1
00:00:04,400 --> 00:00:07,880
Welcome to Crying Out Cloud 
podcast that will make you 

2
00:00:07,880 --> 00:00:11,760
laugh, cry and reconsider all of
your cloud security fears. 

3
00:00:11,840 --> 00:00:14,480
I'm Eden. 
I'm here with an amazing Co host

4
00:00:14,480 --> 00:00:16,239
Amitai. 
Hello. 

5
00:00:16,960 --> 00:00:20,360
And today we have a super 
special guest and we're going to

6
00:00:20,360 --> 00:00:22,720
dive deep into detection 
engineering. 

7
00:00:22,720 --> 00:00:26,080
And yeah, I'm excited. 
Welcome Alex Furtado. 

8
00:00:26,960 --> 00:00:29,040
Hello, thank you so much for 
having me. 

9
00:00:29,040 --> 00:00:32,680
I this is super, super awesome 
being on this side of the house.

10
00:00:32,680 --> 00:00:36,560
I think it's my first time after
hosting, gosh, I don't know, 

11
00:00:36,560 --> 00:00:39,880
like 50 dispatch episodes. 
It's just super, super hyped to 

12
00:00:39,880 --> 00:00:41,800
be here. 
Like yes, that we get you for 

13
00:00:41,800 --> 00:00:43,280
your first time being the 
interview. 

14
00:00:43,480 --> 00:00:45,320
Where are you calling us from 
today? 

15
00:00:45,920 --> 00:00:49,800
I am calling in from Chicago. 
We we just had our first snow. 

16
00:00:49,800 --> 00:00:52,760
So a lot of lot of firsts today.
Yeah, you can see it in the 

17
00:00:52,760 --> 00:00:54,400
background. 
We were expecting a big 

18
00:00:54,400 --> 00:00:56,440
snowstorm, but it's actually 
really sunny. 

19
00:00:56,440 --> 00:01:00,200
It's really clear skies. 
So very, very stoked, very very 

20
00:01:00,200 --> 00:01:02,440
good day for her to film A Wiz 
podcast. 

21
00:01:02,800 --> 00:01:03,920
We love it. 
We love. 

22
00:01:03,920 --> 00:01:04,879
That. 
Magical day. 

23
00:01:05,200 --> 00:01:06,880
Magical day. 
Cool. 

24
00:01:06,880 --> 00:01:09,680
So for those who don't know, 
Alex is currently the head of 

25
00:01:09,680 --> 00:01:13,560
Community detection engineering 
podcast host and a technical 

26
00:01:13,560 --> 00:01:14,960
writer. 
And the logic. 

27
00:01:15,320 --> 00:01:19,000
She also authored their State of
Detection Engineering report in 

28
00:01:19,000 --> 00:01:22,760
collaboration with the Sand 
Institute and various pieces for

29
00:01:22,760 --> 00:01:25,560
the Thor Collective. 
Alex sits at the rear 

30
00:01:25,560 --> 00:01:28,360
intersection of technical 
product and threat research 

31
00:01:28,360 --> 00:01:31,880
marketing, specifically within 
the Detection engineering 

32
00:01:31,880 --> 00:01:34,400
program. 
In her past, she's been a seam 

33
00:01:34,400 --> 00:01:39,080
analyst, CMSC and detection 
engineering content creator. 

34
00:01:40,800 --> 00:01:43,120
That's right, that's right. 
We're, we're actually in the in 

35
00:01:43,120 --> 00:01:47,600
the midst of shipping out our 
our state of detection 

36
00:01:47,600 --> 00:01:50,120
engineering survey. 
So for those tuning in, 

37
00:01:50,120 --> 00:01:53,200
definitely got in front of that 
survey so that we can get the 

38
00:01:53,200 --> 00:01:56,080
best, the best data there is as 
we all know very well. 

39
00:01:56,400 --> 00:02:01,040
OK, so we like to kick off 
episodes with guests with 

40
00:02:01,040 --> 00:02:04,800
something we call Cloud 
Confessions to get to know the 

41
00:02:04,840 --> 00:02:08,560
people on our podcast. 
Are you up for a little Cloud 

42
00:02:08,560 --> 00:02:12,040
Confessions? 
Let's let's see if I if I know 

43
00:02:12,040 --> 00:02:16,800
any. 
OK, so you have an article 

44
00:02:16,800 --> 00:02:20,520
called Quiet Loud and in the log
files where you reflect on 

45
00:02:20,720 --> 00:02:24,360
introverts and extroverts in 
detection engineering. 

46
00:02:25,160 --> 00:02:27,760
Are you? 
Would you describe yourself as 

47
00:02:27,760 --> 00:02:32,800
an introvert or an extrovert? 
See, I like to think of myself 

48
00:02:32,800 --> 00:02:38,600
as an introverted extrovert, but
everyone who knows me will turn 

49
00:02:39,280 --> 00:02:42,760
literally burn that down. 
They think that's very much an 

50
00:02:42,760 --> 00:02:45,000
extrovert. 
Although I, I do, I, it's, it's 

51
00:02:45,040 --> 00:02:48,320
really in the article, I talk 
about what that really means. 

52
00:02:48,320 --> 00:02:51,440
And it's really about how you 
recharge your energy. 

53
00:02:51,960 --> 00:02:57,240
And I do like to kind of go 
leave the crowd, recharge and 

54
00:02:57,240 --> 00:03:01,040
then and then get back at it. 
So I but but in the traditional 

55
00:03:01,040 --> 00:03:03,040
sense, I think I am officially 
an extrovert. 

56
00:03:03,360 --> 00:03:06,600
I'm also a social introvert. 
That's how I describe myself. 

57
00:03:07,040 --> 00:03:08,600
Yeah, and and what about you, 
Amitai? 

58
00:03:08,760 --> 00:03:12,000
Oh, I get. 
I get tired very quickly in any 

59
00:03:12,000 --> 00:03:17,200
social interaction, as as Eden 
and our producer well know. 

60
00:03:17,760 --> 00:03:20,960
The social battery runs out 
interest Interesting for a for a

61
00:03:20,960 --> 00:03:22,920
podcast. 
Oh yes. 

62
00:03:23,320 --> 00:03:25,840
We just we sliced our timing 
wisely. 

63
00:03:26,600 --> 00:03:31,760
Next question, where do you 
stand on the term thrunting? 

64
00:03:33,200 --> 00:03:37,120
Very controversial topic, huh? 
Well, at first I was against the

65
00:03:37,120 --> 00:03:39,720
idea. 
I was against not making 

66
00:03:39,720 --> 00:03:45,040
thrunting a thing. 
But as I've met the crew over at

67
00:03:45,040 --> 00:03:49,440
the Thor Collective and I was 
shipped some pretty, pretty sick

68
00:03:49,440 --> 00:03:54,080
stickers for my laptop, it's had
the term has now grown on me. 

69
00:03:54,280 --> 00:03:58,360
I think we should make light of 
the situation, not take it not, 

70
00:03:58,360 --> 00:04:01,160
you know, not take it things so 
seriously and just have a little

71
00:04:01,160 --> 00:04:03,480
fun with it. 
So I'm officially a part of the 

72
00:04:03,480 --> 00:04:08,800
thrunting crew. 
Right, I'm also trying to to 

73
00:04:08,800 --> 00:04:12,040
personally adopt within with 
both thrunting and Thrintel. 

74
00:04:12,560 --> 00:04:14,360
But Thrintel, that's nice 
meeting meeting. 

75
00:04:14,360 --> 00:04:16,959
Some push back, though. 
OK, next question. 

76
00:04:16,959 --> 00:04:20,079
You're an avid trekker we we 
hear. 

77
00:04:20,720 --> 00:04:23,560
What's your favorite track you 
did ever or recently? 

78
00:04:23,560 --> 00:04:27,720
If you can't think of ever. 
I so I've recently I did the 

79
00:04:28,880 --> 00:04:31,640
kawaii track. 
I forgot what it's called. 

80
00:04:31,640 --> 00:04:35,440
It's a really long name, a 
really long Hawaiian name, but 

81
00:04:35,440 --> 00:04:42,440
it was a total of 8 hours across
different climates, across 

82
00:04:42,480 --> 00:04:46,400
uphill and mountain, right by a 
lake up up towards waterfall. 

83
00:04:46,440 --> 00:04:48,320
And it was, it was truly 
magical. 

84
00:04:48,600 --> 00:04:52,480
But it still does not beat the 
best track I've ever done, which

85
00:04:52,480 --> 00:04:56,960
it was through the Salcante 
Andes Mountains in Peru. 

86
00:04:57,320 --> 00:05:01,760
So we've we've read you describe
yourself as in our research as 

87
00:05:01,760 --> 00:05:03,560
an unprofessional parallel 
Parker. 

88
00:05:04,400 --> 00:05:06,720
Where, where did you guys, where
are you guys finding this? 

89
00:05:06,720 --> 00:05:09,400
We we can't reveal our sources. 
Those are confidential. 

90
00:05:09,760 --> 00:05:14,600
However, a parallel Parker 
sounds to me like it's talking 

91
00:05:14,600 --> 00:05:17,080
about Spider Man from another 
dimension. 

92
00:05:17,720 --> 00:05:18,200
Oh. 
I see. 

93
00:05:18,200 --> 00:05:19,800
Yeah. 
B. 

94
00:05:20,360 --> 00:05:23,720
What are your What are some of 
your best parallel parking 

95
00:05:23,800 --> 00:05:26,480
experiences? 
There's a lot of parallel 

96
00:05:26,480 --> 00:05:29,920
parking situations you'll find 
yourself in big urban cities 

97
00:05:29,920 --> 00:05:35,760
like Chicago and every single 
time I've tried to parallel park

98
00:05:36,320 --> 00:05:39,840
by using the playbook that they 
teach you in driver school when 

99
00:05:39,840 --> 00:05:42,400
you're 16 years old. 
Till this day, I can't figure it

100
00:05:42,400 --> 00:05:44,480
out. 
I don't know how I managed to 

101
00:05:44,480 --> 00:05:48,760
pass to pass it when I was 16. 
I, I always have to have 

102
00:05:48,760 --> 00:05:52,080
whoever's accompanying me like 
me get out of the car them do it

103
00:05:52,080 --> 00:05:54,480
for me. 
I I never have figured it out. 

104
00:05:55,080 --> 00:05:57,480
OK. 
Last question is, if you are a 

105
00:05:57,480 --> 00:05:59,520
vulnerability, what type would 
you be? 

106
00:05:59,720 --> 00:06:03,320
Oh that's a good one. 
I recently saw your mashup which

107
00:06:03,320 --> 00:06:08,960
I really loved but you know what
I was thinking about has that 

108
00:06:08,960 --> 00:06:14,160
has anyone ever thought about 
being in a vulnerability for a 

109
00:06:14,160 --> 00:06:17,440
SIM? 
Like if you bring down the 

110
00:06:17,440 --> 00:06:21,640
watchdog, right, the monitor, 
you're kind of, it really turns 

111
00:06:21,640 --> 00:06:23,880
it off for everybody. 
You can really do some damage, 

112
00:06:24,160 --> 00:06:25,320
right? 
It's like a movie scene. 

113
00:06:25,320 --> 00:06:27,600
You break into the control room,
you knock out the guy that's 

114
00:06:27,600 --> 00:06:29,240
monitoring, that's monitoring 
everything. 

115
00:06:29,240 --> 00:06:32,640
You're free reign for a while. 
So you might be like an like a 

116
00:06:33,680 --> 00:06:37,400
authentication bypass in Splunk.
Yeah, exactly. 

117
00:06:37,760 --> 00:06:42,200
OK, graduated from Cloud 
Confessions on to real 

118
00:06:42,200 --> 00:06:45,760
questions. 
At most I think of our listeners

119
00:06:45,840 --> 00:06:48,320
are in the cloud security 
demographic. 

120
00:06:50,240 --> 00:06:54,280
So to start, what do you think 
some of the key differences or 

121
00:06:54,280 --> 00:06:58,760
challenges are when it comes to 
detection engineering in the 

122
00:06:58,760 --> 00:07:01,960
cloud versus other types of 
environments? 

123
00:07:02,840 --> 00:07:05,040
Yeah, it's, it's a good 
question. 

124
00:07:05,120 --> 00:07:10,000
The thing about cloud is that it
really does break all of the 

125
00:07:10,080 --> 00:07:13,520
traditional detection 
engineering rules, right, with 

126
00:07:14,400 --> 00:07:17,880
as it relates to the predictable
perimeters, right, the static 

127
00:07:17,880 --> 00:07:22,480
IP's, the logs that really stay 
put and in the cloud with 

128
00:07:22,480 --> 00:07:26,040
everything so ephemeral and 
decentralized and abstracted 

129
00:07:26,040 --> 00:07:29,520
behind AP is detection 
engineering really becomes less 

130
00:07:29,520 --> 00:07:34,040
about chasing signatures, more 
about, you know, piercing 

131
00:07:34,040 --> 00:07:38,160
together activities of millions 
of moving parts, right? 

132
00:07:39,320 --> 00:07:42,360
I really four things identity 
sprawl. 

133
00:07:42,560 --> 00:07:48,360
We saw in the Mandia M trends 
that stolen credentials leapfrog

134
00:07:48,640 --> 00:07:51,200
phishing, right? 
Phishing the the primary attack 

135
00:07:51,200 --> 00:07:54,400
vector in endpoint environments 
and identity really has become 

136
00:07:54,400 --> 00:07:59,120
the new way in and you're not 
just obviously watching 

137
00:07:59,120 --> 00:08:01,120
endpoints anymore. 
You're watching, you know, 

138
00:08:01,120 --> 00:08:06,440
service accounts and federation 
users and IAM roles and that 

139
00:08:06,440 --> 00:08:10,360
just has levelled up the game 
really for detection engineers. 

140
00:08:10,560 --> 00:08:13,480
Not to mention that it's so 
temporary, right? 

141
00:08:13,480 --> 00:08:15,920
What what detection can be run 
against. 

142
00:08:16,440 --> 00:08:19,520
With everything being so elastic
and workloads spinning up and 

143
00:08:19,520 --> 00:08:23,240
down all the time, you really 
can't rely on, you know, static,

144
00:08:23,360 --> 00:08:26,280
static artifacts. 
The time to live on those are 

145
00:08:26,280 --> 00:08:30,000
just so short lived, right? 
So you have to correlate 

146
00:08:30,000 --> 00:08:33,480
activity by cloud, resource by 
cloud account and adapt in real 

147
00:08:33,480 --> 00:08:35,039
time. 
Otherwise you're kind of toast. 

148
00:08:35,159 --> 00:08:37,880
What comes to mind too is the 
vendor certs for, sorry, the 

149
00:08:37,880 --> 00:08:40,840
cloud certs. 
No better time to really freshen

150
00:08:40,840 --> 00:08:45,160
up on those because each flavor 
of cloud comes with its own 

151
00:08:45,160 --> 00:08:48,320
nuances, right? 
And I've had customers that have

152
00:08:48,480 --> 00:08:51,520
all of them, like everyone can 
sit with us like all inclusive 

153
00:08:51,520 --> 00:08:54,920
of all clouds and which I'm 
really jealous of their budget. 

154
00:08:54,920 --> 00:08:59,760
But each cloud provider has its 
own what like schema, its own 

155
00:08:59,760 --> 00:09:03,000
telemetry, nomenclature, API 
behavior. 

156
00:09:03,360 --> 00:09:06,400
And its own like thousands of 
services, with hundreds more 

157
00:09:06,400 --> 00:09:08,240
coming out every year. 
Yeah. 

158
00:09:08,240 --> 00:09:15,760
And while AWSGCP Azure might all
log the same event, it still 

159
00:09:15,760 --> 00:09:19,080
does it differently, right? 
And standardizing that at scale 

160
00:09:19,080 --> 00:09:22,840
is really tough. 
So yeah, you, I think you 

161
00:09:23,000 --> 00:09:26,600
detection engineers have to rely
on behavior driven logic, 

162
00:09:26,600 --> 00:09:31,040
identity aware detections, 
temporary aware detections time,

163
00:09:31,040 --> 00:09:33,920
the timing of everything is, is 
more important now more than 

164
00:09:33,920 --> 00:09:35,360
ever. 
And just your detection 

165
00:09:35,360 --> 00:09:37,680
engineering chops just have to 
really stay up up to date. 

166
00:09:37,680 --> 00:09:43,480
So cloud certs are the way you. 
Think there are any advantages 

167
00:09:43,480 --> 00:09:46,440
like things that might be easier
in the cloud as opposed to to 

168
00:09:46,440 --> 00:09:50,080
let's say on Prem? 
Yes, in terms of I think the 

169
00:09:50,080 --> 00:09:53,920
docks, the docks have are much, 
much better in the cloud. 

170
00:09:53,920 --> 00:09:58,680
The the accesses support to them
is much better as well. 

171
00:09:59,400 --> 00:10:03,520
And then the community around 
around it all too, more more 

172
00:10:03,520 --> 00:10:07,000
than more than other legacy on 
Prem, you know, things. 

173
00:10:07,000 --> 00:10:09,400
So that that's one of those. 
That's why I'm super big on 

174
00:10:09,400 --> 00:10:12,000
community because this is all 
we're writing it in real time, 

175
00:10:12,000 --> 00:10:14,160
the playbooks and and learning 
all together. 

176
00:10:14,160 --> 00:10:16,560
And the more that we do that, 
the more we the more it can be 

177
00:10:16,560 --> 00:10:19,480
searched against. 
It's not like you have different

178
00:10:19,480 --> 00:10:22,160
architectures, but the 
components are a bit more 

179
00:10:22,160 --> 00:10:26,720
uniform which means that there's
more like transferable skills 

180
00:10:26,720 --> 00:10:29,880
and knowledge between different 
organizations that might be like

181
00:10:29,880 --> 00:10:33,560
all be using AWS for example. 
Totally. 

182
00:10:33,800 --> 00:10:37,520
I guess like in the latest buzz 
of a lot of what I think also 

183
00:10:37,520 --> 00:10:41,080
requires a lot of community but 
generative AI. 

184
00:10:41,080 --> 00:10:44,160
What are your thoughts? 
Like AI assisted or AI enabled 

185
00:10:44,240 --> 00:10:46,600
threat hunting and detection 
engineering. 

186
00:10:47,280 --> 00:10:50,440
Yeah. 
So my thing with AI is that I 

187
00:10:50,440 --> 00:10:53,560
think we should be really 
specific about the kind of AI 

188
00:10:53,560 --> 00:10:57,560
that we are using and for what 
And like for what for, right, 

189
00:10:57,880 --> 00:11:00,960
Because that really should 
dictate when it makes sense to 

190
00:11:00,960 --> 00:11:03,320
use it for detection 
engineering, for threat hunting,

191
00:11:03,320 --> 00:11:06,840
whatever. 
Just recently we spoke at Blue 

192
00:11:06,840 --> 00:11:08,920
Team Con. 
Like just because you can use an

193
00:11:08,920 --> 00:11:13,440
LLM for entity extraction, for 
example, doesn't mean you 

194
00:11:13,440 --> 00:11:16,200
should, right? 
Like it's cute, but just use 

195
00:11:16,200 --> 00:11:18,280
rejects and move on with your 
life, right? 

196
00:11:18,680 --> 00:11:22,240
I, I still think the best, the 
best agents, if we're talking 

197
00:11:22,240 --> 00:11:24,520
about Gen. tech, AI is still 
your team. 

198
00:11:25,320 --> 00:11:28,400
It's got a plethora of tribal 
knowledge and business contacts 

199
00:11:28,400 --> 00:11:33,440
and just just plain curiosity 
that that that's never going to 

200
00:11:33,440 --> 00:11:37,360
be ever like replicated. 
That's not to say it can't be 

201
00:11:37,360 --> 00:11:41,400
useful, right? 
It it can truly be useful in 

202
00:11:41,640 --> 00:11:45,200
pattern recognition. 
For example, I think AI eats 

203
00:11:45,200 --> 00:11:49,320
repetitive things for breakfast 
repetitive patterns right, it 

204
00:11:49,320 --> 00:11:51,000
doesn't get bored it doesn't get
tired. 

205
00:11:51,480 --> 00:11:55,640
I'll go even as far as saying it
can offset our, you know, 

206
00:11:55,640 --> 00:11:58,080
analyst long lunch hours or 
vacation time, right? 

207
00:11:58,080 --> 00:12:03,360
I really do think it can. 
But I do see as well the 

208
00:12:03,360 --> 00:12:06,480
pressure it's put against 
detection engineers. 

209
00:12:06,480 --> 00:12:10,560
I've talked about this recently 
with Jay Johnson on the podcast.

210
00:12:11,120 --> 00:12:15,320
But analysts are feeling the 
pressure to level up with AI to 

211
00:12:15,760 --> 00:12:18,560
avoid getting automated out of a
job, right? 

212
00:12:19,520 --> 00:12:23,440
Fear is really valid. 
But I think if we look at it as 

213
00:12:23,440 --> 00:12:27,280
not getting replaced but 
delegating the drudge work, 

214
00:12:27,600 --> 00:12:30,560
maybe starting with the things 
that nobody on your team likes 

215
00:12:30,560 --> 00:12:32,760
doing and trying to automate 
that because that just makes 

216
00:12:32,760 --> 00:12:34,840
everybody happier. 
If you have the choice between 

217
00:12:34,840 --> 00:12:38,840
hiring someone who has made 
themselves themselves very 

218
00:12:38,840 --> 00:12:42,120
efficient by adopting AI and 
teaching themselves to use 

219
00:12:42,120 --> 00:12:45,760
generative AI tooling as opposed
to someone who hasn't and might 

220
00:12:45,760 --> 00:12:50,040
be wasting time on doing things,
then you know, you might you 

221
00:12:50,040 --> 00:12:53,440
might lean towards hiring the 
person who's who's learned to 

222
00:12:53,440 --> 00:12:56,280
rely rely on AI. 
I think generally the sentiment 

223
00:12:56,280 --> 00:13:00,000
nowadays is we're not replacing 
people with AI, we're replacing 

224
00:13:00,000 --> 00:13:01,440
people with people who use AI. 
The. 

225
00:13:01,480 --> 00:13:03,520
People who use AI, yeah, Oh, I 
love that. 

226
00:13:03,520 --> 00:13:05,120
I've I've seen that many times 
too. 

227
00:13:05,520 --> 00:13:10,320
And I mean, because if you look 
at when it is effective, it 

228
00:13:10,320 --> 00:13:14,440
really is really good at a lot 
of things like for example, the 

229
00:13:14,440 --> 00:13:16,680
R miter attack on version what, 
18. 

230
00:13:16,680 --> 00:13:20,800
Now we all have really accepted 
that as our common language, 

231
00:13:20,800 --> 00:13:22,120
right? 
That's the universal language of

232
00:13:22,120 --> 00:13:24,320
detection engineers. 
That's that's the official one. 

233
00:13:24,320 --> 00:13:28,280
I would go as far as saying, and
AI is really good at making that

234
00:13:28,280 --> 00:13:31,960
it's context, right? 
It's Bible and, and, and, and 

235
00:13:31,960 --> 00:13:34,200
only using that to make its 
decisions. 

236
00:13:34,400 --> 00:13:37,680
Also, like I was talking about a
lot of these, a lot of these 

237
00:13:37,680 --> 00:13:41,720
modern like languages are well 
documented telemetry sources, 

238
00:13:41,720 --> 00:13:45,000
especially cloud are well well 
documented. 

239
00:13:45,000 --> 00:13:47,960
And if you give that as context,
it's cheaper than ever to 

240
00:13:47,960 --> 00:13:50,480
replicate demo data to test 
against your detections. 

241
00:13:50,800 --> 00:13:54,640
Thread Intel, you guys have an 
amazing thread Intel team that's

242
00:13:55,160 --> 00:13:57,800
given us all types of amazing 
cloud SEC research. 

243
00:13:58,040 --> 00:14:00,280
And you pointed, you pointed to 
that too. 

244
00:14:00,320 --> 00:14:03,400
And it you're telling me if you,
if I can prompt an LLM with 

245
00:14:03,400 --> 00:14:07,040
MITRE, with a target data source
with an incredible blog post and

246
00:14:07,040 --> 00:14:09,360
it'll can kick out a detection 
rule, why wouldn't I want that? 

247
00:14:09,640 --> 00:14:13,120
But you can also like highlights
were also then customization has

248
00:14:13,120 --> 00:14:15,840
to come in and has to be 
relevant to your environment. 

249
00:14:15,840 --> 00:14:19,080
I'm it's like the the big depth 
of knowledge that does exist. 

250
00:14:19,080 --> 00:14:22,280
So like how do you then take 
that and still like customize 

251
00:14:22,280 --> 00:14:25,320
and personalize it for your own 
environment? 

252
00:14:25,960 --> 00:14:30,520
Well, that is why a human in the
loop is still so important 

253
00:14:30,800 --> 00:14:34,320
because while an LLM can get 
your detection at least 80% of 

254
00:14:34,320 --> 00:14:37,040
the way there, it's up. 
Now it's up to the detection 

255
00:14:37,040 --> 00:14:40,400
engineer to really make sure 
that it's customized to that 

256
00:14:40,400 --> 00:14:41,520
environment. 
Why? 

257
00:14:41,520 --> 00:14:44,480
Because they know that 
environment the best, right? 

258
00:14:44,480 --> 00:14:47,120
They will. 
A vendor will never know it more

259
00:14:47,120 --> 00:14:49,200
than you do. 
You know where all where 

260
00:14:49,200 --> 00:14:51,200
everything is, you know where 
all your cups are. 

261
00:14:51,440 --> 00:14:54,200
You know where all your jewelry 
in your houses and all the doors

262
00:14:54,280 --> 00:14:57,200
to get their organizations that 
do build their own detections 

263
00:14:57,200 --> 00:14:59,760
are far better equipped to 
protect their org than anybody 

264
00:14:59,760 --> 00:15:01,960
else that just uses 
out-of-the-box content. 

265
00:15:02,240 --> 00:15:05,680
And in the state of of detection
engineering report has shows all

266
00:15:05,680 --> 00:15:07,360
of the data to prove that as 
well. 

267
00:15:07,800 --> 00:15:10,880
So you saw correlation between 
like successful detection 

268
00:15:10,880 --> 00:15:14,000
engineering teams and and teams 
that were using customization? 

269
00:15:14,480 --> 00:15:19,280
Yes, exactly. 
The, the, the data points 

270
00:15:19,280 --> 00:15:25,960
exactly were about 64%. 
The detection engineers that 

271
00:15:25,960 --> 00:15:29,600
took the report, they've 
reported having problems with 

272
00:15:29,800 --> 00:15:32,360
our with high fault with high 
false positive rates from vendor

273
00:15:32,360 --> 00:15:37,680
provided detections and mostly 
relying on the detection 

274
00:15:37,680 --> 00:15:39,760
engineers to build their own 
custom detections. 

275
00:15:40,080 --> 00:15:43,360
Right. 
And is that also I saw a 

276
00:15:43,360 --> 00:15:46,120
correlation with when we asked 
them what they like doing the 

277
00:15:46,120 --> 00:15:48,920
most, they said that they love, 
they love to build, to build 

278
00:15:48,920 --> 00:15:50,080
their own detections in the 1st 
place. 

279
00:15:50,080 --> 00:15:54,360
They love to threat hunt. 
So if you pair that with both 

280
00:15:54,360 --> 00:15:57,160
data points just is a win win 
for everybody in security, 

281
00:15:57,160 --> 00:15:59,680
right? 
I think that really showed the 

282
00:15:59,680 --> 00:16:03,240
rise of this role because just 
10 years ago we didn't really 

283
00:16:03,240 --> 00:16:05,680
see detection engineering roles 
on LinkedIn or people really 

284
00:16:05,680 --> 00:16:09,880
investing in building out that 
that we were probably doing it, 

285
00:16:09,880 --> 00:16:12,000
but maybe they were called 
security analyst which. 

286
00:16:12,240 --> 00:16:14,320
Is which is actually interesting
because I mean like someone had 

287
00:16:14,320 --> 00:16:17,760
to coin that someone had to like
say like this is a new type of 

288
00:16:17,760 --> 00:16:19,040
job. 
I actually don't know the 

289
00:16:19,040 --> 00:16:22,080
history there, like who actually
was the first like group of 

290
00:16:22,080 --> 00:16:24,120
people to say like this is a new
thing. 

291
00:16:24,280 --> 00:16:28,160
Well, I, I like to think we made
a big impact to, to, to the rise

292
00:16:28,160 --> 00:16:30,320
of that role, but I would, it 
would be, it would be a fun 

293
00:16:30,320 --> 00:16:32,240
thing to go back and see who 
actually did that. 

294
00:16:32,240 --> 00:16:34,480
I'm sure the data is there. 
I'm sure I LinkedIn, shout out 

295
00:16:34,480 --> 00:16:36,120
to LinkedIn. 
Hopefully they can, maybe they 

296
00:16:36,120 --> 00:16:38,560
can help us figure it out. 
Oh, we have a joiner. 

297
00:16:38,720 --> 00:16:43,520
My, my dog joined the pod today.
I named him Summer after Chicago

298
00:16:43,520 --> 00:16:46,960
summers, which unfortunately 
we've have have ended. 

299
00:16:48,320 --> 00:16:52,520
I had an experience recently 
where we were considering 

300
00:16:53,880 --> 00:16:57,640
basically using like a, a ready 
made solution for a certain set 

301
00:16:57,640 --> 00:17:01,640
of detection engineering rules. 
And we sort of we looked into it

302
00:17:01,640 --> 00:17:04,000
and we tried to see if it might 
fit what we need. 

303
00:17:04,319 --> 00:17:06,839
And at the same time we were 
sort of looking into doing it 

304
00:17:06,960 --> 00:17:10,359
ourselves. 
And in the end we decided to do 

305
00:17:10,359 --> 00:17:12,319
it ourselves. 
And I think that I was sort of 

306
00:17:12,480 --> 00:17:15,280
secretly hoping, hoping that 
that would be what we would 

307
00:17:15,280 --> 00:17:19,000
decide to do just because I was 
having fun working on it, even 

308
00:17:19,000 --> 00:17:21,359
if it wasn't necessarily like 
the most efficient way to do 

309
00:17:21,359 --> 00:17:24,200
things. 
Because I think the tendency is 

310
00:17:24,200 --> 00:17:27,680
always to say like if there is 
an existing, you know, off the 

311
00:17:27,680 --> 00:17:29,160
shelf solution, then why not use
that? 

312
00:17:29,440 --> 00:17:31,960
But in this case, it was 
actually fun to sort of try to 

313
00:17:31,960 --> 00:17:34,840
do it ourselves. 
No, you're not alone. 

314
00:17:34,840 --> 00:17:37,080
That from that's directly from 
the trenches. 

315
00:17:37,120 --> 00:17:40,720
I mean this these vendor 
provided rules, they were never 

316
00:17:40,720 --> 00:17:42,760
designed to work for your 
environment. 

317
00:17:42,760 --> 00:17:47,680
They'll, they're never going to 
work, you know, as as soon as, 

318
00:17:47,680 --> 00:17:50,160
as soon as you deploy them, 
there's always work to be done 

319
00:17:50,160 --> 00:17:53,080
to fit them back into how you 
need them. 

320
00:17:53,400 --> 00:17:55,640
And it's just might as well 
start from scratch, right like 

321
00:17:55,640 --> 00:17:57,280
you did. 
You wrote an article called 

322
00:17:57,280 --> 00:18:00,040
Detection Engineering DIY or 
Daitron. 

323
00:18:00,560 --> 00:18:03,040
What was that about? 
Has that relate to this? 

324
00:18:03,600 --> 00:18:08,480
Exactly this really the just how
off detections are super mid. 

325
00:18:08,840 --> 00:18:11,520
They're not cutting it. 
They're, they're super noisy, 

326
00:18:11,520 --> 00:18:14,880
they're brittle, They're 
they're, they aren't built for, 

327
00:18:14,880 --> 00:18:17,480
for you knowing what you need, 
you know, in mind. 

328
00:18:17,480 --> 00:18:21,000
So the fact, and I think that 
really contributed to the rise 

329
00:18:21,000 --> 00:18:24,120
of detection engineers for to 
have somebody strategically, 

330
00:18:24,120 --> 00:18:27,880
strategically think about what 
gaps you have, What do you need 

331
00:18:27,880 --> 00:18:30,520
to build for right, according to
those gaps? 

332
00:18:30,800 --> 00:18:34,600
And what is the best way to do 
it with given what you know, 

333
00:18:34,640 --> 00:18:37,280
feasibility of data you have, 
you have available. 

334
00:18:37,680 --> 00:18:41,960
And so that is really why the 
core of a detection engineer and

335
00:18:41,960 --> 00:18:46,200
if you get to a mature spot when
you treat detections like code, 

336
00:18:46,720 --> 00:18:49,480
that's my next article. 
That and, and start thinking 

337
00:18:49,480 --> 00:18:52,400
about it like a software dev, 
like the way you push out a lot 

338
00:18:52,400 --> 00:18:55,480
of those principles around the 
software development life cycle 

339
00:18:55,680 --> 00:18:59,480
that can even, you know, up 
level and scale the way that you

340
00:18:59,480 --> 00:19:03,880
ship and manage and maintain 
your organization's detections. 

341
00:19:04,520 --> 00:19:06,840
And that's, that's another, 
another win, win for security. 

342
00:19:06,840 --> 00:19:09,200
That's I think that's like the 
next step is 1. 

343
00:19:09,200 --> 00:19:12,160
Are you in a position to build 
your own detections and not just

344
00:19:12,160 --> 00:19:15,280
keep things working or keep the 
lights on based on old 

345
00:19:15,280 --> 00:19:17,880
detections? 
That's number one, That's step 

346
00:19:17,880 --> 00:19:20,680
one. 
And then Step 2 is OK, now that 

347
00:19:20,680 --> 00:19:24,800
you've, you've got maybe like 
2003 thousand detections running

348
00:19:24,800 --> 00:19:26,520
out in production today. 
Can't. 

349
00:19:26,520 --> 00:19:30,640
Are you in a in a systematic 
type of process where you can 

350
00:19:30,640 --> 00:19:34,480
keep them healthy and treat them
like production code in a way? 

351
00:19:34,640 --> 00:19:37,680
It sounds like what you're 
describing is sort of shifting 

352
00:19:37,680 --> 00:19:40,840
like detection engineering 
towards a sort of DevOps area 

353
00:19:41,200 --> 00:19:44,200
where you're sort of managing 
like the meta of detection 

354
00:19:44,200 --> 00:19:47,960
engineering. 
Yeah, no, I I think that is, I 

355
00:19:47,960 --> 00:19:51,440
mean how what's next? 
Like are you talking 6 like 6 

356
00:19:51,440 --> 00:19:53,920
months? 
Are you talking like 1010 years 

357
00:19:53,920 --> 00:19:55,800
where AM and AGI is going to 
take over? 

358
00:19:57,400 --> 00:19:59,520
I think the our, our, our 
measurement should be like, 

359
00:19:59,520 --> 00:20:03,480
what's the what's the next title
on LinkedIn that doesn't exist 

360
00:20:03,480 --> 00:20:05,240
today, That will exist maybe 
next. 

361
00:20:06,440 --> 00:20:08,720
That's a it's a good question. 
It's a good question. 

362
00:20:09,400 --> 00:20:13,160
I, I think I don't, I think 
because it's still so new, I 

363
00:20:13,160 --> 00:20:17,240
don't think there will be a next
one for another another another 

364
00:20:17,240 --> 00:20:20,360
five years. 
What it will turn is it'll go 

365
00:20:20,360 --> 00:20:25,720
from ad hoc to more, a better 
control, more being more 

366
00:20:25,720 --> 00:20:30,080
proactive, right and and, and 
have product ties. 

367
00:20:30,120 --> 00:20:35,040
Detection engineering like in as
a as a system CICD pipelines for

368
00:20:35,040 --> 00:20:38,800
your detections, version control
detections. 

369
00:20:40,000 --> 00:20:43,080
Yeah, testing exactly. 
Just being more proactive 

370
00:20:43,080 --> 00:20:46,840
instead of just waiting, waiting
on your backlog. 

371
00:20:46,880 --> 00:20:50,200
Right, man, Almost like having 
your backlog managing you 

372
00:20:50,200 --> 00:20:53,080
instead of you managing it 
right. 

373
00:20:53,080 --> 00:20:58,920
So I just just this my past 
episode with Spencer Pratt, he 

374
00:20:58,960 --> 00:21:04,000
really opened up my eyes about 
what the new detection engineer 

375
00:21:04,080 --> 00:21:07,160
role is going to turn into. 
Because before it was all about 

376
00:21:07,160 --> 00:21:10,080
tuning your detections and 
keeping them healthy, right 

377
00:21:10,080 --> 00:21:12,600
every time. 
And a rule contributed to a 

378
00:21:12,600 --> 00:21:17,160
false positive. 
But now his job has morphed into

379
00:21:17,680 --> 00:21:21,880
tuning the LLM he's throwing all
of his previous alerts at. 

380
00:21:22,200 --> 00:21:25,480
And it's making like this 
basically historic archive of 

381
00:21:25,480 --> 00:21:28,320
all of their alerts so that as 
new ones come in, it analyzes 

382
00:21:28,320 --> 00:21:32,440
them from the previous. 
And he now instead of tuning his

383
00:21:32,440 --> 00:21:36,320
his alerts right in the logic, 
he's now is tuning the LLM right

384
00:21:36,320 --> 00:21:38,960
on his rag index. 
And he there's all, there's all 

385
00:21:38,960 --> 00:21:41,080
these things that he wants to 
fix like every single day. 

386
00:21:41,080 --> 00:21:44,280
And I was like, that's that just
made you so much better, like 

387
00:21:44,720 --> 00:21:47,960
higher on the market, like to go
to go after and that's, and 

388
00:21:47,960 --> 00:21:50,080
that's this just shows progress 
as an industry. 

389
00:21:50,480 --> 00:21:53,600
Finally, we, we wanted to talk a
bit about the, the, you 

390
00:21:53,600 --> 00:21:56,280
mentioned this a bit before, 
like the importance of community

391
00:21:56,520 --> 00:21:59,000
and detection engineering. 
Oh yeah, sharing knowledge. 

392
00:21:59,760 --> 00:22:03,720
We wanted to hear a bit of your 
take on that and how people can 

393
00:22:03,720 --> 00:22:06,600
sort of participate in in 
community in that sense, like if

394
00:22:06,600 --> 00:22:07,960
they don't know where, where to 
turn. 

395
00:22:08,320 --> 00:22:11,320
Yeah. 
So for dispatch, I'll speak from

396
00:22:11,360 --> 00:22:15,320
from dispatch POV and but also 
Thor Collective is a huge 

397
00:22:15,320 --> 00:22:18,200
proponent of this as well. 
So I will also lead folks down 

398
00:22:18,200 --> 00:22:21,120
that path. 
But for dispatch community is 

399
00:22:21,160 --> 00:22:24,440
everything it because detection 
engineering is still a 

400
00:22:24,440 --> 00:22:30,000
relatively new discipline. 
We are as a, as a whole group of

401
00:22:30,040 --> 00:22:32,680
within the practice, writing it 
in real time. 

402
00:22:32,680 --> 00:22:36,000
When you're building a practice 
that it didn't exist 10 years 

403
00:22:36,000 --> 00:22:38,280
ago, you can't do this alone, 
right? 

404
00:22:38,280 --> 00:22:41,760
So I like, it's not really a 
support group, right, But it's 

405
00:22:41,760 --> 00:22:46,120
like an engine for the, for how 
you build upon things, right. 

406
00:22:46,240 --> 00:22:48,720
And everything. 
If you look at art and in the 

407
00:22:48,720 --> 00:22:50,960
history of time, it's always 
like you build upon each other. 

408
00:22:50,960 --> 00:22:53,040
You take inspiration from 
different things and you build 

409
00:22:53,040 --> 00:22:56,160
upon that as well as code. 
Are you kind of copy a lot of 

410
00:22:56,160 --> 00:22:57,920
other people's functions and 
vice versa. 

411
00:22:57,920 --> 00:23:05,000
So the, the when I talk about 
kind of building together, it's 

412
00:23:05,000 --> 00:23:07,600
sharing patterns, right, that 
are meant to be iterated on. 

413
00:23:07,800 --> 00:23:11,720
One person figures out how to 
catch an SAML abuse, another 

414
00:23:11,720 --> 00:23:13,960
person should tune it better 
because they, they have another 

415
00:23:13,960 --> 00:23:16,920
perspective on it. 
The feedback loops is really 

416
00:23:16,920 --> 00:23:20,880
where everyone learns. 
And if we fail collectively, we 

417
00:23:20,880 --> 00:23:24,920
all learn as well collectively. 
So I think that that if you have

418
00:23:24,920 --> 00:23:26,840
the opportunity to get involved,
do it. 

419
00:23:27,360 --> 00:23:31,080
It's luckily that this this 
particular industry needs so 

420
00:23:31,080 --> 00:23:35,440
much more of it because it's 
honest. 

421
00:23:35,440 --> 00:23:39,680
It's the best way to level up 
analyst now that Chachi BT is in

422
00:23:39,680 --> 00:23:43,160
the mix, right? 
And you can you can do that by 

423
00:23:43,160 --> 00:23:46,880
contributing just writing what 
projects you're working on on 

424
00:23:46,880 --> 00:23:51,200
medium, on sub stack. 
There's, there's already like 

425
00:23:51,800 --> 00:23:56,000
established programs like Thor 
Collective that there are 

426
00:23:56,160 --> 00:23:59,000
absolutely willing to take your 
research and showcase it. 

427
00:23:59,320 --> 00:24:02,520
I'm willing to to share to 
showcase if you've got a cool 

428
00:24:02,520 --> 00:24:05,720
use case that you've built in 
the sock on our podcast. 

429
00:24:06,240 --> 00:24:08,480
And yeah, just pretty much 
getting loud and asking 

430
00:24:08,480 --> 00:24:11,920
questions and being curious 
about it is, is a way to do it. 

431
00:24:11,920 --> 00:24:14,160
No one is going to turn down 
like that Curiosity. 

432
00:24:14,680 --> 00:24:17,400
So if people have ideas they 
should, they should, they should

433
00:24:17,400 --> 00:24:19,040
reach out to you. 
How? 

434
00:24:19,040 --> 00:24:21,800
How can they find the Thor 
Collective besides Googling it I

435
00:24:21,840 --> 00:24:23,640
guess? 
I think, well, they can reach 

436
00:24:23,680 --> 00:24:25,160
out to me. 
They can reach out to Sydney 

437
00:24:25,160 --> 00:24:27,240
Maroney. 
Shout out to her. 

438
00:24:27,800 --> 00:24:30,040
She's the original founder of 
the Thor Collective. 

439
00:24:30,560 --> 00:24:33,680
She will, she will, you know, 
take take your Google doc. 

440
00:24:33,680 --> 00:24:37,200
She'll get it ready prep for you
and and she'll publish it. 

441
00:24:37,720 --> 00:24:41,320
Anything that's that that gets 
something out to the community 

442
00:24:41,320 --> 00:24:43,880
that for an an open to feedback.
I know some people may be a 

443
00:24:43,880 --> 00:24:46,720
little bit scared of the 
feedback that they may get, but 

444
00:24:46,720 --> 00:24:49,840
I haven't had any, any, any bad 
experience with it so far. 

445
00:24:50,080 --> 00:24:52,280
But yeah, they could totally 
submit their research there. 

446
00:24:53,160 --> 00:24:53,920
That sounds great. 
Cool. 

447
00:24:55,160 --> 00:24:57,680
Amazing, thank you for joining 
us. 

448
00:24:57,680 --> 00:25:03,600
It was so fun having you on our 
podcast and I I learned a lot of

449
00:25:03,600 --> 00:25:07,200
time of our listeners will enjoy
and also check out your podcast 

450
00:25:07,200 --> 00:25:11,000
will link it so they can check 
it out as well and thank you. 

451
00:25:11,440 --> 00:25:15,160
Thank you so much for having me.
And to all of the Wiz fans and 

452
00:25:15,280 --> 00:25:18,520
Cloud SEC nerds tuning in, Wiz 
is a real deal. 

453
00:25:18,760 --> 00:25:20,760
They'll never be another Wiz, we
all know it. 

454
00:25:21,800 --> 00:25:25,480
If you enjoyed the show, be sure
to subscribe and share a link to

455
00:25:25,480 --> 00:25:28,160
the podcast, but not your cloud 
keys. 

456
00:25:28,720 --> 00:25:31,960
And as always, if your cloud 
security strategy is making you 

457
00:25:31,960 --> 00:25:33,840
cry, don't worry. 
Just cry out.

