สำเนาบทสนทนา 🔗
WEBVTT
00:00:00.000 --> 00:00:01.586
We always joke a zero-day
00:00:01.586 --> 00:00:03.235
was about 25 grand before,
00:00:03.235 --> 00:00:05.092
you know, Mythos and all this stuff,
00:00:05.092 --> 00:00:07.000
and now it's 11 cents in token costs.
00:00:07.000 --> 00:00:08.846
You know, we talk about in the software
00:00:08.846 --> 00:00:10.833
development life cycle, software engineers
00:00:10.833 --> 00:00:12.538
becoming force multipliers,
00:00:12.538 --> 00:00:15.000
and the same applies to the bad actors.
00:00:15.000 --> 00:00:16.367
What we're increasingly finding
00:00:16.367 --> 00:00:17.778
is the pace is almost impossible
00:00:17.778 --> 00:00:20.000
for any human team to keep up with now.
00:00:20.000 --> 00:00:21.667
I think we let the agents be responsible
00:00:21.667 --> 00:00:23.000
for reviewing the pull requests.
00:00:23.000 --> 00:00:24.934
If I'm sitting working on reviewing
00:00:24.934 --> 00:00:26.591
some code agents have written,
00:00:26.591 --> 00:00:29.922
I just have to get up and do a lap at some
00:00:29.922 --> 00:00:32.778
point because there's so much there,
00:00:32.778 --> 00:00:35.044
it can be overwhelming.
00:00:45.000 --> 00:00:47.893
Yo, yo, yo. Welcome back to another
00:00:47.893 --> 00:00:50.455
episode of Ready, Set, Runtime,
00:00:50.455 --> 00:00:53.104
a podcast from AHEAD. As a reminder,
00:00:53.104 --> 00:00:55.385
we're talking about accelerated
00:00:55.385 --> 00:00:57.728
development and how it's changing
00:00:57.728 --> 00:01:00.000
how we deliver in the IT estate.
00:01:00.000 --> 00:01:02.297
Today I'm joined by two colleagues
00:01:02.297 --> 00:01:05.000
who work on building AI and securing it.
00:01:05.000 --> 00:01:07.250
Tim Curless is a Principal Architect
00:01:07.250 --> 00:01:09.000
for AI Engineering at AHEAD.
00:01:09.000 --> 00:01:10.286
He leads enterprise AI
00:01:10.286 --> 00:01:11.923
platform engineering, brings
00:01:11.923 --> 00:01:14.181
AI into the software development life
00:01:14.181 --> 00:01:16.500
cycle to improve security, governance,
00:01:16.500 --> 00:01:18.611
and effectiveness, and he builds
00:01:18.611 --> 00:01:21.250
agentic systems like DevSecOps Sentinel.
00:01:21.250 --> 00:01:23.352
He's also building a quantum
00:01:23.352 --> 00:01:25.455
computing lab on NVIDIA DGX.
00:01:25.455 --> 00:01:28.182
Also joining us is Felix Vargas,
00:01:28.182 --> 00:01:29.471
Senior Director of Security
00:01:29.471 --> 00:01:31.000
Specialist Solution Engineering.
00:01:31.000 --> 00:01:34.250
With 23 years in technology, he leads
00:01:34.250 --> 00:01:37.500
our security pre-sales team, advising
00:01:37.500 --> 00:01:41.295
enterprises on securing their AI platforms
00:01:41.295 --> 00:01:45.000
and using AI to strengthen cyber defense.
00:01:45.000 --> 00:01:45.662
Tim, Felix,
00:01:45.662 --> 00:01:47.647
thank you so much for joining us.
00:01:47.647 --> 00:01:51.125
You are both right now on the front
00:01:51.125 --> 00:01:55.000
lines of the most topical news stories.
00:01:55.000 --> 00:01:57.786
Everyone is talking about the security,
00:01:57.786 --> 00:01:59.500
let's say, opportunities
00:01:59.500 --> 00:02:02.552
and challenges associated with how AI
00:02:02.552 --> 00:02:05.357
is proliferating in the IT estate.
00:02:05.357 --> 00:02:08.213
And just to start, for both of you,
00:02:08.213 --> 00:02:10.417
but let's start with Felix,
00:02:10.417 --> 00:02:12.672
how have your security concerns
00:02:12.672 --> 00:02:15.000
changed over the last few years,
00:02:15.000 --> 00:02:18.000
and how has AI changed your role?
00:02:18.000 --> 00:02:20.116
Yeah, no. First of all,
00:02:20.116 --> 00:02:22.692
thank you for having me, MK.
00:02:22.692 --> 00:02:24.309
I've been a longtime
00:02:24.309 --> 00:02:26.250
listener of the podcast,
00:02:26.250 --> 00:02:27.466
first-time participant,
00:02:27.466 --> 00:02:29.000
so I'm very happy to be here.
00:02:29.000 --> 00:02:31.143
So yes, lots of changes. I mean,
00:02:31.143 --> 00:02:33.889
I would say most cybersecurity folks will
00:02:33.889 --> 00:02:36.549
tell you that we've had a very long year,
00:02:36.549 --> 00:02:38.235
a very long several years.
00:02:38.235 --> 00:02:40.451
We just had our security executive
00:02:40.451 --> 00:02:42.667
advisory board in Florida a couple
00:02:42.667 --> 00:02:45.424
weeks ago, and that sentiment was shared
00:02:45.424 --> 00:02:48.250
across the CISO community that joined us.
00:02:48.250 --> 00:02:49.356
But I would say,
00:02:49.356 --> 00:02:51.154
you know, a couple things.
00:02:51.154 --> 00:02:53.134
I would say before AI,
00:02:53.134 --> 00:02:55.385
you needed to be somewhat
00:02:55.385 --> 00:02:57.992
specialized in order to conduct some
00:02:57.992 --> 00:03:00.526
of these attacks that we're seeing.
00:03:00.526 --> 00:03:02.555
There was an element of sophistication
00:03:02.555 --> 00:03:04.211
that had to come with the role.
00:03:04.211 --> 00:03:06.144
There were some organizations that looked
00:03:06.144 --> 00:03:08.125
to abstract that complexity and say, "Hey,
00:03:08.125 --> 00:03:10.248
we have a ransomware-as-a-service
00:03:10.248 --> 00:03:12.500
solution for unsophisticated threat
00:03:12.500 --> 00:03:14.431
actors to leverage this platform
00:03:14.431 --> 00:03:16.000
to conduct these attacks."
00:03:16.000 --> 00:03:17.333
That was before AI.
00:03:17.333 --> 00:03:19.926
And now we're in a situation where you can
00:03:19.926 --> 00:03:22.333
have an unsophisticated threat actor...
00:03:22.333 --> 00:03:24.325
And we always joke a zero-day
00:03:24.325 --> 00:03:26.111
was about 25 grand before,
00:03:26.111 --> 00:03:28.910
you know, Mythos and all this stuff,
00:03:28.910 --> 00:03:31.786
and now it's 11 cents in token costs.
00:03:31.786 --> 00:03:33.929
So you could have an unsophisticated
00:03:33.929 --> 00:03:34.915
threat actor conduct
00:03:34.915 --> 00:03:36.000
sophisticated attacks.
00:03:36.000 --> 00:03:39.231
And what
Load full transcript
that has done, it has increased
00:03:39.231 --> 00:03:42.059
the number of threat actors, right?
00:03:42.059 --> 00:03:44.589
Because the barrier of entry to being
00:03:44.589 --> 00:03:47.188
a threat actor now is very low, right?
00:03:47.188 --> 00:03:48.750
So we have an increase.
00:03:48.750 --> 00:03:51.100
And then also that's affecting our mean
00:03:51.100 --> 00:03:53.571
time to detection, mean time to response.
00:03:53.571 --> 00:03:56.148
So CrowdStrike has it at 29...
00:03:56.148 --> 00:03:59.583
29 minutes is the average breakout time,
00:03:59.583 --> 00:04:01.525
and the shortest they've seen
00:04:01.525 --> 00:04:03.333
is about 27 seconds, right?
00:04:03.333 --> 00:04:05.966
Breakout time. And we're still struggling
00:04:05.966 --> 00:04:08.214
as a cybersecurity community to get
00:04:08.214 --> 00:04:11.234
to a mean time to detection and a mean
00:04:11.234 --> 00:04:14.333
time to response in the minutes, right?
00:04:14.333 --> 00:04:16.667
So most organizations, according to IBM,
00:04:16.667 --> 00:04:19.829
are still in the 200-day range as far
00:04:19.829 --> 00:04:23.077
as detecting and containing an attack.
00:04:23.077 --> 00:04:25.802
So we are being very largely
00:04:25.802 --> 00:04:28.333
outpaced by the, you know,
00:04:28.333 --> 00:04:29.448
cybersecurity ecosystem
00:04:29.448 --> 00:04:31.000
from a threat actor perspective.
00:04:31.000 --> 00:04:33.127
Now we talk about in the software
00:04:33.127 --> 00:04:35.833
development life cycle, you know, software
00:04:35.833 --> 00:04:37.862
engineers becoming force multipliers,
00:04:37.862 --> 00:04:40.000
and the same applies to the bad actors.
00:04:40.000 --> 00:04:42.000
They are also force multipliers, right?
00:04:42.000 --> 00:04:44.199
Tim, what about you in the last
00:04:44.199 --> 00:04:45.333
couple of years?
00:04:45.333 --> 00:04:47.632
You know, how has security and AI
00:04:47.632 --> 00:04:50.000
changed how you come to your role?
00:04:50.000 --> 00:04:51.875
Yeah, well, again, as well,
00:04:51.875 --> 00:04:53.333
thanks for having me.
00:04:53.333 --> 00:04:56.022
I'm very glad to be here and very excited.
00:04:56.022 --> 00:04:57.750
So the theme I think you'll
00:04:57.750 --> 00:04:59.931
probably hear from me a few times today
00:04:59.931 --> 00:05:02.000
is that AI's accelerating everything.
00:05:02.000 --> 00:05:05.032
It's pouring gas on what we've done as an
00:05:05.032 --> 00:05:07.917
enterprise. So the effort that has been
00:05:07.917 --> 00:05:10.202
increasingly required and the onus
00:05:10.202 --> 00:05:12.353
that has been put on developers,
00:05:12.353 --> 00:05:14.421
obviously as well as security engineers,
00:05:14.421 --> 00:05:16.333
to partake in security has amplified.
00:05:16.333 --> 00:05:18.728
And I think for most organizations we work
00:05:18.728 --> 00:05:20.952
with, the default is we're gonna handle
00:05:20.952 --> 00:05:22.552
this like we've handled things
00:05:22.552 --> 00:05:24.524
in the past, which is we have a scan,
00:05:24.524 --> 00:05:26.081
it tells us some things are vulnerable,
00:05:26.081 --> 00:05:27.000
and we go and fix them.
00:05:27.000 --> 00:05:28.762
But what we're increasingly
00:05:28.762 --> 00:05:30.263
finding is the pace is,
00:05:30.263 --> 00:05:32.077
you know, almost impossible for any
00:05:32.077 --> 00:05:33.684
human team to keep up with now.
00:05:33.684 --> 00:05:36.343
So it's putting this pressure
00:05:36.343 --> 00:05:39.643
on from both sides in terms of, yes,
00:05:39.643 --> 00:05:42.077
the attacks are becoming force-multiplied
00:05:42.077 --> 00:05:44.333
by AI. We need to change our processes
00:05:44.333 --> 00:05:45.871
and our approach to combat
00:05:45.871 --> 00:05:48.000
that in a like cause or like manner.
00:05:48.000 --> 00:05:50.805
So I think we'll dig into that more,
00:05:50.805 --> 00:05:54.000
but that's my top-level thesis, I'll say.
00:05:54.000 --> 00:05:55.801
Just to add, you know, to that,
00:05:55.801 --> 00:05:58.182
and I'm completely aligned with Tim here,
00:05:58.182 --> 00:05:59.234
you know, from a force
00:05:59.234 --> 00:06:00.333
multiplier perspective.
00:06:00.333 --> 00:06:01.855
The theme, right, and,
00:06:01.855 --> 00:06:04.000
you know, the fundamental theme
00:06:04.000 --> 00:06:06.688
that I think we wanna get across here
00:06:06.688 --> 00:06:09.375
is we do have to do the fundamentals,
00:06:09.375 --> 00:06:11.000
but we have to move at machine speed.
00:06:11.000 --> 00:06:13.072
If you think about the threat actors,
00:06:13.072 --> 00:06:15.312
they're moving at machine speed already,
00:06:15.312 --> 00:06:17.188
right? We talked about that, right?
00:06:17.188 --> 00:06:18.444
There's more of them.
00:06:18.444 --> 00:06:20.000
They have access to tools.
00:06:20.000 --> 00:06:21.000
They, to some degree,
00:06:21.000 --> 00:06:22.143
got a head start, right?
00:06:22.143 --> 00:06:23.140
And so now we need
00:06:23.140 --> 00:06:24.524
to move at machine speed,
00:06:24.524 --> 00:06:26.524
and what that means for every organization
00:06:26.524 --> 00:06:28.000
that is out there is different.
00:06:28.000 --> 00:06:33.636
And so machine speed could mean, you know,
00:06:33.636 --> 00:06:34.807
validating exploitability
00:06:34.807 --> 00:06:36.071
of exposures faster, right?
00:06:36.071 --> 00:06:38.214
Or adding agents into the SOC, right?
00:06:38.214 --> 00:06:40.231
And so it's all a matter
00:06:40.231 --> 00:06:42.500
of determining what machine
00:06:42.500 --> 00:06:44.606
speed means for my organization
00:06:44.606 --> 00:06:47.188
and how I can actually action on that.
00:06:47.188 --> 00:06:48.884
And, you know, do I have the resources
00:06:48.884 --> 00:06:50.000
to move at machine speed?
00:06:50.000 --> 00:06:53.303
And there's a whole sort of conversation
00:06:53.303 --> 00:06:56.111
there around build versus buy and,
00:06:56.111 --> 00:06:57.602
you know, we really get
00:06:57.602 -->
00:07:00.000
into the specifics of how to do that,
00:07:00.000 --> 00:07:01.000
but that's really what...
00:07:01.000 --> 00:07:02.210
what we're facing right now
00:07:02.210 --> 00:07:03.600
is how do we do more with less,
00:07:03.600 --> 00:07:06.000
and how do we use AI to fight AI?
00:07:06.000 --> 00:07:07.576
One of the things
00:07:07.576 --> 00:07:10.357
that I think about a lot is...
00:07:10.357 --> 00:07:12.000
and Josh Perkins talked
00:07:12.000 --> 00:07:14.286
about this when we met with him,
00:07:14.286 --> 00:07:18.120
is that companies who had really pursued
00:07:18.120 --> 00:07:21.667
good hygiene around security, DevOps,
00:07:21.667 --> 00:07:25.037
SecOps, are in a better position,
00:07:25.037 --> 00:07:27.692
and those that are behind,
00:07:27.692 --> 00:07:30.000
they're in a substantially worse position.
00:07:30.000 --> 00:07:32.687
Like the gap between readiness
00:07:32.687 --> 00:07:36.000
and not ready, my concern is growing.
00:07:36.000 --> 00:07:38.145
And in some ways, when I think about,
00:07:38.145 --> 00:07:40.000
let's say, financial services...
00:07:40.000 --> 00:07:41.500
In financial services,
00:07:41.500 --> 00:07:45.205
they've been so focused on protection for
00:07:45.205 --> 00:07:49.000
so long that I actually have some optimism
00:07:49.000 --> 00:07:51.600
about the ability in financial services
00:07:51.600 --> 00:07:54.000
to continue to protect their assets.
00:07:54.000 --> 00:07:56.647
But similarly for the areas
00:07:56.647 --> 00:07:59.000
that are less regulated,
00:07:59.000 --> 00:08:01.221
maybe less focused and concerned,
00:08:01.221 --> 00:08:03.846
I'm wildly concerned about their assets
00:08:03.846 --> 00:08:07.243
and their, you know, being attacked
00:08:07.243 --> 00:08:10.833
without the proper controls in place.
00:08:10.833 --> 00:08:13.063
And so when you think about who
00:08:13.063 --> 00:08:15.364
is responsible for security now,
00:08:15.364 --> 00:08:17.000
you know, I think we used to silo it.
00:08:17.000 --> 00:08:18.167
We'd be like, "Well,
00:08:18.167 --> 00:08:20.385
the CISO is responsible for security,"
00:08:20.385 --> 00:08:22.471
and I've always taken the opinion
00:08:22.471 --> 00:08:25.000
that we're all responsible for security.
00:08:25.000 --> 00:08:27.607
But Felix, like now when you think about,
00:08:27.607 --> 00:08:30.278
you know, who is responsible for security,
00:08:30.278 --> 00:08:32.669
how do you think you answer that, and is
00:08:32.669 --> 00:08:35.000
it a different answer than in the past?
00:08:35.000 --> 00:08:38.000
I think it's the same answer as the past.
00:08:38.000 --> 00:08:40.554
I don't think... 'cause I would categorize
00:08:40.554 --> 00:08:42.500
that under a fundamental, right?
00:08:42.500 --> 00:08:43.942
It's always been people,
00:08:43.942 --> 00:08:45.385
process, and technology,
00:08:45.385 --> 00:08:48.462
and if you think about people, you might
00:08:48.462 --> 00:08:51.086
see the CISO as ultimately managing
00:08:51.086 --> 00:08:53.636
the risk within that organization.
00:08:53.636 --> 00:08:56.867
But every other stakeholder below
00:08:56.867 --> 00:09:00.000
the CISO in adjacent disciplines
00:09:00.000 --> 00:09:02.571
are just as responsible in cybersecurity
00:09:02.571 --> 00:09:04.500
and have been for a long time.
00:09:04.500 --> 00:09:07.374
So I don't think that has changed,
00:09:07.374 --> 00:09:10.333
but what has changed is how quickly
00:09:10.333 --> 00:09:11.970
we need to come together,
00:09:11.970 --> 00:09:14.000
right, and sort of collaborate.
00:09:14.000 --> 00:09:15.847
Because before, maybe we could get away
00:09:15.847 --> 00:09:17.647
with poor collaboration between teams,
00:09:17.647 --> 00:09:19.288
and we're living in a world
00:09:19.288 --> 00:09:20.625
where we can't, right?
00:09:20.625 --> 00:09:22.959
If you think about Mythos and exposure
00:09:22.959 --> 00:09:25.231
management, I can have a tool to scan
00:09:25.231 --> 00:09:26.437
and discover an asset, but,
00:09:26.437 --> 00:09:28.000
you know, an exposure and an asset.
00:09:28.000 --> 00:09:31.457
But, you know, remediating that exposure
00:09:31.457 --> 00:09:35.000
involves probably three or four different
00:09:35.000 --> 00:09:37.600
distinct teams within an organization.
00:09:37.600 --> 00:09:39.583
So we have to really increase
00:09:39.583 --> 00:09:40.404
our collaboration,
00:09:40.404 --> 00:09:42.000
and that's really what has changed.
00:09:42.000 --> 00:09:43.925
That's an amazing point. I mean,
00:09:43.925 --> 00:09:45.429
I think it's so right on.
00:09:45.429 --> 00:09:48.000
You know, the risk is growing.
00:09:48.000 --> 00:09:50.474
You've described sort of the time
00:09:50.474 --> 00:09:52.273
to address is shrinking,
00:09:52.273 --> 00:09:55.065
and the need to be together to protect
00:09:55.065 --> 00:09:57.857
ourselves is more important than ever.
00:09:57.857 --> 00:10:00.184
Tim, when you think about the same thing,
00:10:00.184 --> 00:10:02.000
you know, how do you come at it?
00:10:02.000 --> 00:10:04.098
Yeah. I was gonna say, I think my heart
00:10:04.098 --> 00:10:06.143
wants to agree that it's everyone's...
00:10:06.143 --> 00:10:08.082
security is everyone's responsibility,
00:10:08.082 --> 00:10:09.000
and I think it is.
00:10:09.000 --> 00:10:10.897
But we all know in practice the rubber
00:10:10.897 --> 00:10:12.895
meets the road a little bit differently.
00:10:12.895 --> 00:10:15.257
And, you know, I think for an organization
00:10:15.257 --> 00:10:17.000
that has weak practices, again,
00:10:17.000 --> 00:10:18.483
it's going to accelerate that.
00:10:18.483 --> 00:10:20.312
It's gonna exacerbate that challenge.
00:10:20.312 --> 00:10:23.493
So if I have a business unit or a group
00:10:23.493 --> 00:10:26.429
that stands up my code base, GitHub,
00:10:26.429 --> 00:10:28.766
and gives everyone admin access,
00:10:28.766 --> 00:10:29.643
for example,
00:10:29.643 --> 00:10:30.643
you're gonna have the Replit incident.
00:10:30.643 --> 00:10:32.570
You're gonna have agents that come
00:10:32.570 --> 00:10:34.667
in and are onboarded like humans are,
00:10:34.667 --> 00:10:36.070
and that means they also
00:10:36.070 --> 00:10:37.941
have admin access to everything.
00:10:37.941 --> 00:10:41.060
So I'm a firm believer that we need
00:10:41.060 --> 00:10:44.000
to shift the security piece down.
00:10:44.000 --> 00:10:45.173
We've been talking shift
00:10:45.173 --> 00:10:46.200
left for a long time,
00:10:46.200 --> 00:10:47.100
and that's putting security
00:10:47.100 --> 00:10:48.000
on developers to an extent.
00:10:48.000 --> 00:10:49.611
I think we need to shift down
00:10:49.611 --> 00:10:51.667
so that it's covered in the platform,
00:10:51.667 --> 00:10:53.775
and that is then applied the same
00:10:53.775 --> 00:10:55.500
way to agents or to humans.
00:10:55.500 --> 00:10:57.989
And I think that's one of the key pieces.
00:10:57.989 --> 00:11:00.357
The other thing I think about is the...
00:11:00.357 --> 00:11:02.531
I agree the regulated industries
00:11:02.531 --> 00:11:05.385
have an advantage, but the challenge still
00:11:05.385 --> 00:11:08.192
is there's not a clear picture of what
00:11:08.192 --> 00:11:11.000
the kind of AI regulation needs to be.
00:11:11.000 --> 00:11:12.481
It's sort of just taken on face
00:11:12.481 --> 00:11:14.200
value that it's your responsibility,
00:11:14.200 --> 00:11:15.590
but what does that mean
00:11:15.590 --> 00:11:17.222
I have to protect, and how?
00:11:17.222 --> 00:11:18.889
So I think organizations that already
00:11:18.889 --> 00:11:20.600
kind of have that protection framework
00:11:20.600 --> 00:11:23.000
in place are absolutely at an advantage.
00:11:23.000 --> 00:11:24.355
One of the things
00:11:24.355 --> 00:11:26.429
that I always think about,
00:11:26.429 --> 00:11:28.748
and it's been the core of security
00:11:28.748 --> 00:11:31.000
for my whole career, is identity.
00:11:31.000 --> 00:11:34.245
And we're in such a bizarre spot
00:11:34.245 --> 00:11:37.692
as it relates to identity because,
00:11:37.692 --> 00:11:39.338
you know, agents not...
00:11:39.338 --> 00:11:42.273
let's say non-human identities or actors,
00:11:42.273 --> 00:11:46.666
they need to have an accountable owner.
00:11:46.666 --> 00:11:50.833
And that's people, process, you know,
00:11:50.833 --> 00:11:52.106
and hopefully a little
00:11:52.106 --> 00:11:53.611
bit of tech to support it.
00:11:53.611 --> 00:11:55.242
But when you think about identity
00:11:55.242 --> 00:11:56.429
and identity challenges,
00:11:56.429 --> 00:11:58.625
how do you see identity evolving, you
00:11:58.625 --> 00:12:01.000
know, over the next year or three years?
00:12:01.000 --> 00:12:03.000
Felix, we can start with you.
00:12:03.000 --> 00:12:04.129
Yeah. You know,
00:12:04.129 --> 00:12:06.765
it's interesting you bring this up.
00:12:06.765 --> 00:12:08.394
In our security executive advisory
00:12:08.394 --> 00:12:10.263
board last week, we posed this question
00:12:10.263 --> 00:12:12.944
to the group, and there was one CISO
00:12:12.944 --> 00:12:16.071
that had a quote that I thought was great.
00:12:16.071 --> 00:12:17.649
The conversation was around
00:12:17.649 --> 00:12:19.286
moving at machine speed and,
00:12:19.286 --> 00:12:20.797
you know, pointing machine
00:12:20.797 --> 00:12:22.308
speed at the fundamentals,
00:12:22.308 --> 00:12:24.319
and identity came up,
00:12:24.319 --> 00:12:27.000
and she said, "I can't worry
00:12:27.000 --> 00:12:29.606
about non-human identities if I still
00:12:29.606 --> 00:12:32.000
have not sorted human identities."
00:12:32.000 --> 00:12:35.455
And so I couldn't agree more.
00:12:35.455 --> 00:12:37.830
Identity is a conversation
00:12:37.830 --> 00:12:40.938
we are having more and more often.
00:12:40.938 --> 00:12:43.576
It's always been a topic of conversation,
00:12:43.576 --> 00:12:45.250
but more so now than ever.
00:12:45.250 --> 00:12:47.750
And I remember when I was on stage,
00:12:47.750 --> 00:12:49.750
you know, a year ago, right,
00:12:49.750 --> 00:12:51.806
in February, and I said, you know,
00:12:51.806 --> 00:12:54.286
"Non-human identities to human identities
00:12:54.286 --> 00:12:56.179
are gonna drastically grow
00:12:56.179 --> 00:12:58.000
over the next few years."
00:12:58.000 --> 00:13:01.342
And so the number was 82 to one by the end
00:13:01.342 --> 00:13:03.889
of the year, non-human to human.
00:13:03.889 --> 00:13:06.250
And what I didn't say at the time,
00:13:06.250 --> 00:13:08.190
and the context of the comment
00:13:08.190 --> 00:13:10.000
was around these are a force
00:13:10.000 --> 00:13:11.995
multiplier and they could, you know,
00:13:11.995 --> 00:13:13.824
do things for good or bad, right?
00:13:13.824 --> 00:13:15.184
Depending on its amplification
00:13:15.184 --> 00:13:16.000
of human behavior.
00:13:16.000 --> 00:13:19.500
And the context of it was that a human
00:13:19.500 --> 00:13:21.250
controlling agents,
00:13:21.250 --> 00:13:24.718
their identity could be hijacked, and then
00:13:24.718 --> 00:13:27.609
their agents could be used to do X.
00:13:27.609 --> 00:13:29.304
And then what we're seeing now is that
00:13:29.304 --> 00:13:31.000
the agents themselves are doing things
00:13:31.000 --> 00:13:33.616
on their own, right? So what we're gonna
00:13:33.616 --> 00:13:36.364
see over the next few years is an emphasis
00:13:36.364 --> 00:13:38.444
on going back to the fundamentals
00:13:38.444 --> 00:13:40.588
around discovery and understanding
00:13:40.588 --> 00:13:43.289
what are all the agents and non-human
00:13:43.289 --> 00:13:45.625
identities and human identities,
00:13:45.625 --> 00:13:47.985
rationalizing them, analyzing them,
00:13:47.985 --> 00:13:50.750
and understanding what is their behavior,
00:13:50.750 --> 00:13:52.302
what should they have access to,
00:13:52.302 --> 00:13:54.000
what do they have access to, right?
00:13:54.000 --> 00:13:56.262
And then get to a point where we're
00:13:56.262 --> 00:13:58.846
then protecting those identities, right?
00:13:58.846 --> 00:14:00.882
And shifting from thinking about human
00:14:00.882 --> 00:14:02.702
and non-human and just going back
00:14:02.702 --> 00:14:04.412
to the fundamentals and saying,
00:14:04.412 --> 00:14:05.930
"Do we have the proper controls
00:14:05.930 --> 00:14:07.105
and protections in place
00:14:07.105 --> 00:14:09.101
to prevent human and non-human identities
00:14:09.101 --> 00:14:11.000
from doing things that they shouldn't?"
00:14:11.000 --> 00:14:12.832
And one key piece there that is often
00:14:12.832 --> 00:14:14.714
missed is behavioral analytics, right?
00:14:14.714 --> 00:14:17.323
As we monitor these agents, right? You
00:14:17.323 --> 00:14:20.000
know, I could be in my house right now,
00:14:20.000 --> 00:14:21.722
and I could have guardrails and say,
00:14:21.722 --> 00:14:23.636
"Hey, I have AIs or humans in the house.
00:14:23.636 --> 00:14:25.643
They can't leave," or, you know,
00:14:25.643 --> 00:14:28.214
only a subset of individuals can come in.
00:14:28.214 --> 00:14:29.939
And that could be dictated
00:14:29.939 --> 00:14:31.000
by a front door.
00:14:31.000 --> 00:14:32.410
I also need to look
00:14:32.410 --> 00:14:34.636
at the behavior itself, right?
00:14:34.636 --> 00:14:37.234
'Cause someone could use another way
00:14:37.234 --> 00:14:39.615
of leaving, and I need to monitor
00:14:39.615 --> 00:14:40.975
the agent itself, not just worry
00:14:40.975 --> 00:14:42.292
about what they have access to.
00:14:42.292 --> 00:14:43.333
So I think that...
00:14:43.333 --> 00:14:46.312
what we're gonna see over the next couple
00:14:46.312 --> 00:14:49.000
of years is an emphasis on behavioral
00:14:49.000 --> 00:14:51.375
analytics of human and non-human identity
00:14:51.375 --> 00:14:53.750
so that we can detect anomalous behavior,
00:14:53.750 --> 00:14:55.414
because we're not gonna solve
00:14:55.414 --> 00:14:57.308
the problem by having guardrails,
00:14:57.308 --> 00:14:59.054
even though guardrails are critical
00:14:59.054 --> 00:15:01.000
and they're part of those fundamentals.
00:15:01.000 --> 00:15:03.035
It's a defense-in-depth approach,
00:15:03.035 --> 00:15:05.625
and the behavior is something that I don't
00:15:05.625 --> 00:15:06.275
believe we're doing a good
00:15:06.275 --> 00:15:07.000
job as an industry right now.
00:15:07.000 --> 00:15:10.676
In audits, identity has been
00:15:10.676 --> 00:15:14.615
the easiest thing for auditors
00:15:14.615 --> 00:15:16.662
to look for issues with. So,
00:15:16.662 --> 00:15:19.000
you know, I worked at a company,
00:15:19.000 --> 00:15:20.855
and I left the company,
00:15:20.855 --> 00:15:24.000
and they turned off my AD, essentially.
00:15:24.000 --> 00:15:27.510
But they left on my ability to access,
00:15:27.510 --> 00:15:31.389
let's say, a specific business app, right?
00:15:31.389 --> 00:15:33.957
So that's a miss in identity, and that's
00:15:33.957 --> 00:15:36.333
an easy find for an auditor in a way.
00:15:36.333 --> 00:15:39.033
Now we're gonna ask auditors, and this is
00:15:39.033 --> 00:15:41.667
really where it's gonna get interesting.
00:15:41.667 --> 00:15:44.247
Auditors are gonna be struggling
00:15:44.247 --> 00:15:46.667
to understand identity because
00:15:46.667 --> 00:15:49.552
of the proliferation of identity,
00:15:49.552 --> 00:15:52.000
and we'll see how that goes.
00:15:52.000 --> 00:15:54.481
I expect that that is gonna force
00:15:54.481 --> 00:15:57.188
some things over the next few years.
00:15:57.188 --> 00:15:59.743
Tim, what are you seeing as it relates
00:15:59.743 --> 00:16:02.500
to identity and how people are, you know,
00:16:02.500 --> 00:16:05.444
thinking about the agents in the world,
00:16:05.444 --> 00:16:07.646
and how they're protecting
00:16:07.646 --> 00:16:09.000
their IT estate?
00:16:09.000 --> 00:16:12.350
Yeah, and I think we need to, on that last
00:16:12.350 --> 00:16:15.462
point, identify or create a fundamental
00:16:15.462 --> 00:16:16.791
understanding of what an identity
00:16:16.791 --> 00:16:18.000
needs to look like for agents.
00:16:18.000 --> 00:16:20.838
I think we did that on the human side,
00:16:20.838 --> 00:16:23.750
but, you know, onboarding, offboarding,
00:16:23.750 --> 00:16:25.682
what does that look like?
00:16:25.682 --> 00:16:27.159
You know, the identity package
00:16:27.159 --> 00:16:28.636
and what we need to attest to,
00:16:28.636 --> 00:16:31.133
I think, is gonna be crucial to define
00:16:31.133 --> 00:16:33.235
that. And folks are starting to.
00:16:33.235 --> 00:16:35.714
But the other piece of this is,
00:16:35.714 --> 00:16:39.018
you know, everyone begins with developers
00:16:39.018 --> 00:16:42.000
using copilots and coding assistants,
00:16:42.000 --> 00:16:43.273
and for the most part, those
00:16:43.273 --> 00:16:45.000
are identified based on the developer.
00:16:45.000 --> 00:16:48.000
But it brings up kind of the next point,
00:16:48.000 --> 00:16:50.457
which is that the settings in an IDE
00:16:50.457 --> 00:16:52.368
are ultimately a suggestion.
00:16:52.368 --> 00:16:54.400
We can't necessarily rely on those,
00:16:54.400 --> 00:16:55.909
not that we shouldn't try.
00:16:55.909 --> 00:16:59.291
But I think having the system in place,
00:16:59.291 --> 00:17:02.500
AI gateway is a good example of that,
00:17:02.500 --> 00:17:04.210
where I have a point of control
00:17:04.210 --> 00:17:06.471
of what my governance needs to look like.
00:17:06.471 --> 00:17:08.652
Is really key. And I think the other thing
00:17:08.652 --> 00:17:10.625
that I'm starting to see and recommend
00:17:10.625 --> 00:17:12.969
more is, let's make sure we have,
00:17:12.969 --> 00:17:15.385
at every stage of the SDLC anyway,
00:17:15.385 --> 00:17:18.120
agents that are doing the work, there's
00:17:18.120 --> 00:17:20.714
an identity that's created associated
00:17:20.714 --> 00:17:22.886
with them, token with a blast radius.
00:17:22.886 --> 00:17:25.000
It is checked out, the work is done.
00:17:25.000 --> 00:17:28.846
It's then, you know, the session is ended.
00:17:28.846 --> 00:17:31.535
That gets stored in even just a JSON
00:17:31.535 --> 00:17:34.000
in an S3 bucket that's immutable.
00:17:34.000 --> 00:17:36.304
Something that is immutable
00:17:36.304 --> 00:17:38.438
and unable to be changed.
00:17:38.438 --> 00:17:41.034
And then when the auditing teams come in,
00:17:41.034 --> 00:17:43.250
it's easy to point to that and say,
00:17:43.250 --> 00:17:45.000
"Here's the evidence that we have,
00:17:45.000 --> 00:17:46.043
and it's standardized
00:17:46.043 --> 00:17:47.333
across all of our agents."
00:17:47.333 --> 00:17:49.269
You know, there's shadow AI, and those
00:17:49.269 --> 00:17:51.000
present difficulties of their own,
00:17:51.000 --> 00:17:53.222
but I think control what we can control
00:17:53.222 --> 00:17:55.500
first and use that as the lesson learned
00:17:55.500 --> 00:17:57.000
to then move out from there.
00:17:57.000 --> 00:17:59.218
We've both been talking about control
00:17:59.218 --> 00:18:01.136
planes and governance throughout
00:18:01.136 --> 00:18:02.265
our conversation, and Felix,
00:18:02.265 --> 00:18:03.636
I like your metaphor of your house
00:18:03.636 --> 00:18:05.253
and how you think of kind
00:18:05.253 --> 00:18:07.000
of the control plane there.
00:18:07.000 --> 00:18:09.667
Do we need to create new control planes?
00:18:09.667 --> 00:18:11.360
Will we move them? Are there
00:18:11.360 --> 00:18:13.235
gonna be new concepts of gates?
00:18:13.235 --> 00:18:15.212
These are things that you've already
00:18:15.212 --> 00:18:17.353
alluded to, but Tim, why don't you talk
00:18:17.353 --> 00:18:19.202
us through kind of more specifically
00:18:19.202 --> 00:18:21.154
what you're starting to see leveraged.
00:18:21.154 --> 00:18:22.674
And it is clearly
00:18:22.674 --> 00:18:25.000
gonna be part of the SDLC,
00:18:25.000 --> 00:18:26.000
what we're about to talk about, right?
00:18:26.000 --> 00:18:27.394
You mentioned IDEs sort
00:18:27.394 --> 00:18:28.667
of not being perfect,
00:18:28.667 --> 00:18:31.085
but when you think about using
00:18:31.085 --> 00:18:33.182
harnesses or golden paths,
00:18:33.182 --> 00:18:35.269
you know, we can start to try
00:18:35.269 --> 00:18:37.500
to control these things through
00:18:37.500 --> 00:18:39.227
a combination of policy and hopefully
00:18:39.227 --> 00:18:41.000
real, you know, protective guardrails.
00:18:41.000 --> 00:18:43.222
Yeah, I mean, I think the inner loop,
00:18:43.222 --> 00:18:45.263
so the developer, where they spend
00:18:45.263 --> 00:18:47.266
their time doing their coding loop
00:18:47.266 --> 00:18:49.211
and so forth, some local testing,
00:18:49.211 --> 00:18:51.595
that is the place where organizations
00:18:51.595 --> 00:18:53.400
are finding speed right now.
00:18:53.400 --> 00:18:55.450
The outer loop, the handoff to, "I've
00:18:55.450 --> 00:18:57.500
done my work, it's in a pull request,
00:18:57.500 --> 00:19:00.038
and now I need to get it deployed," so all
00:19:00.038 --> 00:19:02.333
the steps it takes to get it deployed,
00:19:02.333 --> 00:19:03.208
including testing,
00:19:03.208 --> 00:19:04.667
that's where you prove safety.
00:19:04.667 --> 00:19:06.634
And not that you don't wanna go
00:19:06.634 --> 00:19:08.158
faster there, but you...
00:19:08.158 --> 00:19:09.538
that's where we wanna focus on how
00:19:09.538 --> 00:19:11.000
we implement some of these controls.
00:19:11.000 --> 00:19:13.471
I mean, the focus right now is,
00:19:13.471 --> 00:19:15.385
from a human standpoint,
00:19:15.385 --> 00:19:17.249
that the amount of reviews
00:19:17.249 --> 00:19:19.615
has really become the bottleneck.
00:19:19.615 --> 00:19:20.603
And so there's a whole
00:19:20.603 --> 00:19:21.591
challenge around that.
00:19:21.591 --> 00:19:23.831
But I think as we work to solve that,
00:19:23.831 --> 00:19:26.071
what we're seeing is a lot of humans,
00:19:26.071 --> 00:19:27.857
not maliciously by any means,
00:19:27.857 --> 00:19:30.209
just through the pressures of their day
00:19:30.209 --> 00:19:32.500
job, kind of being asked to do so many
00:19:32.500 --> 00:19:34.133
more reviews that things are just starting
00:19:34.133 --> 00:19:35.417
to get passed through unreviewed.
00:19:35.417 --> 00:19:37.852
And without the controls in place,
00:19:37.852 --> 00:19:40.000
we're seeing 125% increase per
00:19:40.000 --> 00:19:41.849
month in incidents caused
00:19:41.849 --> 00:19:43.846
by code agents are writing.
00:19:43.846 --> 00:19:46.392
So it's important to include the checks
00:19:46.392 --> 00:19:48.611
and balances and prove the safety,
00:19:48.611 --> 00:19:50.417
and become confident in the safety
00:19:50.417 --> 00:19:52.647
in the outer loop, where we're taking code
00:19:52.647 --> 00:19:54.515
from developers who've done their work
00:19:54.515 --> 00:19:56.333
through the QA process to deployment.
00:19:56.333 --> 00:19:58.726
That's not the whole picture, obviously,
00:19:58.726 --> 00:20:01.000
but that's the picture I'm focused on.
00:20:01.000 --> 00:20:02.944
When do you think we let the agents
00:20:02.944 --> 00:20:05.000
be responsible for reviewing the pull
00:20:05.000 --> 00:20:06.559
requests? When do we transfer...
00:20:06.559 --> 00:20:07.632
Because that is the...
00:20:07.632 --> 00:20:09.214
you know, that's the bottleneck now.
00:20:09.214 --> 00:20:10.357
That's the new bottleneck.
00:20:10.357 --> 00:20:12.712
We're delivering all this new code,
00:20:12.712 --> 00:20:15.000
but it needs to be human-reviewed.
00:20:15.000 --> 00:20:16.920
When do you think we get
00:20:16.920 --> 00:20:19.000
to the other side of that?
00:20:19.000 --> 00:20:21.360
Yeah, and one of the challenges is AI
00:20:21.360 --> 00:20:23.529
writing code so much more quickly,
00:20:23.529 -->
00:20:25.544
pull requests are becoming larger,
00:20:25.544 --> 00:20:27.500
code reviews are becoming larger,
00:20:27.500 --> 00:20:29.062
they're becoming more frequent.
00:20:29.062 --> 00:20:30.690
And there's a real... I felt this myself,
00:20:30.690 --> 00:20:32.000
there's a real degree of fatigue.
00:20:32.000 --> 00:20:34.176
If I'm sitting working on reviewing
00:20:34.176 --> 00:20:36.042
some code agents have written,
00:20:36.042 --> 00:20:38.472
I just have to get up and do a lap at some
00:20:38.472 --> 00:20:40.556
point because there's so much there,
00:20:40.556 --> 00:20:41.772
it can be overwhelming.
00:20:41.772 --> 00:20:43.889
And I think what I'm starting to believe
00:20:43.889 --> 00:20:45.854
is that no human team ultimately
00:20:45.854 --> 00:20:47.941
will be able to keep up with this.
00:20:47.941 --> 00:20:50.455
So I think about the iceberg analogy,
00:20:50.455 --> 00:20:52.045
and sort of the 80% of the mass
00:20:52.045 --> 00:20:53.636
of the iceberg under the water,
00:20:53.636 --> 00:20:54.603
that is the review
00:20:54.603 --> 00:20:56.000
that agents can help with.
00:20:56.000 --> 00:20:58.232
That's just basic hygiene,
00:20:58.232 --> 00:21:01.667
following standards, and it does go back
00:21:01.667 --> 00:21:03.418
to having well-defined standards
00:21:03.418 --> 00:21:05.278
in your repository AGENTS.md file,
00:21:05.278 --> 00:21:06.698
you know, that the team
00:21:06.698 --> 00:21:08.056
has sort of agreed on.
00:21:08.056 --> 00:21:10.475
But then pass the 20% of the mass above
00:21:10.475 --> 00:21:12.647
the water to the humans, and that's
00:21:12.647 --> 00:21:14.322
where you're looking for business
00:21:14.322 --> 00:21:16.250
logic failures and, you know, overt...
00:21:16.250 --> 00:21:18.603
does this match what the criteria was
00:21:18.603 --> 00:21:21.211
for getting this job or this ticket done?
00:21:21.211 --> 00:21:22.263
Those are, I think,
00:21:22.263 --> 00:21:23.736
the types of things that we're
00:21:23.736 --> 00:21:25.357
gonna start to see more and more.
00:21:25.357 --> 00:21:27.883
So I'm not necessarily of the opinion
00:21:27.883 --> 00:21:30.000
that we'll ever have AI agents,
00:21:30.000 --> 00:21:31.548
or should have AI agents,
00:21:31.548 --> 00:21:33.529
doing 100% of all the reviewing.
00:21:33.529 --> 00:21:35.448
But I do think a hybrid approach is much
00:21:35.448 --> 00:21:37.174
better than an "I'm overwhelmed, I'm
00:21:37.174 --> 00:21:39.618
just gonna pass it through 'cause I need
00:21:39.618 --> 00:21:42.000
to get these PRs out" type of approach.
00:21:42.000 --> 00:21:44.153
Felix, when you hear this, and I think
00:21:44.153 --> 00:21:46.250
you mentioned this earlier, you know,
00:21:46.250 --> 00:21:48.750
I expect companies are figuring out
00:21:48.750 --> 00:21:50.871
their own plan for protection or creating
00:21:50.871 --> 00:21:52.941
their own governance and control planes,
00:21:52.941 --> 00:21:55.461
and I expect the vendors in the space are
00:21:55.461 --> 00:21:57.857
also trying to productize those things.
00:21:57.857 --> 00:22:02.083
You know, what do you see in the market
00:22:02.083 --> 00:22:05.333
right now for addressing this?
00:22:05.333 --> 00:22:07.500
Are you seeing more companies sort of try
00:22:07.500 --> 00:22:09.667
to independently and autonomously address
00:22:09.667 --> 00:22:12.090
it, or are they going to partners
00:22:12.090 --> 00:22:14.000
to solve for this problem?
00:22:14.000 --> 00:22:16.944
Yeah, I mean, it's a combination, right?
00:22:16.944 --> 00:22:19.132
I always say, you know, people, process,
00:22:19.132 --> 00:22:21.429
and technology. So the first is, you know,
00:22:21.429 --> 00:22:24.416
making sure that we have standards
00:22:24.416 --> 00:22:27.667
that are well drafted and understood.
00:22:27.667 --> 00:22:30.383
And for me, you know, the pull request
00:22:30.383 --> 00:22:32.812
example, we see that all the time,
00:22:32.812 --> 00:22:34.869
where, you know, by whatever degree, pull
00:22:34.869 --> 00:22:36.875
requests increase 'cause developers are,
00:22:36.875 --> 00:22:39.691
you know, writing code now with a coding
00:22:39.691 --> 00:22:42.647
assistant, so you have, you know, a higher
00:22:42.647 --> 00:22:45.139
volume of code that is being written,
00:22:45.139 --> 00:22:47.632
which is affecting the pull requests,
00:22:47.632 --> 00:22:49.437
you know, the number of pull
00:22:49.437 --> 00:22:51.500
requests that we have to review.
00:22:51.500 --> 00:22:53.364
There are organizations
00:22:53.364 --> 00:22:55.714
that are targeting, you know,
00:22:55.714 --> 00:22:58.351
the shift-left mantra and saying, "Well,
00:22:58.351 --> 00:23:00.789
you know, we shouldn't," for example,
00:23:00.789 --> 00:23:02.043
picking on secrets in code,
00:23:02.043 --> 00:23:03.947
we shouldn't have secrets in code, right?
00:23:03.947 --> 00:23:06.237
And so we had an organization
00:23:06.237 --> 00:23:09.000
that we were working with that had,
00:23:09.000 --> 00:23:10.899
you know, roughly 20,000 pull requests,
00:23:10.899 --> 00:23:12.750
and half of them were secrets in code.
00:23:12.750 --> 00:23:15.097
And, you know, I kind of struggle
00:23:15.097 --> 00:23:16.875
with that because, again,
00:23:16.875 --> 00:23:18.791
we can talk all day about streamlining
00:23:18.791 --> 00:23:20.556
pull requests and prioritizing them
00:23:20.556 --> 00:23:22.806
and traceability and how we handle that,
00:23:22.806 --> 00:23:25.000
but there's a conversation that we also
00:23:25.000 --> 00:23:26.737
need to have around how did
00:23:26.737 --> 00:23:28.409
we get here to begin with.
00:23:28.409 --> 00:23:30.000
And there's a lot that we can do.
00:23:30.000 --> 00:23:31.953
I always tell this story.
00:23:31.953 --> 00:23:35.000
My brother and I were talking about AI,
00:23:35.000 --> 00:23:36.924
and he was a bit of a cynic on it.
00:23:36.924 --> 00:23:39.130
He said, "AI, you know, it's all
hype."
00:23:39.130 --> 00:23:40.732
And, you know, I said, "Well, we can
00:23:40.732 --> 00:23:42.600
write a World Cup application in the next,
00:23:42.600 --> 00:23:44.237
you know, two hours if we wanted to."
00:23:44.237 --> 00:23:45.476
And he said, "No, we can't."
00:23:45.476 --> 00:23:47.212
And two hours later, we had a World
00:23:47.212 --> 00:23:49.048
Cup application, and as part of that,
00:23:49.048 --> 00:23:51.207
I tried to integrate my World Cup app
00:23:51.207 --> 00:23:53.250
with Anthropic, and I had a secret,
00:23:53.250 --> 00:23:55.000
an API key that I wanted to write in code.
00:23:55.000 --> 00:23:56.727
And here I am trying to move
00:23:56.727 --> 00:23:58.824
quickly to prove my brother wrong,
00:23:58.824 --> 00:24:01.223
trying to add that into the code,
00:24:01.223 --> 00:24:02.895
and it didn't allow me.
00:24:02.895 --> 00:24:05.366
It said, "You need a vault." And so here's
00:24:05.366 --> 00:24:07.778
the vault, and it created a vault for me.
00:24:07.778 --> 00:24:10.061
And so the harness that I was using
00:24:10.061 --> 00:24:12.083
had that auto-blocking feature,
00:24:12.083 --> 00:24:14.815
that blocking gate to prevent, you know,
00:24:14.815 --> 00:24:17.000
me from putting secrets in code.
00:24:17.000 --> 00:24:19.378
And listen, I'm not a developer,
00:24:19.378 --> 00:24:22.500
but I think there's a lot of organizations
00:24:22.500 --> 00:24:25.471
that are focused on how can we prevent
00:24:25.471 --> 00:24:27.895
secrets from making it to code?
00:24:27.895 --> 00:24:29.646
And are there tools there
00:24:29.646 --> 00:24:31.818
that we can add to the harness?
00:24:31.818 --> 00:24:32.474
Is it an agent?
00:24:32.474 --> 00:24:34.091
Maybe your harness doesn't have that,
00:24:34.091 --> 00:24:34.847
but is there an agent
00:24:34.847 --> 00:24:36.000
that we could position in there?
00:24:36.000 --> 00:24:38.026
That's a really hot space right now.
00:24:38.026 --> 00:24:39.714
And even further left to that,
00:24:39.714 --> 00:24:41.872
think about... everything right now
00:24:41.872 --> 00:24:44.091
from an agentic security perspective
00:24:44.091 --> 00:24:46.042
is focused on the supply chain, right?
00:24:46.042 --> 00:24:47.978
If you think about a lot of the incidents
00:24:47.978 --> 00:24:49.583
that are happening, and, you know,
00:24:49.583 --> 00:24:51.414
we call it taint propagation,
00:24:51.414 --> 00:24:53.750
where one agent's context window gets
00:24:53.750 --> 00:24:55.406
tainted by the supply chain,
00:24:55.406 --> 00:24:56.944
either a skill, et cetera,
00:24:56.944 --> 00:24:58.804
and then that propagates to another agent,
00:24:58.804 --> 00:25:00.000
and then bad things happen.
00:25:00.000 --> 00:25:01.708
Look at the developer ecosystem and
00:25:01.708 --> 00:25:03.611
the supply chain that they have, right?
00:25:03.611 --> 00:25:05.323
And, you know, whether it's,
00:25:05.323 --> 00:25:07.647
you know, the IDE, like Tim mentioned,
00:25:07.647 -->
00:25:09.581
or the packages and libraries and images
00:25:09.581 --> 00:25:11.176
that they're using, and binaries.
00:25:11.176 --> 00:25:14.118
How do... And to answer your question,
00:25:14.118 --> 00:25:15.515
there are particular startups
00:25:15.515 --> 00:25:17.105
that are focused there on saying,
00:25:17.105 --> 00:25:19.288
"Hey, maybe we can take that off
00:25:19.288 --> 00:25:21.538
the plate and provide a pristine,
00:25:21.538 --> 00:25:24.472
you know, supply chain for the developer,
00:25:24.472 --> 00:25:27.333
where the packages are secure and signed
00:25:27.333 --> 00:25:29.462
off by a third party, to help decrease
00:25:29.462 --> 00:25:31.647
the amount of exposures that, you know,
00:25:31.647 --> 00:25:32.739
we might find and decrease
00:25:32.739 --> 00:25:34.000
that number of pull requests."
00:25:34.000 --> 00:25:36.021
So again, you know, we're sort of shifting
00:25:36.021 --> 00:25:37.609
left and moving at machine speed,
00:25:37.609 --> 00:25:39.175
and if you look at, you know,
00:25:39.175 --> 00:25:41.389
all the way from starting with the supply
00:25:41.389 --> 00:25:43.021
chain and the developer ecosystem
00:25:43.021 --> 00:25:45.000
to just the fundamentals of not putting,
00:25:45.000 --> 00:25:47.000
you know, secrets in code and, you know,
00:25:47.000 --> 00:25:48.644
all the other fundamentals from a coding
00:25:48.644 --> 00:25:50.000
perspective that we need to have.
00:25:50.000 --> 00:25:52.000
There's a lot that we can do there.
00:25:52.000 --> 00:25:54.222
There's a lot of startups and innovation
00:25:54.222 --> 00:25:56.000
that is happening in that space.
00:25:56.000 --> 00:25:58.714
And so, you know, it's a good segue
00:25:58.714 --> 00:26:01.429
into kind of the SDLC and the roles
00:26:01.429 --> 00:26:03.129
that people play traditionally
00:26:03.129 --> 00:26:05.000
and how those roles are changing.
00:26:05.000 --> 00:26:05.424
You know,
00:26:05.424 --> 00:26:07.118
we talk about two-pizza-sized teams.
00:26:07.118 --> 00:26:07.924
We've been talking
00:26:07.924 --> 00:26:09.000
about that for 15 years.
00:26:09.000 --> 00:26:11.667
I think Tim and I, when we met,
00:26:11.667 --> 00:26:13.861
we recently talked about squads, rather
00:26:13.861 --> 00:26:16.000
than thinking about sort of big teams.
00:26:16.000 --> 00:26:18.000
One of the things that I've been
00:26:18.000 --> 00:26:20.000
thinking about, though, is that,
00:26:20.000 --> 00:26:21.750
you know, there's this idea,
00:26:21.750 --> 00:26:23.438
as you described, you know,
00:26:23.438 --> 00:26:25.612
we can vibe code stuff,
00:26:25.612 --> 00:26:28.636
and your example is a great one.
00:26:28.636 --> 00:26:32.696
I think there is a differentiation between
00:26:32.696 --> 00:26:36.562
something that I vibe code and something
00:26:36.562 --> 00:26:38.112
that is really enterprise-ready,
00:26:38.112 --> 00:26:40.000
and that's a bunch of different things.
00:26:40.000 --> 00:26:42.766
You know, it could be our standards
00:26:42.766 --> 00:26:45.294
and practices, the tools we use,
00:26:45.294 --> 00:26:46.923
but also security. And like in
00:26:46.923 --> 00:26:48.824
your example, it's a great example,
00:26:48.824 --> 00:26:50.368
there was something that stopped
00:26:50.368 --> 00:26:52.105
you from including a secret in code.
00:26:52.105 --> 00:26:54.525
But I... One of the things that I'm
00:26:54.525 --> 00:26:57.083
a little concerned about is this idea
00:26:57.083 --> 00:26:59.630
that everybody can code and that everybody
00:26:59.630 --> 00:27:01.389
can move stuff to production,
00:27:01.389 --> 00:27:02.823
and I know we've talked
00:27:02.823 --> 00:27:04.444
about how we prevent that,
00:27:04.444 --> 00:27:06.250
but I do think that architecture
00:27:06.250 --> 00:27:08.000
still plays a significant role.
00:27:08.000 --> 00:27:11.308
And when we're coding, when we talk
00:27:11.308 --> 00:27:14.615
about shift left, what we literally
00:27:14.615 --> 00:27:17.308
mean is thinking security first, and it's
00:27:17.308 --> 00:27:20.000
what I'm talking to every customer about.
00:27:20.000 --> 00:27:22.059
Tim, how do you see that evolving?
00:27:22.059 --> 00:27:24.211
You know, and is there starting
00:27:24.211 --> 00:27:26.364
to be a recognition that, like,
00:27:26.364 --> 00:27:28.182
enterprise software has different
00:27:28.182 --> 00:27:30.000
conditions and requirements than,
00:27:30.000 --> 00:27:31.000
you know, like the World Cup
00:27:31.000 --> 00:27:32.000
app that Felix built, right?
00:27:32.000 --> 00:27:33.047
They're just...
00:27:33.047 --> 00:27:35.000
They're different use cases.
00:27:35.000 --> 00:27:37.002
Yeah, well, and I mean, first of all,
00:27:37.002 --> 00:27:39.167
defining what those standards should be.
00:27:39.167 --> 00:27:40.787
Not every organization
00:27:40.787 --> 00:27:42.333
has gotten there yet.
00:27:42.333 --> 00:27:44.125
And I think the grassroots
00:27:44.125 --> 00:27:45.227
nature of how...
00:27:45.227 --> 00:27:46.843
if we go back to the team sizes,
00:27:46.843 --> 00:27:48.864
for example, like what we started to see
00:27:48.864 --> 00:27:51.125
with coding assistants is the grassroots
00:27:51.125 --> 00:27:53.500
nature of just teams started dispersing...
00:27:53.500 --> 00:27:56.217
Not dispersing. They started forming sort
00:27:56.217 --> 00:27:59.000
of these subgroups where a couple of folks
00:27:59.000 --> 00:28:01.471
would be working with a fleet of agents,
00:28:01.471 --> 00:28:04.581
and that's just kind of indicative of the
00:28:04.581 --> 00:28:07.692
way things are going from these two-pizza
00:28:07.692 --> 00:28:10.655
teams to more like I'm a human actor
00:28:10.655 --> 00:28:13.947
with a squad of agents that I work with,
00:28:13.947 --> 00:28:15.839
which is a great model and is, back
00:28:15.839 --> 00:28:18.000
to the earlier term, a force multiplier.
00:28:18.000 --> 00:28:20.278
But again, it just accelerates.
00:28:20.278
--> 00:28:21.917
If we don't have the standards
00:28:21.917 --> 00:28:23.611
in place and have that defined,
00:28:23.611 --> 00:28:26.228
it's going to accelerate the challenges.
00:28:26.228 --> 00:28:27.667
And then even more so,
00:28:27.667 --> 00:28:30.337
like if you have sort of business citizen
00:28:30.337 --> 00:28:32.812
developers, as we call them sometimes.
00:28:32.812 --> 00:28:35.561
You know, if I have an environmental
00:28:35.561 --> 00:28:38.462
sciences degree and I'm writing an app
00:28:38.462 --> 00:28:40.943
using a Claude, for example, I'm probably
00:28:40.943 --> 00:28:43.000
not thinking a lot about security.
00:28:43.000 --> 00:28:44.500
Not that I don't want to,
00:28:44.500 --> 00:28:47.356
it just hasn't been part of my life
00:28:47.356 --> 00:28:50.294
like it has kind of in the IT space.
00:28:50.294 --> 00:28:52.617
I would... You know, I'm not a security
00:28:52.617 --> 00:28:55.000
engineer, but, you know, Felix, tell me,
00:28:55.000 --> 00:28:56.610
like, it's been, I think, enough
00:28:56.610 --> 00:28:58.421
of a challenge as security engineers
00:28:58.421 --> 00:29:00.316
getting the rest of IT to understand
00:29:00.316 --> 00:29:02.105
and care about security, let alone
00:29:02.105 --> 00:29:03.611
people that are maybe picking up coding
00:29:03.611 --> 00:29:05.000
for the first time and writing apps.
00:29:05.000 --> 00:29:07.389
So yeah, it just...
00:29:07.389 --> 00:29:10.909
my thinking continues to be,
00:29:10.909 --> 00:29:12.273
let's shift a lot of that onus
00:29:12.273 --> 00:29:13.409
into the platform itself.
00:29:13.409 --> 00:29:15.988
So the platform we give them to do this,
00:29:15.988 --> 00:29:18.438
you know, whatever, if it's a low-code
00:29:18.438 --> 00:29:20.936
platform, and then the hosting to take
00:29:20.936 --> 00:29:23.500
their app and sort of productionize it.
00:29:23.500 --> 00:29:25.000
Let's put the controls into that,
00:29:25.000 --> 00:29:26.213
and that way they don't have to think
00:29:26.213 --> 00:29:27.000
about it, which I'm sure
00:29:27.000 --> 00:29:28.091
they'll be happy with,
00:29:28.091 --> 00:29:29.947
and we get our sleep-at-night factor,
00:29:29.947 --> 00:29:31.000
as I like to call it.
00:29:31.000 --> 00:29:32.600
One other, you know,
00:29:32.600 --> 00:29:35.438
comment around that is we are seeing
00:29:35.438 --> 00:29:38.750
better collaboration between teams because
00:29:38.750 --> 00:29:42.239
now a developer has access to an AI agent
00:29:42.239 --> 00:29:45.643
that can guide them on the cybersecurity
00:29:45.643 --> 00:29:46.990
questions and concerns that they can
00:29:46.990 --> 00:29:48.000
bring to the security team.
00:29:48.000 --> 00:29:51.169
And so we are seeing examples of that,
00:29:51.169 --> 00:29:53.421
which is, you know, it's...
00:29:53.421 --> 00:29:55.827
I wouldn't... I don't wanna say rare,
00:29:55.827 --> 00:29:57.778
but we need to see more of it.
00:29:57.778 --> 00:29:59.312
But more informed developers now
00:29:59.312 --> 00:30:00.750
that they have access to tools
00:30:00.750 --> 00:30:03.000
that can help educate them on, you know,
00:30:03.000 --> 00:30:04.474
cybersecurity best practices
00:30:04.474 --> 00:30:06.000
as it relates to development.
00:30:06.000 --> 00:30:08.788
One of the things I think about as you
00:30:08.788 --> 00:30:11.429
guys are speaking is the longer tail
00:30:11.429 --> 00:30:15.907
of security responsibility because of the
00:30:15.907 --> 00:30:20.385
potentials for AI drift or model changes.
00:30:20.385 --> 00:30:22.604
You know, I think the way
00:30:22.604 --> 00:30:25.000
we think of those things...
00:30:25.000 --> 00:30:27.070
In some ways, when I deploy,
00:30:27.070 --> 00:30:29.583
I've met my security requirements,
00:30:29.583 --> 00:30:31.000
I move on to my next project.
00:30:31.000 --> 00:30:32.712
I'm not really thinking
00:30:32.712 --> 00:30:34.200
about that last one.
00:30:34.200 --> 00:30:36.765
We've all had experiences where we had to.
00:30:36.765 --> 00:30:38.348
I built a lot of Struts
00:30:38.348 --> 00:30:40.000
apps in the early 2000s,
00:30:40.000 --> 00:30:42.707
and then actually went someplace and
00:30:42.707 --> 00:30:45.789
rewrote the entire app to solve for that.
00:30:45.789 --> 00:30:48.057
But that, I think, was more the exception.
00:30:48.057 --> 00:30:50.000
That didn't happen, you know, a lot.
00:30:50.000 --> 00:30:52.467
Now, when you think about the long tail,
00:30:52.467 --> 00:30:53.824
how do we manage that?
00:30:53.824 --> 00:30:57.246
How do we have ongoing oversight
00:30:57.246 --> 00:31:00.455
of how the code is performing,
00:31:00.455 --> 00:31:03.236
but if there's AI inside the code,
00:31:03.236 --> 00:31:05.526
how that is performing based
00:31:05.526 --> 00:31:07.950
on how we expect it to when we started?
00:31:07.950 --> 00:31:10.000
Tim, do you wanna give it a shot?
00:31:10.000 --> 00:31:12.759
Yeah, I think for a lot of enterprise
00:31:12.759 --> 00:31:15.667
development and security organizations,
00:31:15.667 --> 00:31:16.624
it's long been,
00:31:16.624 --> 00:31:18.667
especially when we jokingly talk
00:31:18.667 --> 00:31:21.063
about the Struts apps and the Java
00:31:21.063 --> 00:31:23.529
of the early 2000s, it's long been,
00:31:23.529 --> 00:31:25.588
you know, we get this deployed
00:31:25.588 --> 00:31:27.647
out there and then we move on,
00:31:27.647 --> 00:31:29.885
and it's sort of the distinction
00:31:29.885 --> 00:31:32.333
between what is being developed as,
00:31:32.333 --> 00:31:34.142
like, little bug fixes and what lives
00:31:34.142 --> 00:31:36.000
in production is just this huge drift.
00:31:36.000 --> 00:31:38.400
And it could be six months,
00:31:38.400 --> 00:31:40.020
a year, even more, where somebody
00:31:40.020 --> 00:31:41.591
doesn't really touch production.
00:31:41.591 --> 00:31:43.002
And if no one's looking at it,
00:31:43.002 --> 00:31:44.318
we don't, you know, we've...
00:31:44.318 --> 00:31:45.965
So we've had these challenges around
00:31:45.965 --> 00:31:47.750
security of how do we make that secure?
00:31:47.750 --> 00:31:49.287
How do we, you know, have a process
00:31:49.287 --> 00:31:51.000
in place where vulnerabilities come up?
00:31:51.000 --> 00:31:53.691
And by the way, that's really challenging
00:31:53.691 --> 00:31:56.250
with the speed that Mythos is giving us
00:31:56.250 --> 00:31:58.432
vulnerabilities. But, you know,
00:31:58.432 --> 00:32:01.389
so it's just really important to have kind
00:32:01.389 --> 00:32:03.388
of that process in place. But I think
00:32:03.388 --> 00:32:05.333
the modernization of the process now
00:32:05.333 --> 00:32:07.667
is more important than ever because
00:32:07.667 --> 00:32:10.333
the speed is being put on us externally,
00:32:10.333 --> 00:32:12.000
and then also we're exacerbating
00:32:12.000 --> 00:32:14.250
it internally with the speed that we're
00:32:14.250 --> 00:32:16.500
using our coding assistants and agents.
00:32:16.500 --> 00:32:18.500
And so all of this is just compounding.
00:32:18.500 --> 00:32:20.998
So I think it's really important to relook
00:32:20.998 --> 00:32:23.438
at that process and understand, you know,
00:32:23.438 --> 00:32:25.817
what is our threat model, as well as who's
00:32:25.817 --> 00:32:27.857
responsible and who can, you know...
00:32:27.857 --> 00:32:30.718
how can we develop a process and a
00:32:30.718 --> 00:32:34.000
technology stack that can address that.
00:32:34.000 --> 00:32:37.222
Felix, I would think that for CISOs,
00:32:37.222 --> 00:32:41.692
this is really changing the way they think
00:32:41.692 --> 00:32:45.417
and manage the environments, right?
00:32:45.417 --> 00:32:48.559
Like it's a much longer process and a lot
00:32:48.559 --> 00:32:51.471
more training for, like you described,
00:32:51.471 --> 00:32:53.235
people and process to understand
00:32:53.235 --> 00:32:55.000
how they need to protect things.
00:32:55.000 --> 00:32:58.000
Yeah. I think a key point to make is...
00:32:58.000 --> 00:33:00.667
And Tim brought up Mythos, right?
00:33:00.667 --> 00:33:02.910
Is something that changed the way
00:33:02.910 --> 00:33:05.357
we look at, you know, cybersecurity.
00:33:05.357 --> 00:33:07.321
I think pre-Mythos we were having
00:33:07.321 --> 00:33:09.643
different conversations around securing
00:33:09.643 --> 00:33:12.579
AI. AI is the asset, and then post-Mythos
00:33:12.579 --> 00:33:15.300
we were talking about how do we use AI
00:33:15.300 --> 00:33:16.686
from a security operations perspective
00:33:16.686 --> 00:33:18.000
and exposure management perspective.
00:33:18.000 --> 00:33:20.708
And, you know, everything starts
00:33:20.708 --> 00:33:23.500
with people, process, governance,
00:33:23.500 --> 00:33:25.758
standards, and it sort of graduates
00:33:25.758 --> 00:33:27.500
at the fundamentals, right?
00:33:27.500 --> 00:33:29.821
Where we say, what are the fundamentals,
00:33:29.821 --> 00:33:32.083
right? And discovery is very important,
00:33:32.083 --> 00:33:33.386
right? And understanding
00:33:33.386 --> 00:33:35.667
what is internet-facing versus not, right?
00:33:35.667 --> 00:33:37.000
It's an element of prioritization, right?
00:33:37.000 --> 00:33:38.286
Because you wanna make sure
00:33:38.286 --> 00:33:40.000
you focus on what's most vulnerable.
00:33:40.000 --> 00:33:41.695
And sort of from there,
00:33:41.695 --> 00:33:44.643
you're taking the exposures that are out
00:33:44.643 --> 00:33:47.199
there and then further prioritizing them
00:33:47.199 --> 00:33:49.500
by validating exploitability, right?
00:33:49.500 --> 00:33:51.053
And saying, "Is something reachable?"
00:33:51.053 --> 00:33:52.832
Right? Going back to the example
00:33:52.832 --> 00:33:55.000
of my house, I have a safe in my house.
00:33:55.000 --> 00:33:58.019
If the safe lock isn't working,
00:33:58.019 --> 00:34:01.429
that is a very important, you know,
00:34:01.429 --> 00:34:03.573
a SEV 1 or a CVSS 10 exposure,
00:34:03.573 --> 00:34:05.789
but maybe it's not exploitable.
00:34:05.789 --> 00:34:07.895
Maybe I have a door in my office,
00:34:07.895 --> 00:34:09.809
and if I lock that door and I keep
00:34:09.809 --> 00:34:11.667
it locked, you know, you can't...
00:34:11.667 --> 00:34:12.599
It's not exploitable,
00:34:12.599 --> 00:34:14.286
and I could sort of deprioritize that.
00:34:14.286 --> 00:34:15.694
And I think as you sort of go
00:34:15.694 --> 00:34:17.200
through that and you say, "Hey,
00:34:17.200 --> 00:34:18.200
discovery is important," right?
00:34:18.200 --> 00:34:19.128
We have Mythos now.
00:34:19.128 --> 00:34:20.789
We can discover exposures in code.
00:34:20.789 --> 00:34:22.862
We can, you know, sort of at machine speed
00:34:22.862 --> 00:34:24.737
look at our attack surface management,
00:34:24.737 --> 00:34:26.198
what's internet-facing. We also
00:34:26.198 --> 00:34:27.895
need to look at the assessment piece
00:34:27.895 --> 00:34:30.609
and move at machine speed there because
00:34:30.609 --> 00:34:33.462
historically, in order for us to validate
00:34:33.462 --> 00:34:34.712
the exploitability of an exposure,
00:34:34.712 --> 00:34:36.000
that was a human-led effort, right?
00:34:36.000 --> 00:34:40.312
And now there are tools very focused on,
00:34:40.312 --> 00:34:42.221
you know, doing that at machine
00:34:42.221 --> 00:34:44.375
speed at the exposure level, right?
00:34:44.375 --> 00:34:46.453
Which helps deprioritize the amount
00:34:46.453 --> 00:34:48.056
of work that we have to do.
00:34:48.056 --> 00:34:48.999
And then beyond that,
00:34:48.999 --> 00:34:50.706
if something is exploitable, are there
00:34:50.706 --> 00:34:52.376
compensating controls that we can put
00:34:52.376 --> 00:34:54.000
in place to buy us more time, right?
00:34:54.000 --> 00:34:56.235
Is there a door, like I mentioned
00:34:56.235 --> 00:34:58.810
in my office, that I can close to say,
00:34:58.810 --> 00:35:01.464
"Hey, maybe the part is on backorder
00:35:01.464 --> 00:35:03.824
to fix that safe, but, you know,
00:35:03.824 --> 00:35:05.667
I have a compensating control," right?
00:35:05.667 --> 00:35:06.858
And then from a security
00:35:06.858 --> 00:35:08.000
operations perspective,
00:35:08.000 --> 00:35:10.994
I could say, "Well, there is an exposure
00:35:10.994 --> 00:35:14.062
that we have mitigated, so if there's any
00:35:14.062 --> 00:35:16.263
tampering or any intrusion that we can
00:35:16.263 --> 00:35:18.000
detect that affects that safe,
00:35:18.000 --> 00:35:20.045
it's a higher priority than everything
00:35:20.045 --> 00:35:21.929
else 'cause we know that, you know,
00:35:21.929 --> 00:35:23.000
that particular safe is exposed."
00:35:23.000 --> 00:35:25.070
So there's a lot that we can do from a,
00:35:25.070 --> 00:35:26.875
you know, fundamentals perspective
00:35:26.875 --> 00:35:28.912
to decrease the amount of work
00:35:28.912 --> 00:35:31.765
that we have to do in a post-Mythos world,
00:35:31.765 --> 00:35:34.034
and none of these things are new.
00:35:34.034 --> 00:35:36.786
We've known about them for years, right?
00:35:36.786 --> 00:35:38.920
You know, attack surface management, you
00:35:38.920 --> 00:35:41.000
know, prioritization, business context.
00:35:41.000 --> 00:35:43.101
Is it tier zero? Is it tier one?
00:35:43.101 --> 00:35:45.333
Is it tier two? Is it exploitable?
00:35:45.333 --> 00:35:46.783
Compensating controls and then
00:35:46.783 --> 00:35:48.667
remediation. These are all things that,
00:35:48.667 --> 00:35:51.167
you know, have been backed by,
00:35:51.167 --> 00:35:53.750
you know, frameworks like CTEM,
00:35:53.750 --> 00:35:55.469
Continuous Threat Exposure Management,
00:35:55.469 --> 00:35:57.368
that, you know, Gartner released. However,
00:35:57.368 --> 00:35:59.206
what has changed is now we need to move at
00:35:59.206 --> 00:36:01.000
machine speed across all of those, right?
00:36:01.000 --> 00:36:03.286
And so that is affecting people,
00:36:03.286 --> 00:36:05.000
process, and technology.
00:36:05.000 --> 00:36:07.778
I do hear, you know,
00:36:07.778 --> 00:36:09.741
two sides of vulnerability
00:36:09.741 --> 00:36:12.308
management in a post-Mythos world.
00:36:12.308 --> 00:36:13.608
One is, "Oh my God,
00:36:13.608 --> 00:36:16.071
there are all these vulnerabilities.
00:36:16.071 --> 00:36:18.369
Everybody should be very frightened."
00:36:18.369 --> 00:36:20.667
And the other is, like you just said,
00:36:20.667 --> 00:36:22.365
a lot of these vulnerabilities
00:36:22.365 --> 00:36:23.667
we've been living with,
00:36:23.667 --> 00:36:28.094
and they might not be as exposed
00:36:28.094 --> 00:36:31.000
or risky as we think.
00:36:31.000 --> 00:36:33.187
And I think, you know, the prioritization
00:36:33.187 --> 00:36:35.000
part of what you described is key,
00:36:35.000 --> 00:36:37.217
is understanding the context of what's
00:36:37.217 --> 00:36:39.375
happening and not just to
sort of run
00:36:39.375 --> 00:36:41.166
around in circles and be, "Oh my God,
00:36:41.166 --> 00:36:42.619
we're never gonna solve this."
00:36:42.619 --> 00:36:44.048
I think we've been solving it.
00:36:44.048 --> 00:36:46.103
I think that's the one part, and Felix,
00:36:46.103 --> 00:36:48.000
I'm curious about your view on this.
00:36:48.000 --> 00:36:50.120
We've gotten really good at security,
00:36:50.120 --> 00:36:51.667
and I'm not being arrogant,
00:36:51.667 --> 00:36:53.238
and I'm not saying that we should
00:36:53.238 --> 00:36:54.286
rest on those laurels,
00:36:54.286 --> 00:36:56.337
but we have gotten really good at it,
00:36:56.337 --> 00:36:58.333
and we're learning now, and I expect
00:36:58.333 --> 00:37:00.290
us to continue to be really good
00:37:00.290 --> 00:37:02.857
at it despite the challenges we're facing.
00:37:02.857 --> 00:37:05.000
Are you optimistic about that, Felix?
00:37:05.000 --> 00:37:07.000
Or, you know, where are you on that?
00:37:07.000 --> 00:37:11.139
Yeah, I'm optimistic that we know what we
00:37:11.139 --> 00:37:15.278
have to do in order to secure our assets.
00:37:15.278 --> 00:37:17.500
I think there are a lot of frameworks
00:37:17.500 --> 00:37:19.722
out there that guide us down the path
00:37:19.722 --> 00:37:22.968
of maturity and that, you know, really get
00:37:22.968 --> 00:37:25.750
us down to understanding our current
00:37:25.750 --> 00:37:27.418
state, future state, actionable roadmap
00:37:27.418 --> 00:37:29.000
around all the things that we can do.
00:37:29.000 --> 00:37:31.457
We've known that for a long time.
00:37:31.457 --> 00:37:34.286
I think the concern comes from a skill
00:37:34.286 --> 00:37:36.404
set and capacity perspective
00:37:36.404 --> 00:37:38.750
as it relates to machine speed.
00:37:38.750 --> 00:37:41.733
Generally what we see is organizations
00:37:41.733 --> 00:37:44.167
are investing in AI for reasons
00:37:44.167 --> 00:37:47.154
that are more business-focused, externally
00:37:47.154 --> 00:37:50.000
facing systems that customers are using,
00:37:50.000 --> 00:37:52.444
maybe internal systems
00:37:52.444 --> 00:37:55.000
to increase efficiency,
00:37:55.000 --> 00:37:56.895
decrease, you know, cost as a whole,
00:37:56.895 --> 00:37:58.000
soft cost as a whole.
00:37:58.000 --> 00:38:01.062
And the skill set necessary to move at
00:38:01.062 --> 00:38:04.286
machine speed from a cybersecurity lens,
00:38:04.286 --> 00:38:06.478
there's a bit of a gap there
00:38:06.478 --> 00:38:09.062
because we sort of have to borrow
00:38:09.062 --> 00:38:10.842
potential resources from that investment
00:38:10.842 --> 00:38:12.667
that organizations are making and saying,
00:38:12.667 --> 00:38:14.333
"How do we use those in security,"
00:38:14.333 --> 00:38:16.000
right? To sort of bridge that gap.
00:38:16.000 --> 00:38:19.292
How do we move at machine speed? How do
00:38:19.292 --> 00:38:22.500
we deploy triage agents so that we can
00:38:22.500 --> 00:38:25.431
go and get to a higher-confidence alert
00:38:25.431 --> 00:38:28.438
by having an agent do some of that work,
00:38:28.438 --> 00:38:29.689
front-ending it, and then
00:38:29.689 --> 00:38:31.190
eventually getting to a human.
00:38:31.190 --> 00:38:32.969
And then when it gets to a human,
00:38:32.969 --> 00:38:34.048
then having an agent
00:38:34.048 --> 00:38:36.343
help the human at investigations, and then
00:38:36.343 --> 00:38:38.529
when it comes to, you know, remediation,
00:38:38.529 --> 00:38:40.398
having an agent help a human to remediate,
00:38:40.398 --> 00:38:42.000
always having the human in the loop.
00:38:42.000 --> 00:38:44.534
So to me, it's not a concern
00:38:44.534 --> 00:38:47.250
of knowing what we have to do.
00:38:47.250 --> 00:38:48.826
Going back to the security EAB
00:38:48.826 --> 00:38:50.455
that we had a couple weeks ago,
00:38:50.455 --> 00:38:52.317
I think every CISO there had a really good
00:38:52.317 --> 00:38:54.091
understanding of what needed to be done.
00:38:54.091 --> 00:38:56.831
It was just a matter of do we have the
00:38:56.831 --> 00:38:59.643
resources to execute on that as quickly
00:38:59.643 --> 00:39:02.445
as we need to, and what's the risk
00:39:02.445 --> 00:39:05.000
associated with not doing that?
00:39:05.000 --> 00:39:08.182
Yeah, I'm advocating for everybody to be,
00:39:08.182 --> 00:39:11.053
like, growing their security budgets.
00:39:11.053 --> 00:39:13.896
I mean, it seems to be the most important
00:39:13.896 --> 00:39:16.600
thing an enterprise can do is recognize
00:39:16.600 --> 00:39:17.622
that they're gonna need
00:39:17.622 --> 00:39:19.000
to grow their security budgets.
00:39:19.000 --> 00:39:21.237
On that point, I mean, security budgets,
00:39:21.237 --> 00:39:23.250
and we had this statistic, you know,
00:39:23.250 --> 00:39:26.143
that we were discussing with the CISOs,
00:39:26.143 --> 00:39:27.925
there was a 5% increase
00:39:27.925 --> 00:39:30.250
in security budgets last year.
00:39:30.250 --> 00:39:31.933
You know, and if you look
00:39:31.933 --> 00:39:33.750
at what happened last year,
00:39:33.750 --> 00:39:35.366
you would think there'd be a 100%
00:39:35.366 --> 00:39:36.786
increase in security budgets.
00:39:36.786 --> 00:39:40.789
So we just have to do more with less.
00:39:40.789 --> 00:39:42.365
But what's positive about the whole
00:39:42.365 --> 00:39:44.211
thing is that we have these capabilities,
00:39:44.211 --> 00:39:45.462
we have these agents now,
00:39:45.462 --> 00:39:46.562
we have this AI capability
00:39:46.562 --> 00:39:48.000
that we can implement in security.
00:39:48.000 --> 00:39:50.000
It's just a matter of getting there.
00:39:50.000 --> 00:39:52.377
And I think as a little final
00:39:52.377 --> 00:39:55.000
point to add to that, this fear,
00:39:55.000 --> 00:39:57.892
understandably, around AI taking jobs,
00:39:57.892 --> 00:40:00.556
you know, anecdotally in the field,
00:40:00.556 --> 00:40:02.222
we're
just not seeing it.
00:40:02.222 --> 00:40:03.906
Very few, if any, of our clients
00:40:03.906 --> 00:40:05.800
are using AI in software development
00:40:05.800 --> 00:40:06.900
as a reduction-in-force,
00:40:06.900 --> 00:40:08.000
you know, type of event.
00:40:08.000 --> 00:40:11.562
And I think it goes back to this...
00:40:11.562 --> 00:40:13.737
back to the theme that security
00:40:13.737 --> 00:40:15.000
is everyone's job.
00:40:15.000 --> 00:40:17.657
As developers upskill and are using
00:40:17.657 --> 00:40:20.238
agents to make their lives easier,
00:40:20.238 --> 00:40:22.050
there's still a ton of work there,
00:40:22.050 --> 00:40:23.810
and I think some of that work is,
00:40:23.810 --> 00:40:25.327
as developers, we need to be
00:40:25.327 --> 00:40:27.333
security-conscious more so than ever.
00:40:27.333 --> 00:40:29.208
And we need to also contribute some
00:40:29.208 --> 00:40:31.458
understanding back to the folks making the
00:40:31.458 --> 00:40:33.145
process to help them understand how we can
00:40:33.145 --> 00:40:34.792
all do our jobs and secure assets better.
00:40:34.792 --> 00:40:37.856
So I just think it's an exciting time,
00:40:37.856 --> 00:40:38.824
and I think,
00:40:38.824 --> 00:40:41.034
as much as agreeing with what Felix
00:40:41.034 --> 00:40:43.056
is saying with AI helping, we're
00:40:43.056 --> 00:40:45.159
gonna need humans to be on the receiving
00:40:45.159 --> 00:40:47.000
end of that and helping drive that.
00:40:47.000 --> 00:40:49.281
Just to wrap up here, you know,
00:40:49.281 --> 00:40:51.562
what are the one thing or a few
00:40:51.562 --> 00:40:53.756
things you would tell your customers
00:40:53.756 --> 00:40:56.071
to think about for the next 12 months?
00:40:56.071 --> 00:40:58.999
What would you say that companies should
00:40:58.999 --> 00:41:02.000
focus on? Tim, do you wanna start us out?
00:41:02.000 --> 00:41:04.000
Yeah, I think in 2027,
00:41:04.000 --> 00:41:06.000
the highest-performing
00:41:06.000 --> 00:41:08.127
development organizations will have agents
00:41:08.127 --> 00:41:10.000
in four or more of their SDLC stages.
00:41:10.000 --> 00:41:11.667
Could be requirements,
00:41:11.667 --> 00:41:14.091
could be architecture, could be,
00:41:14.091 --> 00:41:15.903
obviously, coding assistants.
00:41:15.903 --> 00:41:17.778
Testing is gonna be a big one.
00:41:17.778 --> 00:41:20.694
Security is gonna be a big one. And I'm
00:41:20.694 --> 00:41:23.611
a big fan of everything being balanced,
00:41:23.611 --> 00:41:25.994
so, you know, it's valid in statistics
00:41:25.994 --> 00:41:27.812
and measurement and all this.
00:41:27.812 --> 00:41:30.687
But if we're going to implement that much
00:41:30.687 --> 00:41:33.421
AI and that many agents in our software
00:41:33.421 --> 00:41:34.728
development life cycles, I think
00:41:34.728 --> 00:41:36.444
my counterbalance to that is the platform.
00:41:36.444 --> 00:41:38.560
That governance platform that rides
00:41:38.560 --> 00:41:40.556
underneath applies to both humans
00:41:40.556 --> 00:41:42.500
and agents in equal measure.
00:41:42.500 --> 00:41:45.000
That needs to be a really big focus,
00:41:45.000 --> 00:41:46.326
almost in advance of getting
00:41:46.326 --> 00:41:47.368
these agents deployed.
00:41:47.368 --> 00:41:48.377
It's really hard 'cause
00:41:48.377 --> 00:41:49.737
they're already being deployed.
00:41:49.737 --> 00:41:51.661
But we need to, as much as we can, get
00:41:51.661 --> 00:41:53.636
that governance layer and that platform
00:41:53.636 --> 00:41:55.000
riding underneath that as soon as we can.
00:41:55.000 --> 00:41:56.877
Because that is a deterministic
00:41:56.877 --> 00:41:58.571
layer that pairs and couples
00:41:58.571 --> 00:42:01.000
with the probabilistic layer of AI.
00:42:01.000 --> 00:42:04.000
Great. And Felix?
00:42:04.000 --> 00:42:06.889
Yeah, and for me, again, it's the theme
00:42:06.889 --> 00:42:09.333
that we have been discussing, is,
00:42:09.333 --> 00:42:12.290
you know, focus on the fundamentals,
00:42:12.290 --> 00:42:15.000
focus on moving at machine speed.
00:42:15.000 --> 00:42:17.000
Always have a risk-based
00:42:17.000 --> 00:42:19.000
approach to prioritizing
00:42:19.000 --> 00:42:21.500
what you're doing within the business.
00:42:21.500 --> 00:42:23.593
Focusing on all exposures
00:42:23.593 --> 00:42:25.938
at once does not make sense.
00:42:25.938 --> 00:42:27.560
It's the riskiest exposures
00:42:27.560 --> 00:42:29.062
that we need to focus on,
00:42:29.062 --> 00:42:32.218
and understanding what's risky starts
00:42:32.218 --> 00:42:35.714
with GRC and doing some level of analysis
00:42:35.714 --> 00:42:37.982
on business impact and understanding,
00:42:37.982 --> 00:42:40.556
you know, what are the tier one, tier two,
00:42:40.556 --> 00:42:41.846
tier three applications,
00:42:41.846 --> 00:42:43.889
and what is the impact to the business
00:42:43.889 --> 00:42:45.468
of having those applications going
00:42:45.468 --> 00:42:47.000
down from a monetary perspective?
00:42:47.000 --> 00:42:49.109
And then reaching across the aisle to
00:42:49.109 --> 00:42:51.389
other businesses to understand the risk,
00:42:51.389 --> 00:42:53.330
and maybe signing off on that risk
00:42:53.330 --> 00:42:55.500
and saying, "We accept this risk," or,
00:42:55.500 --> 00:42:57.221
"We don't accept it and are willing
00:42:57.221 --> 00:42:58.500
to do something about it."
00:42:58.500 --> 00:42:59.773
So it's making an informed
00:42:59.773 --> 00:43:01.389
decision around what the risk is.
00:43:01.389 --> 00:43:04.251
And then once you understand that and
00:43:04.251 --> 00:43:07.500
where you're gonna focus your energy, it's
00:43:07.500 --> 00:43:09.206
a matter of looking at the fundamentals
00:43:09.206 --> 00:43:11.000
and saying, "We still need data security.
00:43:11.000 --> 00:43:13.228
We still need identity security.
00:43:13.228 --> 00:43:15.526
We still need micro-segmentation.
00:43:15.526 --> 00:43:16.721
We still need zero trust."
00:43:16.721 --> 00:43:18.421
We still need all these fundamentals,
00:43:18.421 --> 00:43:20.461
but now we need to rethink those
00:43:20.461 --> 00:43:22.500
fundamentals and move at machine
00:43:22.500 --> 00:43:25.234
speed 'cause our threat actors are,
00:43:25.234 --> 00:43:28.125
and we do not have time to, you know,
00:43:28.125 --> 00:43:30.526
not evaluate, you know, AI within
00:43:30.526 --> 00:43:33.000
our security operations practices.
00:43:33.000 --> 00:43:34.862
That's great. I really appreciate the
00:43:34.862 --> 00:43:36.875
conversation and both of your expertise.
00:43:36.875 --> 00:43:38.244
I expect we'll continue
00:43:38.244 --> 00:43:39.375
to talk about this,
00:43:39.375 --> 00:43:41.159
how it's changing, how it's impacting
00:43:41.159 --> 00:43:42.750
our day-to-day in the enterprise.
00:43:42.750 --> 00:43:44.442
And I look forward to connecting
00:43:44.442 --> 00:43:45.500
again in the future.
00:43:45.500 --> 00:43:48.468
This has been Ready, Set, Runtime,
00:43:48.468 --> 00:43:51.000
a podcast from AHEAD. I'm MK.
00:43:51.000 --> 00:43:53.364
We'll be back with another
00:43:53.364 --> 00:43:55.000
episode next week.
00:43:55.000 --> 00:43:56.862
Thank you for listening
00:43:56.862 --> 00:43:58.725
to Ready, Set, Runtime.
00:44:00.000 --> 00:44:01.289
If something in today's
00:44:01.289 --> 00:44:02.857
conversation sparked an idea
00:44:02.857 --> 00:44:05.659
or a question about your own environment,
00:44:05.659 --> 00:44:08.462
AHEAD's team would love to hear from you.
00:44:08.462 --> 00:44:10.353
Reach out anytime at ahead.com,
00:44:10.353 --> 00:44:12.000
and if you haven't already,
00:44:12.000 --> 00:44:15.000
follow the show wherever you're tuning in.
00:44:15.000 --> 00:44:17.353
The views shared on this podcast are those
00:44:17.353 --> 00:44:21.316
of the speakers and don't necessarily
00:44:21.316 --> 00:44:25.600
reflect those of AHEAD. Until next time.