:12:37.440 --> 00:12:37.759
Yep.
00:12:38.080 --> 00:12:53.120
So if you have a skilled individual, a human, maybe it's gonna take them three minutes, five minutes tops to discover that they're that they're engaging with with something which isn't quite right.
00:12:53.120 --> 00:12:57.840
AI agents don't have this awareness.
00:12:57.840 --> 00:13:05.519
So we can use those capabilities uh or that lack of capability and lack of awareness against the AI agents.
00:13:05.519 --> 00:13:18.399
Um, like you can have a fake customer talking to a fake customer service agent and they'll just ping-pong between each other, you know, forever, neither of them aware that they're that they're engaging with a with a robot.
00:13:18.799 --> 00:13:26.480
So software projects that create honeypots and like like genuine honeypots, um uh are a thing.
00:13:26.480 --> 00:13:27.919
They're really, really good.
00:13:27.919 --> 00:13:39.360
However, as with any software development, it takes a while to develop the software to something which looks um um reasonable, behaves reasonably.
00:13:39.360 --> 00:13:44.639
Great thing
about AI is that speed of development.
00:13:44.639 --> 00:13:57.200
If you're creating a mission critical application, you know, like maybe AI development vibe coding isn't necessarily what you want, you know, that there's two schools of thought on this.
00:13:57.360 --> 00:14:01.120
But when you're creating a honeypot, it doesn't really matter.
00:14:01.120 --> 00:14:19.600
Um, and we can actually, in the same way that you can, I mean, sometimes I don't know about you, but um sometimes when I procrastinate, I ask an AI agent, you know, be Gandalf and give me um you know wisdom to to encourage me on my quest.
00:14:19.600 --> 00:14:22.960
And it and it talks to me in the guise of Gandalf.
00:14:22.960 --> 00:14:32.240
Um, or something like Winston Churchill is another one, it'll give me a long speech about don't you don't waste a second, young man of your precious life.
00:14:32.240 --> 00:14:34.480
You know, and it goes and it goes on like that.
00:14:34.480 --> 00:14:35.679
And it and it's great fun.
00:14:35.679 --> 00:14:45.360
But in the same way, we can also get the AI to say, pretend that you're a Linux kernel and talk to it using Linux commands.
00:14:45.360 --> 00:14:49.759
Um or as I wrote in the blog, pretend to be a smart fridge.
00:14:49.759 --> 00:14:55.840
And um it'll spin up a file system that looks like a smart fridge.
00:14:56.159 --> 00:15:07.200
It's not going to fool anyone who knows what they're what they're doing or what they're looking for for longer than two minutes, but it will fool an AI agent, and that's what you've got to do.
00:15:07.200 --> 00:15:17.679
Ties up the attacker's resources, tells you that there's someone in your environment looking for something, and tells you what they're doing, how they're doing it, and what their IP address is.
00:15:18.000 --> 00:15:27.519
Yeah, I mean, that's pretty fantastic for, I would say, relatively little work other than the upfront cost of setting it up.
00:15:27.519 --> 00:15:28.000
Yeah.
00:15:28.000 --> 00:15:28.960
That's fantastic.
00:15:29.200 --> 00:15:32.960
And the great, the great thing, you know, approach approach these things as disposable.
00:15:32.960 --> 00:15:36.000
Um, it's something you can run up in a morning.
00:15:36.000 --> 00:15:53.200
Um, you know, don't think of it as as a long-term thing that's really absolutely going to protect your um you know, your your precious AI-enabled fridges or if alarm and or make it you know look like a nuclear power station.
00:15:53.200 --> 00:15:54.799
Um that's not what it's for.
00:15:54.799 --> 00:15:57.360
It's about giving you an early heads up.
00:15:57.360 --> 00:16:04.639
Yeah, something's wrong, something's engaging with my honeypot that superficially looks like a key system.
00:16:04.639 --> 00:16:07.440
Um, that should ring alarm bells.
00:16:07.440 --> 00:16:15.600
We we need to um set off our instant response um procedure here because we can we can detect it early.
00:16:15.919 --> 00:16:28.559
Yeah, and
I'm sure for people it is kind of hard to quantify like how much time an attacker didn't spend on our network because of a tar pit or honey credentials.
00:16:28.559 --> 00:16:34.000
So, like, how should teams be thinking about like measuring the success of these things?
00:16:34.000 --> 00:16:39.759
Is it, you know, the information that you are able to get from an adversary interacting with it?
00:16:39.759 --> 00:16:41.759
Is it the lack thereof?
00:16:41.759 --> 00:16:43.679
Um, what what are your thoughts?
00:16:44.000 --> 00:16:57.200
I would justify it to management that every interaction you get with a threat actor with your honeypot is an interaction that wasn't made with a key system.
00:16:57.200 --> 00:17:10.000
And if you can frame it in that way, is we we, you know, this attacker was looking for an executive, or this attacker was looking for part of our critical national infrastructure.
00:17:10.000 --> 00:17:15.680
Instead, they found a fake profile, they found a fake system.
00:17:15.680 --> 00:17:19.759
So that attack was actually thwarted.
00:17:19.759 --> 00:17:24.720
Um and and that that's the real benefit.
00:17:24.720 --> 00:17:42.880
And when you start framing things in those ways, um, if you know the average cost of um resolving an incident when it's affected your key system or when it's got access to the company bank account, um, and you can point out, well, actually, we saved you this.
00:17:42.880 --> 00:17:48.559
That's that's where you really start demonstrating return on investment for these things.
00:17:49.599 --> 00:17:50.079
Wow.
00:17:50.079 --> 00:17:53.759
Any any final words of wisdom, things that you'd like to say?
00:17:54.160 --> 00:17:55.759
I think most importantly, have fun.
00:17:55.759 --> 00:17:56.799
It's a laugh.
00:17:56.799 --> 00:18:00.799
I love setting breadcrumbs like like this.
00:18:00.799 --> 00:18:26.240
Um, and I think in you know, in a world, and I think we're moving to that eventually, where where all the attackers are agentic AI, yeah, if we're using AI to create a world full of fake systems, the whole premise of cyber attacks falls down because if you've got you know 100 systems and only one of those is genuine, you never go, you know.
00:18:26.240 --> 00:18:34.799
It's like you've got a 99% chance of not getting it right, of one in a one in a hundred chance of of connecting to the right system.
00:18:34.799 --> 00:18:45.200
Suddenly, yeah, cyber attacks become untenable because all you're doing is interacting with um uh with fake systems.
00:18:45.200 --> 00:18:59.279
Um, you know, it's a lot of social media now is also fake uh profiles saying fake things, and you can spend your life, you know, arguing politics with a fake uh um AI uh propaganda bot.
00:18:59.279 --> 00:19:00.720
I mean, why, why, why bother?
00:19:00.720 --> 00:19:07.279
Just set up another fake um AI propaganda bot and happily each other.
00:19:07.759 --> 00:19:15.119
Thank God I'm not like 14, 15, or 16 in this time when like I used to spend so much time arguing on the internet.
00:19:15.680 --> 00:19:18.000
Yeah, yeah, yeah, absolutely.
00:19:18.000 --> 00:19:19.440
Um yeah, why?
00:19:19.440 --> 00:19:20.960
Don't don't don't even bother.
00:19:20.960 --> 00:19:31.119
But but we can tie up the resources of those who who who are doing that, make it a less lucrative um attack for them.
00:19:31.119 --> 00:19:34.160
Um and yeah, and have fun.
00:19:34.319 --> 00:19:34.480
Yeah.
00:19:35.200 --> 00:19:35.519
Have fun.
00:19:35.519 --> 00:19:36.240
Be alive.
00:19:36.960 --> 00:19:40.720
How often do you get to mess around with attackers like this?
00:19:40.720 --> 00:19:41.759
Come on.
00:19:42.400 --> 00:19:52.319
Do you know what I think it's like people take AI seriously and think AI is like this fountain of wisdom, you know, giving them information.
00:19:52.319 --> 00:19:56.960
We need to be reverential to it and polite because it's going to take over the world.
00:19:56.960 --> 00:20:00.400
It's like, no, get into this stupid stuff, you know.
00:20:00.400 --> 00:20:10.480
Train the AI to be a fridge, train it, you know, train it to be a fake nuclear power station, um, you know, a fake pump, uh a dam, you know, whatever.
00:20:10.480 --> 00:20:18.079
Um, or or a fake executive, um, a fake uh executive assistant who has budgetful budgetary control.
00:20:18.079 --> 00:20:26.319
Um, yeah, create create use use AI for laying traps for the bad guys.
00:20:26.960 --> 00:20:29.279
Martin, thank you so much for joining me today.
00:20:29.279 --> 00:20:32.559
We miss you so much, but it was great to have you back.
00:20:32.880 --> 00:20:34.720
Oh, it's lovely talking to you.
00:20:34.720 --> 00:20:35.920
Yeah, it's great.
00:20:36.559 --> 00:20:37.119
Yeah.
00:20:37.119 --> 00:20:40.480
Well, listeners, we have so much planned for this month.
00:20:40.480 --> 00:20:46.559
We have a blog with contributions and advice from different researchers out, including Martin.
00:20:46.559 --> 00:20:48.720
I'll put the link to that in the notes.
00:20:48.720 --> 00:20:56.000
Um, we have several videos, podcasts, and at the end of the month, we also have the IR quarterly trends report coming out.
00:20:56.000 --> 00:20:57.440
So stay tuned.
00:20:57.440 --> 00:21:00.079
A lot of useful info is coming your way.
00:21:00.079 --> 00:21:03.039
Thanks for listening and stay safe out there.