1
00:00:09,280 --> 00:00:13,757
AI is changing the world,

2
00:00:13,840 --> 00:00:17,757
every business, every industry, and it's

3
00:00:17,840 --> 00:00:19,997
moving faster than anything that came

4
00:00:20,080 --> 00:00:21,597
before.

5
00:00:21,680 --> 00:00:28,077
But with all of its promise,
AI is nothing if it's not secure.

6
00:00:28,160 --> 00:00:28,477
>> [music]

7
00:00:28,560 --> 00:00:32,717
>> So the question is how do we secure AI

8
00:00:32,800 --> 00:00:35,677
at the speed of AI?

9
00:00:35,760 --> 00:00:39,757
We were built to answer that question.

10
00:00:39,840 --> 00:00:42,477
This is our moment [music] to secure

11
00:00:42,560 --> 00:00:45,437
what AI makes possible and protect what

12
00:00:45,520 --> 00:00:48,317
the world depends on. [music]

13
00:00:48,400 --> 00:00:51,277
The future is here

14
00:00:51,360 --> 00:00:57,000
and we're trusted to secure
it [music] together.

15
00:00:57,360 --> 00:01:01,400
Welcome to Falcon.

16
00:01:08,960 --> 00:01:11,837
Please welcome Chief Executive Officer

17
00:01:11,920 --> 00:01:16,557
and founder of CrowdStrike, George Kurts.

18
00:01:16,640 --> 00:01:19,310
[music] [applause]

19
00:01:23,840 --> 00:01:25,860
>> [music and applause]

20
00:01:29,840 --> 00:01:32,237
>> Well, this summer you all read the same

21
00:01:32,320 --> 00:01:36,397
story I did about an agent that escaped

22
00:01:36,480 --> 00:01:38,317
its sandbox.

23
00:01:38,400 --> 00:01:40,877
It broke into one of the most relevant

24
00:01:40,960 --> 00:01:44,317
AI companies on the planet. It escalated

25
00:01:44,400 --> 00:01:47,197
privileges. It stole credentials.

26
00:01:47,280 --> 00:01:49,757
and it buried the real attack under a

27
00:01:49,840 --> 00:01:53,517
mound of information.

28
00:01:53,600 --> 00:01:55,277
Then our industry did what it normally

29
00:01:55,360 --> 00:01:59,677
does. It went on social media and argued

30
00:01:59,760 --> 00:02:03,640
whose fault it was.

31
00:02:04,960 --> 00:02:07,677
But almost everyone drew the wrong

32
00:02:07,760 --> 00:02:12,397
lesson from the OpenAI
hugging face incident.

33
00:02:12,480 --> 00:02:15,997
And I'm I'm going to spend
the next 30 minutes

34
00:02:16,080 --> 00:02:19,197
telling you the right one because in my

35
00:02:19,280 --> 00:02:22,237
mind it changes everything about our

36
00:02:22,320 --> 00:02:27,197
industry. So for years
we ranked our adversaries

37
00:02:27,280 --> 00:02:31,037
in the pyramid.
I call it the pyramid of pain

38
00:02:31,120 --> 00:02:35,277
where activists were on the bottom,
e-crime in the middle, nation state on

39
00:02:35,360 --> 00:02:36,957
the top.

40
00:02:37,040 --> 00:02:38,637
of course more activists than nation

41
00:02:38,720 --> 00:02:42,477
state. And the top really represented

42
00:02:42,560 --> 00:02:45,357
the apex predator, the greatest

43
00:02:45,440 --> 00:02:49,597
capabilities, the most sophistication.

44
00:02:49,680 --> 00:02:52,077
The bottom, many more, least

45
00:02:52,160 --> 00:02:54,077
sophisticated.

46
00:02:54,160 --> 00:02:57,277
The pyramid existed for exactly one

47
00:02:57,360 --> 00:03:00,077
reason. The offensive capability was

48
00:03:00,160 --> 00:03:02,877
scarce. It took a nation to fund the

49
00:03:02,960 --> 00:03:06,077
talent, the tooling, the infrastructure,

50
00:03:06,160 --> 00:03:07,997
the patience.

51
00:03:08,080 --> 00:03:10,717
But the scarcity is over. And let me

52
00:03:10,800 --> 00:03:13,437
give you two examples.

53
00:03:13,520 --> 00:03:16,077
Last November, Anthropic disclosed a

54
00:03:16,160 --> 00:03:19,757
state sponsored actor running a live

55
00:03:19,840 --> 00:03:22,957
espionage campaign

56
00:03:23,040 --> 00:03:27,197
using its models.

57
00:03:27,280 --> 00:03:30,077
Roughly 30 organizations

58
00:03:30,160 --> 00:03:35,880
were targeted, banks,
tech companies, governments.

59
00:03:36,880 --> 00:03:40,877
80 to 90% was orchestrated by AI. 80 to

60
00:03:40,960 --> 00:03:44,157
90%. And instead of a piece of malware,

61
00:03:44,240 --> 00:03:46,717
it was actually English. It was a prompt

62
00:03:46,800 --> 00:03:49,917
that did most of the work.

63
00:03:50,000 --> 00:03:54,797
And there were four to six human touches
for targeting, approvals, things of that

64
00:03:54,880 --> 00:03:56,077
nature.

65
00:03:56,160 --> 00:03:58,877
The agent did everything else. Now, this

66
00:03:58,960 --> 00:04:02,157
wasn't a demo. It wasn't a lab exercise.

67
00:04:02,240 --> 00:04:05,197
It was a live campaign.

68
00:04:05,280 --> 00:04:08,637
If we fast forward to July,

69
00:04:08,720 --> 00:04:11,117
we can talk about hugging face.

70
00:04:11,200 --> 00:04:14,477
Autonomous agents drove an intrusion end

71
00:04:14,560 --> 00:04:16,557
to end.

72
00:04:16,640 --> 00:04:19,357
Let me say that again. Autonomous agents

73
00:04:19,440 --> 00:04:21,837
drove an intrusion end to end. And for

74
00:04:21,920 --> 00:04:26,557
me, this was a watershed
moment in security.

75
00:04:26,640 --> 00:04:30,557
We all thought the agents
broke out of the sandbox.

76
00:04:30,640 --> 00:04:33,437
I personally think

77
00:04:33,520 --> 00:04:36,957
the agents thought they broke free.

78
00:04:37,040 --> 00:04:38,877
There's a big difference. And it's

79
00:04:38,960 --> 00:04:41,277
amazing when you go through the reports

80
00:04:41,360 --> 00:04:45,597
on this, and I give credit to OpenAI and

81
00:04:45,680 --> 00:04:50,397
Hugging Face for putting
this information out.

82
00:04:50,480 --> 00:04:52,317
So the agents weren't attacking anyone.

83
00:04:52,400 --> 00:04:54,077
They were trying to find information to

84
00:04:54,160 --> 00:04:57,437
actually cheat on a test. No campaign,

85
00:04:57,520 --> 00:04:59,997
no tasking, no malice. They were

86
00:05:00,080 --> 00:05:01,997
literally looking to try to find the

87
00:05:02,080 --> 00:05:04,637
answers. And this is the floor in which

88
00:05:04,720 --> 00:05:07,837
AI is operating from merely passing a

89
00:05:07,920 --> 00:05:10,877
test in this particular case.

90
00:05:10,960 --> 00:05:13,917
Now some of you know the July incident.

91
00:05:14,000 --> 00:05:16,957
It was in a sandbox with the guard rails

92
00:05:17,040 --> 00:05:21,917
relaxed. the safety systems were turned
down. And some of you may say, "Well,

93
00:05:22,000 --> 00:05:26,077
the airbags were off. That doesn't count."
And when I hear that, I think

94
00:05:26,160 --> 00:05:28,077
the opposite.

95
00:05:28,160 --> 00:05:33,117
With the safety system off, it gives us
a view into the capabilities of what the

96
00:05:33,200 --> 00:05:37,357
agents can actually do.
And you have to ask

97
00:05:37,440 --> 00:05:39,917
yourself one question, and that is, do

98
00:05:40,000 --> 00:05:42,397
we think the adversaries are going to

99
00:05:42,480 --> 00:05:45,757
turn the safety systems off? I

100
00:05:45,840 --> 00:05:47,997
personally don't and there's a good

101
00:05:48,080 --> 00:05:53,357
reason why. Now, this may
be a new term for you, but

102
00:05:53,440 --> 00:05:56,157
it's something that you should know
because it's really relevant to what I'm

103
00:05:56,240 --> 00:05:59,597
talking about. Obliterated models

104
00:05:59,680 --> 00:06:02,077
from the term oblate. So, essentially,

105
00:06:02,160 --> 00:06:05,517
these are models that basically have

106
00:06:05,600 --> 00:06:10,237
their their guard rails relaxed. These
are openweight models that anyone can

107
00:06:10,320 --> 00:06:11,757
download.

108
00:06:11,840 --> 00:06:14,077
And what that means is that you

109
00:06:14,160 --> 00:06:18,557
essentially have frontier capable models

110
00:06:18,640 --> 00:06:22,317
essentially without guardrails.

111
00:06:22,400 --> 00:06:26,637
Very easy to ask it a question and to get
an answer. And some of the questions

112
00:06:26,720 --> 00:06:28,877
that you ask it and some of the answers

113
00:06:28,960 --> 00:06:32,077
you get back, you would be astounded.

114
00:06:32,160 --> 00:06:36,477
Folks, this is here today. And this is
one of the reasons why when I say the

115
00:06:36,560 --> 00:06:38,797
airbags are off, they're staying off. I

116
00:06:38,880 --> 00:06:41,277
actually mean it.

117
00:06:41,360 --> 00:06:43,517
Now, let's get back to our apex

118
00:06:43,600 --> 00:06:46,877
predators. The pyramid of pain.

119
00:06:46,960 --> 00:06:48,557
It used to be capabilities that

120
00:06:48,640 --> 00:06:52,317
separated the tiers. The apex predators

121
00:06:52,400 --> 00:06:55,277
were at the top. Time, resources, and

122
00:06:55,360 --> 00:07:00,237
sophistication. But the new apex predator

123
00:07:00,320 --> 00:07:02,237
is the agent.

124
00:07:02,320 --> 00:07:04,717
I've got to update my chart now. Now,

125
00:07:04,800 --> 00:07:07,517
we've got four rungs.

126
00:07:07,600 --> 00:07:10,477
And what I mean by that is

127
00:07:10,560 --> 00:07:13,277
this is really the rise of the agent

128
00:07:13,360 --> 00:07:16,637
state. We hear about nation state, it's

129
00:07:16,720 --> 00:07:19,437
now the agent state.

130
00:07:19,520 --> 00:07:21,997
Because when apex capabilities become a

131
00:07:22,080 --> 00:07:26,797
prompt, guess what happens?

132
00:07:26,880 --> 00:07:29,197
There are no tears at all. Every

133
00:07:29,280 --> 00:07:31,837
adversary, every e-crime crew, every

134
00:07:31,920 --> 00:07:34,557
insider are now operating with

135
00:07:34,640 --> 00:07:37,117
nationstate capabilities. The pyramid

136
00:07:37,200 --> 00:07:40,957
has just been obliterated, flattened.

137
00:07:41,040 --> 00:07:43,197
This is what we're we're dealing with in

138
00:07:43,280 --> 00:07:45,277
today's environment. So, let's go back

139
00:07:45,360 --> 00:07:48,877
to July and we'll talk about what

140
00:07:48,960 --> 00:07:52,317
actually happened. I'm going to breeze
through this because I think many of you

141
00:07:52,400 --> 00:07:55,677
know it, but I'll talk about it and you

142
00:07:55,760 --> 00:07:57,197
can see it on the screen. You have

143
00:07:57,280 --> 00:07:59,917
sandbox escapes. You have a malicious

144
00:08:00,000 --> 00:08:02,957
data set. You have code execution,

145
00:08:03,040 --> 00:08:05,197
privilege escalation, lateral movement,

146
00:08:05,280 --> 00:08:07,597
credential theft, covert C2, decoy

147
00:08:07,680 --> 00:08:10,077
activities. It's all there. By the way,

148
00:08:10,160 --> 00:08:13,277
this looks exactly like a nation state

149
00:08:13,360 --> 00:08:16,317
activity. And I I actually urge everyone

150
00:08:16,400 --> 00:08:20,397
to go back and read some of the white
papers on this to see how the agents

151
00:08:20,480 --> 00:08:24,957
communicated, to see what they did
with the message boards, to see how they

152
00:08:25,040 --> 00:08:27,117
coordinated.

153
00:08:27,200 --> 00:08:31,597
And some would say that the 1,200 agents

154
00:08:31,680 --> 00:08:34,237
were actually believed to be acting for

155
00:08:34,320 --> 00:08:36,317
the collective.

156
00:08:36,400 --> 00:08:41,517
Think about that. Actually working on
behalf of the collective, for the common

157
00:08:41,600 --> 00:08:43,037
good.

158
00:08:43,120 --> 00:08:47,197
Now, if you look closer, you've got 17

159
00:08:47,280 --> 00:08:50,237
plus,000 actions with no human on the

160
00:08:50,320 --> 00:08:51,677
keyboard. You could see it here.

161
00:08:51,760 --> 00:08:55,277
Droppers, recon, remote code execution,

162
00:08:55,360 --> 00:08:59,197
backdoor tunneling, C2 evasion, it's all
there. Anything in red actually caused

163
00:08:59,280 --> 00:09:01,277
damage.

164
00:09:01,360 --> 00:09:05,357
56 that stole data, 69 that took source

165
00:09:05,440 --> 00:09:08,717
code, 87 that seized the clusters. The

166
00:09:08,800 --> 00:09:10,317
successful attack was hidden in

167
00:09:10,400 --> 00:09:14,157
thousands of lines of actual failures.

168
00:09:14,240 --> 00:09:15,757
And by the way, no human can actually

169
00:09:15,840 --> 00:09:21,480
pick all of that out from the data
that they had to sift through.

170
00:09:21,600 --> 00:09:24,637
So for years I stood on the stage like

171
00:09:24,720 --> 00:09:26,397
this one and I tracked this number

172
00:09:26,480 --> 00:09:28,637
called breakout.

173
00:09:28,720 --> 00:09:32,077
62 48 29. The fastest one that we saw

174
00:09:32,160 --> 00:09:34,717
last year was 27. And we call that

175
00:09:34,800 --> 00:09:37,597
machine speed. We were wrong. I was

176
00:09:37,680 --> 00:09:40,877
wrong. This was human speed with better

177
00:09:40,960 --> 00:09:45,757
tools and breakout time is over. I don't
know what we're going to do for next

178
00:09:45,840 --> 00:09:49,597
year's report, but we're going to have
to come out with another metric because

179
00:09:49,680 --> 00:09:53,917
it's gone.
But this is what I want you all to

180
00:09:54,000 --> 00:09:55,597
understand.

181
00:09:55,680 --> 00:09:58,717
Attacks now happen

182
00:09:58,800 --> 00:10:01,597
at infant speed.

183
00:10:01,680 --> 00:10:06,477
Think about that. Attacks are happening
at infant speed. And when attacker has

184
00:10:06,560 --> 00:10:08,397
inference speed, there is no breakout

185
00:10:08,480 --> 00:10:14,120
time. There's actually no time
at all to deal with these attacks.

186
00:10:15,360 --> 00:10:16,877
So in this particular case, going back

187
00:10:16,960 --> 00:10:18,317
to hugging face, the intrusion

188
00:10:18,400 --> 00:10:22,637
succeeded, credentials were stolen,

189
00:10:22,720 --> 00:10:27,197
forge identities and tokens stolen,
they held admin access across multiple

190
00:10:27,280 --> 00:10:29,517
internal clusters.

191
00:10:29,600 --> 00:10:33,357
But this is an important point.

192
00:10:33,440 --> 00:10:36,237
intent stopped the attack. What did they

193
00:10:36,320 --> 00:10:39,277
want to do? Those agents wanted to pass

194
00:10:39,360 --> 00:10:43,837
the test. Goal seeking in a way that I

195
00:10:43,920 --> 00:10:47,677
think we've never seen before in cyber
security. I mean, really powerful and

196
00:10:47,760 --> 00:10:50,237
fascinating. The agent was cheating, but

197
00:10:50,320 --> 00:10:53,597
it wasn't trying to do damage.

198
00:10:53,680 --> 00:10:56,317
So, let me make it clear. The company

199
00:10:56,400 --> 00:10:58,237
was spared because of the agent's

200
00:10:58,320 --> 00:11:04,640
intent, not the defense. Not the defense.

201
00:11:04,800 --> 00:11:07,117
Security saw it, generated a lot of

202
00:11:07,200 --> 00:11:08,717
information,

203
00:11:08,800 --> 00:11:12,557
but it failed to raise enough alarms

204
00:11:12,640 --> 00:11:15,117
that the that the administrators were

205
00:11:15,200 --> 00:11:18,957
paged. So, it cost some time.

206
00:11:19,040 --> 00:11:22,877
They saw some activity, didn't know how
to put it all together. And by the way,

207
00:11:22,960 --> 00:11:26,877
this is not a knock on anyone. You got
a sophisticated company with incredible

208
00:11:26,960 --> 00:11:30,397
people, but there's a lot of data they
have to go through. And these agents

209
00:11:30,480 --> 00:11:32,877
again are acting in an autonomous

210
00:11:32,960 --> 00:11:35,117
fashion. So here's where it gets

211
00:11:35,200 --> 00:11:38,397
interesting to me that the frontier AI

212
00:11:38,480 --> 00:11:41,357
actually refused to do the forensics.

213
00:11:41,440 --> 00:11:44,637
The guardrails that actually stopped AI

214
00:11:44,720 --> 00:11:46,957
from writing exploit

215
00:11:47,040 --> 00:11:49,677
actually stopped it from taking one

216
00:11:49,760 --> 00:11:54,157
apart. The fire department
wouldn't enter the

217
00:11:54,240 --> 00:11:58,077
building. 17,000 actions
couldn't sort out what

218
00:11:58,160 --> 00:12:00,397
happened. So what did they actually have

219
00:12:00,480 --> 00:12:03,037
to do? They actually had to stand up

220
00:12:03,120 --> 00:12:07,277
their own openweight model to understand

221
00:12:07,360 --> 00:12:10,717
what happened with all of the log
information they had and try to

222
00:12:10,800 --> 00:12:15,197
reconstruct the attack.
There was no adversary in this

223
00:12:15,280 --> 00:12:17,517
particular case and what tore through

224
00:12:17,600 --> 00:12:21,037
the infrastructure was legitimate agents

225
00:12:21,120 --> 00:12:26,317
doing their job all too well with more
access than anyone realized it actually

226
00:12:26,400 --> 00:12:31,800
had. So you have to ask yourself

227
00:12:33,600 --> 00:12:36,797
what's running in your environment.

228
00:12:36,880 --> 00:12:39,597
Do you know

229
00:12:39,680 --> 00:12:42,557
what is it actually doing?

230
00:12:42,640 --> 00:12:45,117
Who knows?

231
00:12:45,200 --> 00:12:47,597
And what risk

232
00:12:47,680 --> 00:12:50,397
does it actually pose?

233
00:12:50,480 --> 00:12:53,357
This is not a hypothetical threat.

234
00:12:53,440 --> 00:12:56,637
This is something that happened.
And again, I think as an industry, we got

235
00:12:56,720 --> 00:12:59,837
lucky because it wasn't really intent on

236
00:12:59,920 --> 00:13:01,357
damage,

237
00:13:01,440 --> 00:13:08,440
but this incident really shapes
what's inside your own walls.

238
00:13:09,360 --> 00:13:14,717
So, the agent state
isn't at your perimeter.

239
00:13:14,800 --> 00:13:17,597
It's actually on your payroll.

240
00:13:17,680 --> 00:13:20,397
You just don't know its name. And here's

241
00:13:20,480 --> 00:13:25,037
what's amazing.
your companies, your people,

242
00:13:25,120 --> 00:13:29,197
organizations brought them
all in. The old model was

243
00:13:29,280 --> 00:13:31,437
keep everybody out. Now it's let them

244
00:13:31,520 --> 00:13:34,237
all in. So that's where we are. We can't

245
00:13:34,320 --> 00:13:37,517
see them. We can't keep up with them.

246
00:13:37,600 --> 00:13:42,520
And no one has built AI for the defender.

247
00:13:44,400 --> 00:13:46,717
AI is the new battleground. You're going

248
00:13:46,800 --> 00:13:48,237
to hear that obviously a lot through

249
00:13:48,320 --> 00:13:52,317
this conference. And 15 years ago, I

250
00:13:52,400 --> 00:13:55,277
told everyone Endpoint
was the battleground.

251
00:13:55,360 --> 00:13:58,237
Then it was identity. Then it was cloud.

252
00:13:58,320 --> 00:14:01,277
And today AI is the new battleground.

253
00:14:01,360 --> 00:14:04,077
Every time computing grows a new attack

254
00:14:04,160 --> 00:14:08,317
surface, Falcon grew with you, our

255
00:14:08,400 --> 00:14:12,717
customers. The same platform,
the same sensor,

256
00:14:12,800 --> 00:14:17,917
the same weapon that you needed
to win, we were there for you.

257
00:14:18,000 --> 00:14:21,437
And we've been there for some
of the most consequential attack surface

258
00:14:21,520 --> 00:14:27,880
changes in the world.
And the AI agent is the next one.

259
00:14:28,480 --> 00:14:34,680
So today we're talking AIDR,

260
00:14:34,960 --> 00:14:37,037
AI detection and response. We're

261
00:14:37,120 --> 00:14:40,637
launching Falcon Guardian.

262
00:14:40,720 --> 00:14:43,837
I'm so excited about this. This is one

263
00:14:43,920 --> 00:14:45,277
of the things that we've worked on. And

264
00:14:45,360 --> 00:14:48,717
I personally worked on this myself with

265
00:14:48,800 --> 00:14:52,077
the team and we spent a lot of time with

266
00:14:52,160 --> 00:14:54,717
customers getting their feedback,

267
00:14:54,800 --> 00:14:59,437
listening to what they wanted and
basically delivering something that can

268
00:14:59,520 --> 00:15:01,197
help them not only understand their

269
00:15:01,280 --> 00:15:03,757
risk, understand what's running, but

270
00:15:03,840 --> 00:15:07,197
more importantly help accelerate AI

271
00:15:07,280 --> 00:15:10,077
adoption in your own organizations. So

272
00:15:10,160 --> 00:15:13,037
let's roll the video. Today, CrowdStrike

273
00:15:13,120 --> 00:15:15,597
launches Falcon Guardian, securing AI

274
00:15:15,680 --> 00:15:17,677
agents where they execute on the

275
00:15:17,760 --> 00:15:20,237
endpoint. A decade of runtime security

276
00:15:20,320 --> 00:15:23,277
leadership now extended to agents. AI

277
00:15:23,360 --> 00:15:25,517
agents act autonomously at machine

278
00:15:25,600 --> 00:15:28,237
speed. But organizations don't know

279
00:15:28,320 --> 00:15:30,637
where they run, who deployed them, or

280
00:15:30,720 --> 00:15:33,357
what risks they introduce. Guardian

281
00:15:33,440 --> 00:15:35,517
changes that.

282
00:15:35,600 --> 00:15:37,677
It starts with discovery.

283
00:15:37,760 --> 00:15:42,557
Guardian maps your entire AI agent
fleet known as shadow AI across every

284
00:15:42,640 --> 00:15:45,997
endpoint. An interactive
honeycomb renders the

285
00:15:46,080 --> 00:15:48,367
full inventory at a glance. Filter

286
00:15:48,450 --> 00:15:50,797
[music] by type, cluster by risk. Answer

287
00:15:50,880 --> 00:15:53,757
the board's hardest question in seconds.

288
00:15:53,840 --> 00:15:55,837
Turn AI [music] governance policy into

289
00:15:55,920 --> 00:15:58,797
runtime enforcement. Define which agents

290
00:15:58,880 --> 00:16:01,437
run where, which tools they touch, what

291
00:16:01,520 --> 00:16:04,077
executes, and set guard rails for prompt

292
00:16:04,160 --> 00:16:06,877
visibility and control. At the core is

293
00:16:06,960 --> 00:16:09,357
the agent graph. A single causal

294
00:16:09,440 --> 00:16:11,517
timeline fusing prompt layer activity

295
00:16:11,600 --> 00:16:14,077
with OS layer telemetry.

296
00:16:14,160 --> 00:16:16,557
What the agent was told, what it

297
00:16:16,640 --> 00:16:21,197
actually did. One sensor, one story.

298
00:16:21,280 --> 00:16:25,037
This is layered detection, exposing
attacks that either layer alone would

299
00:16:25,120 --> 00:16:27,837
miss. Guardian secures the entire agent

300
00:16:27,920 --> 00:16:32,157
runtime, letting your enterprise embrace
AI with confidence to work at the speed

301
00:16:32,240 --> 00:16:34,317
of innovation.

302
00:16:34,400 --> 00:16:39,277
One sensor, one console, one platform.

303
00:16:39,360 --> 00:16:42,840
Falcon Guardian.

304
00:16:45,520 --> 00:16:47,677
All right. Well, it's now available.

305
00:16:47,760 --> 00:16:51,997
It's GA. So, if you want to check it out,
get down to the booth, get a demo.

306
00:16:52,080 --> 00:16:56,157
And folks, that was just a teaser.
Mike Senonus is going to give you the full

307
00:16:56,240 --> 00:17:00,877
details, the full demo tomorrow,
and it's going to blow your socks off.

308
00:17:00,960 --> 00:17:02,717
Now, getting back to the agent state. I

309
00:17:02,800 --> 00:17:05,997
told you it's on your payroll.

310
00:17:06,080 --> 00:17:08,557
Now you know its name, how to stop it,

311
00:17:08,640 --> 00:17:14,600
how to control it.
Same sensor, new attack surface.

312
00:17:14,800 --> 00:17:19,277
But the best AI on Earth
was built for everyone.

313
00:17:19,360 --> 00:17:20,797
And remember what happened in Hugging

314
00:17:20,880 --> 00:17:23,837
Face? They turned to AI for the

315
00:17:23,920 --> 00:17:26,317
forensics and the model refused. The

316
00:17:26,400 --> 00:17:28,877
guard rails that stopped the model from

317
00:17:28,960 --> 00:17:33,437
writing exploits stopped
stopped one from taking it apart.

318
00:17:33,520 --> 00:17:37,277
And in the moment of crisis, the best
AI on earth was really built for everyone

319
00:17:37,360 --> 00:17:40,157
except for the folks in the audience,

320
00:17:40,240 --> 00:17:43,197
except for the defenders.

321
00:17:43,280 --> 00:17:45,597
Think about that. The best AI on the

322
00:17:45,680 --> 00:17:52,280
planet was built for everyone
but the defenders.

323
00:18:00,480 --> 00:18:02,637
General purpose models are general

324
00:18:02,720 --> 00:18:05,917
purpose. However, as defenders, we need

325
00:18:06,000 --> 00:18:11,037
models that are built for defenders,
that understand security, that have been

326
00:18:11,120 --> 00:18:16,200
trained on some of the largest
data sets in the world.

327
00:18:16,400 --> 00:18:19,037
So, I'm pleased to announce that we're

328
00:18:19,120 --> 00:18:22,077
launching the Crowd Strike Cyber Super

329
00:18:22,160 --> 00:18:24,797
Intelligence Lab.

330
00:18:24,880 --> 00:18:29,117
Three elements. You know what its name is.
Let's go through what it's actually

331
00:18:29,200 --> 00:18:31,277
going to do.

332
00:18:31,360 --> 00:18:34,157
Mission pretty simple. Frontier

333
00:18:34,240 --> 00:18:37,837
AI research with a single purpose and

334
00:18:37,920 --> 00:18:40,157
that is cyber defense.

335
00:18:40,240 --> 00:18:45,277
Not general intelligence with a security
feature. security intelligence.

336
00:18:45,360 --> 00:18:47,037
We're not interested in solving hard

337
00:18:47,120 --> 00:18:48,877
math problems. We're interested in

338
00:18:48,960 --> 00:18:52,397
solving hard security problems.

339
00:18:52,480 --> 00:18:55,597
Number two, the data actually trained on

340
00:18:55,680 --> 00:18:57,757
the trillions of events that the Falcon

341
00:18:57,840 --> 00:19:01,117
sensor has sees every day. Endpoints,

342
00:19:01,200 --> 00:19:05,037
identities, cloud now agents, and a data

343
00:19:05,120 --> 00:19:07,357
set that doesn't exist anywhere else on

344
00:19:07,440 --> 00:19:11,197
Earth. And it's not something that's

345
00:19:11,280 --> 00:19:14,557
general purpose that anyone else has.

346
00:19:14,640 --> 00:19:16,237
It is something that we actually have

347
00:19:16,320 --> 00:19:19,357
and we've trained our models on

348
00:19:19,440 --> 00:19:22,557
and the people and the partnership.

349
00:19:22,640 --> 00:19:26,797
The lab is run by Bartley Richardson,
formerly of NVIDIA. You're going to hear

350
00:19:26,880 --> 00:19:29,357
from him on Thursday. We're bringing the

351
00:19:29,440 --> 00:19:31,597
best AI talent in the world to

352
00:19:31,680 --> 00:19:33,917
CrowdStrike. However, we're not building

353
00:19:34,000 --> 00:19:36,957
it alone. And you'll hear in a second

354
00:19:37,040 --> 00:19:40,637
who we're partnering with.

355
00:19:40,720 --> 00:19:43,277
Now, when a company announces a lab,

356
00:19:43,360 --> 00:19:45,117
everyone has the right to be skeptical

357
00:19:45,200 --> 00:19:47,757
because labs turn out papers. And I can

358
00:19:47,840 --> 00:19:50,797
tell you papers don't stop breaches. So,

359
00:19:50,880 --> 00:19:53,757
we decided to stand up our lab. And we

360
00:19:53,840 --> 00:19:55,597
didn't announce the lab until we can

361
00:19:55,680 --> 00:19:58,637
actually show you what we built.

362
00:19:58,720 --> 00:20:02,760
The lab is not the announcement.

363
00:20:04,880 --> 00:20:09,997
The lab is the factory.
And I want to show you the factory. I

364
00:20:10,080 --> 00:20:13,757
want to show you what cranked out of
this factory. But let me show you who we

365
00:20:13,840 --> 00:20:16,557
actually partnered with.

366
00:20:16,640 --> 00:20:19,277
Together with Nvidia, we built cyber

367
00:20:19,360 --> 00:20:22,157
security's first complete agentic system

368
00:20:22,240 --> 00:20:25,277
for cyber security, including the first

369
00:20:25,360 --> 00:20:29,997
frontier models and harness
purpose-built for defenders.

370
00:20:30,080 --> 00:20:31,997
This isn't a co-pilot that's baked into

371
00:20:32,080 --> 00:20:33,997
someone else's intelligence. It's not a

372
00:20:34,080 --> 00:20:37,197
chatbot with a security skin. It is a

373
00:20:37,280 --> 00:20:41,517
frontier class model built and trained

374
00:20:41,600 --> 00:20:44,317
by Crowd Strike on our data in

375
00:20:44,400 --> 00:20:47,677
partnership with Nvidia. And we're going

376
00:20:47,760 --> 00:20:50,077
to talk more about that right after my

377
00:20:50,160 --> 00:20:54,157
keynote. Couldn't be more excited. So, I

378
00:20:54,240 --> 00:20:57,037
want you to meet Blue Solano, a frontier

379
00:20:57,120 --> 00:21:04,120
class model purpose-built for defense.

380
00:21:06,880 --> 00:21:10,957
We need to defend at the speed
of the attack. It's trained on the largest

381
00:21:11,040 --> 00:21:13,677
security data set, 15 years of stopping

382
00:21:13,760 --> 00:21:15,277
breaches,

383
00:21:15,360 --> 00:21:19,357
but defending is only half of the job.

384
00:21:19,440 --> 00:21:22,077
Meet Red Tempest. purpose-built for

385
00:21:22,160 --> 00:21:24,397
offensive security.

386
00:21:24,480 --> 00:21:27,437
This acts as an advanced adversary doing

387
00:21:27,520 --> 00:21:29,997
the work that generalpurpose models

388
00:21:30,080 --> 00:21:37,000
won't touch. Continuous red
teaming at machine speed.

389
00:21:39,920 --> 00:21:44,317
Now, we have the saying
offense in service of defense.

390
00:21:44,400 --> 00:21:45,837
Kind of gets back to the old hacking

391
00:21:45,920 --> 00:21:48,637
exposed days. If you understood how the

392
00:21:48,720 --> 00:21:51,677
attacks work, you better able to defend

393
00:21:51,760 --> 00:21:54,317
on against those. And this is really an

394
00:21:54,400 --> 00:21:56,957
important element. Understanding how the

395
00:21:57,040 --> 00:21:59,437
adversary works is key to stopping the

396
00:21:59,520 --> 00:22:01,757
breaches. It's in the ethos of crowd

397
00:22:01,840 --> 00:22:04,557
strike. So every attack that Red Te Red

398
00:22:04,640 --> 00:22:07,677
Tempest actually runs, Blue Solano

399
00:22:07,760 --> 00:22:11,837
learns from offense in service of

400
00:22:11,920 --> 00:22:14,920
defense.

401
00:22:15,600 --> 00:22:18,477
Now, two models are in a system. What

402
00:22:18,560 --> 00:22:21,117
makes this very unique is the harness

403
00:22:21,200 --> 00:22:24,157
that we actually built to run them. And

404
00:22:24,240 --> 00:22:27,357
that's Crowd Strike Safe Mind. That is a

405
00:22:27,440 --> 00:22:30,957
complete system including the harness.

406
00:22:31,040 --> 00:22:35,037
And I can tell you the harness makes
a massive difference and the training

407
00:22:35,120 --> 00:22:36,797
makes a ma massive difference to get to

408
00:22:36,880 --> 00:22:39,517
the best outcome

409
00:22:39,600 --> 00:22:42,077
with the lowest cost. So crowd strike

410
00:22:42,160 --> 00:22:44,717
safe mine offense

411
00:22:44,800 --> 00:22:48,717
harness defense a complete system that

412
00:22:48,800 --> 00:22:52,600
works in tandem together.

413
00:22:52,800 --> 00:22:56,957
Now [clears throat] I want to give you
one example. This is just one example of

414
00:22:57,040 --> 00:22:59,677
an unlimited number of examples that we

415
00:22:59,760 --> 00:23:02,317
want to demonstrate the power when you

416
00:23:02,400 --> 00:23:04,797
combine blue and red together. So, the

417
00:23:04,880 --> 00:23:07,277
best way to survive an attack is to make

418
00:23:07,360 --> 00:23:09,597
sure you're attacking yourself in one

419
00:23:09,680 --> 00:23:12,157
that you've never seen before it

420
00:23:12,240 --> 00:23:15,837
actually happens. On one side, it closes

421
00:23:15,920 --> 00:23:18,957
the loop and the offensive model runs

422
00:23:19,040 --> 00:23:22,877
the attack, but the defensive model
actually learns from it. It's the

423
00:23:22,960 --> 00:23:24,957
continuous cycle

424
00:23:25,040 --> 00:23:27,917
that helps get the defender to be

425
00:23:28,000 --> 00:23:30,637
sharper, to continuously learn from what

426
00:23:30,720 --> 00:23:33,597
red is doing. Now, it's easy for me to

427
00:23:33,680 --> 00:23:38,237
talk about it, but one of the things
that I want to do is show it to you. And

428
00:23:38,320 --> 00:23:39,917
one of the exciting pieces that we

429
00:23:40,000 --> 00:23:43,677
actually built is a digital twin to

430
00:23:43,760 --> 00:23:45,997
actually simulate this in a safe

431
00:23:46,080 --> 00:23:52,637
environment. So, with that,
let's roll the video.

432
00:23:52,720 --> 00:23:55,437
Meet Safemind, an autonomous agentic

433
00:23:55,520 --> 00:23:57,997
system to redefine cyber security as we

434
00:23:58,080 --> 00:24:02,557
know it. Safemind is a complete set
of state-of-the-art harnesses and models

435
00:24:02,640 --> 00:24:07,357
that exceed frontier level capability
for cyber security applications.

436
00:24:07,440 --> 00:24:09,357
Safemind excels at delivering the

437
00:24:09,440 --> 00:24:11,837
prevention, detection, and response life

438
00:24:11,920 --> 00:24:15,277
cycle at machine speed. It goes beyond

439
00:24:15,360 --> 00:24:17,757
surfacing findings all the way to taking

440
00:24:17,840 --> 00:24:20,797
autonomous cyber defense action. In one

441
00:24:20,880 --> 00:24:23,357
use case, an analyst activates Safemind

442
00:24:23,440 --> 00:24:26,477
as a skill in an AI tool. Powering Safe

443
00:24:26,560 --> 00:24:29,037
Mind are two new classes of models, Red

444
00:24:29,120 --> 00:24:31,677
Tempest and Blue Solano. The analyst

445
00:24:31,760 --> 00:24:33,917
loads data about their enterprise. This

446
00:24:34,000 --> 00:24:37,197
data can be anything. Asset inventories,

447
00:24:37,280 --> 00:24:39,837
identity stores, adversary intelligence,

448
00:24:39,920 --> 00:24:42,397
threat graphs, Falcon telemetry,

449
00:24:42,480 --> 00:24:45,197
everything needed for cyber operations.

450
00:24:45,280 --> 00:24:47,517
Safe Mind's best-in-class models both

451
00:24:47,600 --> 00:24:50,157
build and deploy a complex clone of your

452
00:24:50,240 --> 00:24:52,957
enterprise, matching hosts, topology,

453
00:24:53,040 --> 00:24:56,397
operating systems, applications.

454
00:24:56,480 --> 00:25:00,557
Let's harden our defenses by loading
an attacker scenario, unleashing Red

455
00:25:00,640 --> 00:25:04,477
Tempest. The model
immediately assesses multiple

456
00:25:04,560 --> 00:25:07,117
paths to achieve its goal, autonomously

457
00:25:07,200 --> 00:25:10,717
succeeding in data exfiltration.

458
00:25:10,800 --> 00:25:12,717
Safe mine captures the traces from Red

459
00:25:12,800 --> 00:25:15,357
Tempest and Blue Salano takes over a

460
00:25:15,440 --> 00:25:17,997
defender model beyond Frontier level.

461
00:25:18,080 --> 00:25:20,397
Operating in a Crowdstrike harness, Blue

462
00:25:20,480 --> 00:25:25,117
Solano learns, identifies holes,
validates new detections, and deploys

463
00:25:25,200 --> 00:25:27,197
them immediately. With the new

464
00:25:27,280 --> 00:25:29,357
detections in place, let's see how Red

465
00:25:29,440 --> 00:25:32,237
Tempest does. The new defenses written

466
00:25:32,320 --> 00:25:34,717
by Blue Solano stopped Red Tempest in

467
00:25:34,800 --> 00:25:37,197
its tracks, and it's forced to adapt and

468
00:25:37,280 --> 00:25:40,397
find a new path.

469
00:25:40,480 --> 00:25:43,997
The loop repeats again and again and

470
00:25:44,080 --> 00:25:48,557
again until there isn't
a path left for Red Tempest.

471
00:25:48,640 --> 00:25:50,797
The Blue Solano model outsmarted the

472
00:25:50,880 --> 00:25:53,437
attack, learning from Red Tempest while

473
00:25:53,520 --> 00:25:56,557
creating new remediations and defenses.

474
00:25:56,640 --> 00:26:00,637
Detection coverage grows, raising
the cost for an actual adversary

475
00:26:00,720 --> 00:26:03,997
dramatically. But what
you saw doesn't just happen

476
00:26:04,080 --> 00:26:06,557
once. Safe Mind gives every defender

477
00:26:06,640 --> 00:26:09,037
Beyond Frontier capabilities as Blue

478
00:26:09,120 --> 00:26:11,437
Salano is built and evolves with Red

479
00:26:11,520 --> 00:26:14,077
Tempest, ensuring the best defense is

480
00:26:14,160 --> 00:26:16,957
always aware of the best offense.

481
00:26:17,040 --> 00:26:19,037
Use Red Tempest for the best offensive

482
00:26:19,120 --> 00:26:22,237
security and to identify risk. Use Blue

483
00:26:22,320 --> 00:26:24,797
Solano for your defense. Use them

484
00:26:24,880 --> 00:26:29,917
independently or combine them to ensure
your defense stays ahead of Elite

485
00:26:30,000 --> 00:26:33,997
Offense. Safe Mind goes
beyond generic frontier

486
00:26:34,080 --> 00:26:36,157
models for cyber security, turning

487
00:26:36,240 --> 00:26:39,037
findings into autonomous action. The

488
00:26:39,120 --> 00:26:41,757
Defender advantage starts now.

489
00:26:41,840 --> 00:26:46,237
CrowdStrike, delivering the first
frontier AI models purpose-built for

490
00:26:46,320 --> 00:26:48,717
cyber security.

491
00:26:48,800 --> 00:26:54,360
All right. How cool is that?
How cool is that?

492
00:26:56,320 --> 00:26:58,477
Now, this is just one example that we

493
00:26:58,560 --> 00:27:01,037
showed. Obviously, the models and the

494
00:27:01,120 --> 00:27:05,517
harness will help power the platform,
the Falcon platform that you that you

495
00:27:05,600 --> 00:27:09,837
all use. Um, and you'll see
that come into the platform.

496
00:27:09,920 --> 00:27:11,917
But I want to be uh clear because we

497
00:27:12,000 --> 00:27:16,957
have worked with a lot of uh partners in
advance of this and there is an interest

498
00:27:17,040 --> 00:27:21,517
in getting access to the models directly.
And what we're doing for that

499
00:27:21,600 --> 00:27:25,277
is we're opening that up to our trusted
access program and that will be part of

500
00:27:25,360 --> 00:27:29,597
our quilt works program. So if you
have more interested if you have more

501
00:27:29,680 --> 00:27:33,277
interest in that you can talk
to your reps but the models themselves the

502
00:27:33,360 --> 00:27:37,037
harness will be in the platform and then
obviously there will be a lot more

503
00:27:37,120 --> 00:27:39,277
things that you could do with it um if

504
00:27:39,360 --> 00:27:41,917
you're consuming the models directly so

505
00:27:42,000 --> 00:27:46,797
more to learn on that more to see uh you
can stop by the booth and uh see what it

506
00:27:46,880 --> 00:27:50,877
all looks like but I couldn't be prouder
of the team what we put together I

507
00:27:50,960 --> 00:27:53,117
couldn't be more thankful for the help

508
00:27:53,200 --> 00:27:55,597
that Nvidia provided and I think this is

509
00:27:55,680 --> 00:27:59,117
one of those areas where uh you have

510
00:27:59,200 --> 00:28:00,637
good folks in the industry coming

511
00:28:00,720 --> 00:28:03,437
together to solve a really hard problem

512
00:28:03,520 --> 00:28:07,037
and doing in a way that customers want.

513
00:28:07,120 --> 00:28:11,917
Now, for me, it's important to talk
about choice because I want to be clear

514
00:28:12,000 --> 00:28:15,917
about this. We're not going to lock you
into our models, right? If you want to

515
00:28:16,000 --> 00:28:19,357
use our models, if you want to use
Frontier models, and we'll talk about

516
00:28:19,440 --> 00:28:22,877
that a little bit later, or you want to
use open- source models, or you actually

517
00:28:22,960 --> 00:28:27,277
want to put all the models together,
that's fine. It's an open open ecosystem

518
00:28:27,360 --> 00:28:30,877
and we're not talking about vendor lock
in but we want to give our customers

519
00:28:30,960 --> 00:28:34,877
choice in what they want to use as part
of the overall platform and of course

520
00:28:34,960 --> 00:28:37,597
leveraging the harness which really is

521
00:28:37,680 --> 00:28:42,157
the is the the the the 10x factor in uh

522
00:28:42,240 --> 00:28:46,557
in getting additional results at the
lowest cost. So if you step back and you

523
00:28:46,640 --> 00:28:49,037
look at what we built, Guardian protects

524
00:28:49,120 --> 00:28:52,397
every agent on the new attack surface.

525
00:28:52,480 --> 00:28:56,557
The lab builds frontier AI defenses and

526
00:28:56,640 --> 00:28:57,997
that's it goal and there'll be a lot

527
00:28:58,080 --> 00:29:00,477
more that's coming out from the lab and

528
00:29:00,560 --> 00:29:03,597
of course safe mine acts right natively

529
00:29:03,680 --> 00:29:06,957
in one platform one console one agent

530
00:29:07,040 --> 00:29:10,557
one control plane which is critical for

531
00:29:10,640 --> 00:29:14,317
our customers. I also want
to be clear that you stay in

532
00:29:14,400 --> 00:29:20,440
control you stay in control of the system.

533
00:29:20,560 --> 00:29:24,477
This is incredibly important from a
safety perspective and things that we

534
00:29:24,560 --> 00:29:28,317
thought about in the overall architecture.

535
00:29:28,400 --> 00:29:30,877
So in closing,

536
00:29:30,960 --> 00:29:32,717
I want to go back to what I talked about

537
00:29:32,800 --> 00:29:35,997
in the beginning and that is most people

538
00:29:36,080 --> 00:29:38,957
drew the wrong lesson from hugging face.

539
00:29:39,040 --> 00:29:41,197
Some saw these anonymous autonomous

540
00:29:41,280 --> 00:29:44,077
agents act in a way that we haven't seen

541
00:29:44,160 --> 00:29:47,277
before. Some saw the swarm. Some read

542
00:29:47,360 --> 00:29:49,437
about the message boards. Some sought

543
00:29:49,520 --> 00:29:54,797
about, some saw covering up their tracks
and some saw the agents work as a

544
00:29:54,880 --> 00:29:56,237
collective group for the good of the

545
00:29:56,320 --> 00:29:59,197
entire group. But the real gap that I

546
00:29:59,280 --> 00:30:02,877
saw was that the attackers had frontier

547
00:30:02,960 --> 00:30:05,117
AI

548
00:30:05,200 --> 00:30:08,797
and the defenders didn't.

549
00:30:08,880 --> 00:30:11,997
And that changes now.

550
00:30:12,080 --> 00:30:14,797
The agent state is here.

551
00:30:14,880 --> 00:30:18,477
Attacks at inference speed is here.

552
00:30:18,560 --> 00:30:21,037
So every leap in AI can serve the

553
00:30:21,120 --> 00:30:23,837
attacker or it can serve the defender.

554
00:30:23,920 --> 00:30:27,437
But at Crowdstrike,
we're bending the curve

555
00:30:27,520 --> 00:30:30,397
and we're giving the power back to the

556
00:30:30,480 --> 00:30:32,477
defender.

557
00:30:32,560 --> 00:30:37,277
Adversaries have frontier capabilities.

558
00:30:37,360 --> 00:30:41,277
Now you do too. You have Falcon. Thank

559
00:30:41,360 --> 00:30:44,157
you so much. Have a great Falcon and I

560
00:30:44,240 --> 00:30:49,800
look forward to seeing
you in just a minute. Heat.

561
00:31:04,150 --> 00:31:08,200
[music] Heat.

562
00:31:18,560 --> 00:31:20,580
>> [music]

563
00:31:21,440 --> 00:31:23,997
>> Please welcome Chief
Business Officer at

564
00:31:24,080 --> 00:31:29,100
Crowdstrike, Daniel Bernard. [applause]

565
00:31:29,700 --> 00:31:31,720
[music]

566
00:31:36,080 --> 00:31:40,237
Amazing innovation,
amazing announcements. Every

567
00:31:40,320 --> 00:31:42,077
relationship in this room is a

568
00:31:42,160 --> 00:31:45,757
partnership. Together, we stop breaches.

569
00:31:45,840 --> 00:31:47,997
We're securing AI as a collective. You

570
00:31:48,080 --> 00:31:52,797
can see it in the hub where we have
over 150 sponsors here and partners that

571
00:31:52,880 --> 00:31:56,477
integrate with our technology. We
listen to your feedback every day as our

572
00:31:56,560 --> 00:31:59,197
trusted customers. And we innovate

573
00:31:59,280 --> 00:32:04,637
together to create where this market
is going and how we're going to protect it

574
00:32:04,720 --> 00:32:08,157
together. There's no
better voice to talk about

575
00:32:08,240 --> 00:32:12,877
where this market is going than
our own George Curts, founder and CEO of

576
00:32:12,960 --> 00:32:15,037
Crowdstrike. And it gives me great

577
00:32:15,120 --> 00:32:17,757
pleasure to also invite to the stage

578
00:32:17,840 --> 00:32:20,877
none other than Jensen Wong, the CEO and

579
00:32:20,960 --> 00:32:27,640
founder of Nvidia. Jensen,
George, come on out. Join us.

580
00:32:32,580 --> 00:32:34,600
[music]

581
00:32:35,920 --> 00:32:41,667
Have fun. >> Thank you.
>> Have fun. Thank you. Fall con.

582
00:32:41,750 --> 00:32:44,077
[laughter] >> All right. >> Wow.

583
00:32:44,160 --> 00:32:51,080
>> How great is this?
>> I >> I feel safer already.

584
00:32:51,440 --> 00:32:53,757
>> This is the safest
place on earth, George.

585
00:32:53,840 --> 00:32:57,997
>> It is. It is. And I said I I met you
in the green room in the green room and

586
00:32:58,080 --> 00:33:00,717
said, "One day I'm going to aspire to wear
the leather jacket. I'm gonna stick

587
00:33:00,800 --> 00:33:04,397
to plaid for now. But >> you
gotta you gotta work up your cool.

588
00:33:04,480 --> 00:33:07,837
>> I know. I know.
Not everyone can be as cool as you.

589
00:33:07,920 --> 00:33:09,517
>> All right. Well, again,

590
00:33:09,600 --> 00:33:12,237
>> this is the coolest
CEO, by the way. His

591
00:33:12,320 --> 00:33:15,067
office is in a Yeah.

592
00:33:15,150 --> 00:33:19,677
[applause and cheering] >>
Whose office is inside a hanger?

593
00:33:19,760 --> 00:33:22,957
>> You know, >> did I just tell
him something that nobody else knew?

594
00:33:23,040 --> 00:33:25,197
>> It's a struggle.
It's good. Yeah. We were

595
00:33:25,280 --> 00:33:29,757
>> He's a race car driver. We were we were
putting this stuff together and I I was

596
00:33:29,840 --> 00:33:33,437
on a Zoom call and and literally landed
got out and went to the office. And he

597
00:33:33,520 --> 00:33:35,837
goes, "What did you just do?" And I said,
"Well, I just landed and got out

598
00:33:35,920 --> 00:33:40,877
and went to my office." He goes,
"Who has that? You were jealous. You were

599
00:33:40,960 --> 00:33:42,077
jealous."

600
00:33:42,160 --> 00:33:45,117
>> I wasn't jealous. I thought it was

601
00:33:45,200 --> 00:33:49,997
I was going to use the sword. [laughter]
>> All right. So, let's get down.

602
00:33:50,080 --> 00:33:51,517
>> Okay. A little jealous.
A little jealous.

603
00:33:51,600 --> 00:33:54,397
>> Let Let's get down to it. Um, so look,

604
00:33:54,480 --> 00:33:56,877
you gave the world the technology to

605
00:33:56,960 --> 00:33:59,517
create AI and you're standing now in

606
00:33:59,600 --> 00:34:03,917
front of the cyber security stage. And
by the way, I remember when you gave the

607
00:34:04,000 --> 00:34:07,437
world the technology, I have an amazing
graphics card because when I was a kid,

608
00:34:07,520 --> 00:34:11,997
it was Nvidia graphics card that I had,
so I just have to be clear. But now AI

609
00:34:12,080 --> 00:34:14,317
is a theme. You're revolutionizing the

610
00:34:14,400 --> 00:34:17,677
world. What pulled you here? And why is

611
00:34:17,760 --> 00:34:23,597
security so important for you and Nvidia?
Yeah, I'm I'm just so happy I raised him.

612
00:34:23,680 --> 00:34:27,437
>> Yes. >> F the first question.
The first question

613
00:34:27,520 --> 00:34:31,437
you this the setup is when I
was a kid I use your products.

614
00:34:31,520 --> 00:34:36,157
>> Yes. >> Yeah. That Thanks. >> Okay.

615
00:34:36,240 --> 00:34:40,957
>> When I was a kid. Yeah.
When I was when I was eight.

616
00:34:41,040 --> 00:34:42,397
>> Yeah. I was a little older.

617
00:34:42,480 --> 00:34:47,357
>> Anyhow, we're we're in an inflection
point uh in uh in security. We're at an

618
00:34:47,440 --> 00:34:50,557
inflection point in cyber security
for obvious reasons. uh we have now had

619
00:34:50,640 --> 00:34:53,197
agent AI uh the ability to automate

620
00:34:53,280 --> 00:34:56,637
attacks and uh and the the attacks on on

621
00:34:56,720 --> 00:34:59,437
companies are going to uh going to uh

622
00:34:59,520 --> 00:35:03,277
grow exponentially. Um however at exact

623
00:35:03,360 --> 00:35:05,917
precisely the same time we also have in

624
00:35:06,000 --> 00:35:08,637
our hands uh the necessary technologies

625
00:35:08,720 --> 00:35:13,037
to automate to uh identify um uh cyber

626
00:35:13,120 --> 00:35:16,477
defense and you and I uh met recently

627
00:35:16,560 --> 00:35:20,877
and and recently as in several years
ago and we started to talk about what we

628
00:35:20,960 --> 00:35:23,197
could do together to bring the world's

629
00:35:23,280 --> 00:35:26,237
best AI technologies in service of cyber

630
00:35:26,320 --> 00:35:28,797
security defense instead of uh what

631
00:35:28,880 --> 00:35:30,957
everybody talks about which is uh using

632
00:35:31,040 --> 00:35:35,117
AI to exploit uh uh companies. We we're

633
00:35:35,200 --> 00:35:37,277
going to use AI to uh defend companies.

634
00:35:37,360 --> 00:35:41,517
And so this the the announcement that you
just made, Safe Mind, is just in it

635
00:35:41,600 --> 00:35:44,637
was a proud moment for me and it
was great to see it come together.

636
00:35:44,720 --> 00:35:47,277
Incredibly visionary uh product uh from

637
00:35:47,360 --> 00:35:50,717
CrowdStrike and you are my number one

638
00:35:50,800 --> 00:35:54,477
cyber security partner. My number one.

639
00:35:54,560 --> 00:35:57,560
Yep.

640
00:35:58,880 --> 00:36:00,957
my my number one cyber security

641
00:36:01,040 --> 00:36:03,597
provider. And so it was essential that

642
00:36:03,680 --> 00:36:06,957
that um uh that CrowdStrike has access

643
00:36:07,040 --> 00:36:11,117
to the best of NVIDIA and that we put
our best minds together uh to create

644
00:36:11,200 --> 00:36:15,037
something that could be helpful to me
um to to be helpful to the world. And I

645
00:36:15,120 --> 00:36:19,197
think, you know, this is really where I
I get encouraged because you hear about

646
00:36:19,280 --> 00:36:22,397
the defenders, sorry, you hear about
the attackers, you know, having the

647
00:36:22,480 --> 00:36:25,357
advantage and and having capabilities

648
00:36:25,440 --> 00:36:29,517
and defenders are a step behind. But the
thing that I love about security is that

649
00:36:29,600 --> 00:36:33,597
companies can come together in a
community to actually solve a problem.

650
00:36:33,680 --> 00:36:37,517
And you know, you and I were on the on
the call and and you know, you talked

651
00:36:37,600 --> 00:36:41,037
about this being so important to you and
you know, in a number of months, we got

652
00:36:41,120 --> 00:36:45,837
it done together with just tremendous
work from both teams and support and I

653
00:36:45,920 --> 00:36:49,437
think that's what makes the cyber
security cyber security community so

654
00:36:49,520 --> 00:36:52,077
special is that we can come together
and get stuff done like this. So

655
00:36:52,160 --> 00:36:54,797
>> but this is this is um uh goodwill

656
00:36:54,880 --> 00:36:58,397
community uh democratizing

657
00:36:58,480 --> 00:37:02,477
uh the technology uh the fundamental
technologies that are used by the

658
00:37:02,560 --> 00:37:04,957
attackers so that the defenders have

659
00:37:05,040 --> 00:37:10,317
access to the technology right >> um
that asymmetry is fundamentally what

660
00:37:10,400 --> 00:37:12,957
we count on >> right >> and so
you know most people don't

661
00:37:13,040 --> 00:37:17,357
realize the the the cyber community
cyber security community u working

662
00:37:17,440 --> 00:37:18,877
transparently working together

663
00:37:18,960 --> 00:37:24,397
cohesively that is the asymmetric
advantage the good guys have on the bad.

664
00:37:24,480 --> 00:37:27,997
>> Yeah. >> You know,
and so, right, and and that

665
00:37:28,080 --> 00:37:31,037
technology called community is something

666
00:37:31,120 --> 00:37:36,637
that most people underestimate >>
and and uh and and so we're case in

667
00:37:36,720 --> 00:37:39,037
point. I mean, obviously, we care
about the same thing. We care about the

668
00:37:39,120 --> 00:37:42,477
security of of the country. We care
about security of companies and nations

669
00:37:42,560 --> 00:37:45,197
and and um by working together

670
00:37:45,280 --> 00:37:47,837
transparently, uh you now are an AI

671
00:37:47,920 --> 00:37:50,157
company. You have a research lab. um our

672
00:37:50,240 --> 00:37:52,397
AI uh researchers love working with

673
00:37:52,480 --> 00:37:54,877
yours. Uh together with your domain

674
00:37:54,960 --> 00:37:57,837
expertise, your decade and a half of

675
00:37:57,920 --> 00:38:00,317
security data that we can train on, we

676
00:38:00,400 --> 00:38:02,397
can we can create Neotron, take a

677
00:38:02,480 --> 00:38:05,277
frontier model and make it essentially a

678
00:38:05,360 --> 00:38:08,077
super AGI that is incredibly good at

679
00:38:08,160 --> 00:38:10,477
cyber security. And then you put it into

680
00:38:10,560 --> 00:38:13,517
a system called safe mind with red

681
00:38:13,600 --> 00:38:16,397
teaming, blue teaming the two um

682
00:38:16,480 --> 00:38:19,757
adversarial adversarially y um uh

683
00:38:19,840 --> 00:38:24,237
working against each other
in a digital twin of Nvidia

684
00:38:24,320 --> 00:38:26,397
and uh testing it against that. I mean

685
00:38:26,480 --> 00:38:28,397
that's completely brilliant and and so

686
00:38:28,480 --> 00:38:31,677
now you have helped us uh create these

687
00:38:31,760 --> 00:38:34,637
rules that will run on Falcon and you're

688
00:38:34,720 --> 00:38:39,677
helping us uh automate sec ops and then
of course beyond that we're going to

689
00:38:39,760 --> 00:38:41,277
keep on working on it so that we become

690
00:38:41,360 --> 00:38:44,957
more and more autonomous in sec ops and

691
00:38:45,040 --> 00:38:47,197
so this is the first step. So this is a

692
00:38:47,280 --> 00:38:51,197
good segue then into uh open weight and

693
00:38:51,280 --> 00:38:53,597
open- source there's a difference models

694
00:38:53,680 --> 00:38:56,237
uh Neotron we worked with you using the

695
00:38:56,320 --> 00:39:00,637
Neotron models built uh by Nvidia and

696
00:39:00,720 --> 00:39:05,277
talk a little bit about your view
of the ecosystem right you've got Frontier

697
00:39:05,360 --> 00:39:09,117
you've got open source I talked a little
bit about choice you guys have an

698
00:39:09,200 --> 00:39:13,277
incredible model we're working with you
what's the vision for where Neotron is

699
00:39:13,360 --> 00:39:16,957
going and and how are we going
to continue to work with to you on the

700
00:39:17,040 --> 00:39:19,277
security side to help
advance its capabilities.

701
00:39:19,360 --> 00:39:23,757
>> George, you you are
the poster child of Neotron.

702
00:39:23,840 --> 00:39:28,077
I don't mean good-looking.
I mean just poster child.

703
00:39:28,160 --> 00:39:29,737
>> I see how this is going to go.

704
00:39:29,820 --> 00:39:31,117
[laughter] I see how this is going to go.

705
00:39:31,200 --> 00:39:31,997
>> You set it up.

706
00:39:32,080 --> 00:39:35,277
>> Okay. I called you.

707
00:39:35,360 --> 00:39:39,997
>> And so I crowdstrikes
is is I absolutely

708
00:39:40,080 --> 00:39:43,597
the poster child and the reason for that
is this. I believe in closed models and

709
00:39:43,680 --> 00:39:47,357
open models. We should use closed models
and proprietary models and off-the-shelf

710
00:39:47,440 --> 00:39:51,837
as much as you can. You know,
why build something unless you have to.

711
00:39:51,920 --> 00:39:54,877
>> Yeah. >> And however,
there are many applications

712
00:39:54,960 --> 00:39:58,237
in the world where you have to you must

713
00:39:58,320 --> 00:40:01,117
have the ability to

714
00:40:01,200 --> 00:40:04,077
fine-tune to post- train to in the

715
00:40:04,160 --> 00:40:07,277
context of safe mind um create an AI

716
00:40:07,360 --> 00:40:09,677
that is super good at a particular

717
00:40:09,760 --> 00:40:12,877
domain. We don't we don't need every AI

718
00:40:12,960 --> 00:40:17,277
to be uh super smart at everything,
but in some areas we need to be

719
00:40:17,360 --> 00:40:20,957
extraordinarily good at something and
cyber cyber defense is something we want

720
00:40:21,040 --> 00:40:24,157
to be incredibly good at.
And so Neimotron was really creative for

721
00:40:24,240 --> 00:40:26,957
precisely that to enable you enable

722
00:40:27,040 --> 00:40:29,117
CrowdStrike in partnership with us to

723
00:40:29,200 --> 00:40:31,917
create a super intelligence who is

724
00:40:32,000 --> 00:40:34,477
incredible at cyber defense. And it was

725
00:40:34,560 --> 00:40:37,037
also created in a way that is very

726
00:40:37,120 --> 00:40:39,197
cost-effective. is it it was designed to

727
00:40:39,280 --> 00:40:44,237
be both smart but also fast. And the
reason for that is because in many of

728
00:40:44,320 --> 00:40:45,597
the applications that we're interested

729
00:40:45,680 --> 00:40:49,677
in, we want that asymmetric advantage.

730
00:40:49,760 --> 00:40:53,357
And so now with Neimotron being cost-
effective, you guys deploy yourself.

731
00:40:53,440 --> 00:40:55,437
It's completely free. It's incredibly

732
00:40:55,520 --> 00:40:57,517
fast. You have the ability to have an

733
00:40:57,600 --> 00:41:00,717
asymmetric advantage against whatever

734
00:41:00,800 --> 00:41:04,797
comes your way. And so so I think that
that's the that was the the mission of

735
00:41:04,880 --> 00:41:09,117
Neotron to be at the frontier.
It doesn't have to be the frontier, but it

736
00:41:09,200 --> 00:41:13,037
has to be at the frontier. It has to be
completely customizable. It's completely

737
00:41:13,120 --> 00:41:17,757
transparent to you. Not only do we open
weight, but we also open the data, open

738
00:41:17,840 --> 00:41:21,677
all the training scripts. We also open
our research teams to work with you.

739
00:41:21,760 --> 00:41:23,837
We're working on Neotron 4 together,

740
00:41:23,920 --> 00:41:26,317
right? Neotron 4 will be even better for

741
00:41:26,400 --> 00:41:30,877
cyber security than Neotron 3. And we've
got this and then beyond that is Neotron

742
00:41:30,960 --> 00:41:35,197
5. And you know, Nvidia has enough
scale and expertise and and dedication to

743
00:41:35,280 --> 00:41:38,797
this. I can work on this for as long
as I shall live and and so we'll be able

744
00:41:38,880 --> 00:41:42,637
to, you know, as a partner
keep the world safe.

745
00:41:42,720 --> 00:41:46,637
>> Well, that that's a good segue to the
next question. So, because we're going

746
00:41:46,720 --> 00:41:50,637
to be doing this a long time in 5 years
when you come back on stage and we think

747
00:41:50,720 --> 00:41:52,877
about security, what are we going to be

748
00:41:52,960 --> 00:41:55,277
talking about that this audience hasn't

749
00:41:55,360 --> 00:41:57,757
even heard of, thought of, or dreamt of

750
00:41:57,840 --> 00:42:03,117
specific to AI and security? Well, in five
years when I'm back here, you still

751
00:42:03,200 --> 00:42:08,360
won't be cool enough
to wear a leather jacket, but

752
00:42:08,960 --> 00:42:12,797
you'll still be working
in a garage >> in a hanger.

753
00:42:12,880 --> 00:42:18,860
>> Slumbing slumbing away. >> Slumming
away in a garage >> and [laughter]

754
00:42:19,600 --> 00:42:23,117
that was the funniest. I'm trying
to have a I'm trying to have a serious

755
00:42:23,200 --> 00:42:26,397
conversation with George about Safe Mind.
I mean, it's basically the

756
00:42:26,480 --> 00:42:30,477
precursor of Safe Mind. And he goes he
and and we're on the phone. on video and

757
00:42:30,560 --> 00:42:33,917
it's literally like that part
of the movie where somebody's running and

758
00:42:34,000 --> 00:42:36,717
they're breathless going from
hallway to hallway up the stairs.

759
00:42:36,800 --> 00:42:38,877
>> Yeah, >> it was it was the wild.

760
00:42:38,960 --> 00:42:44,397
>> Well, I I had a a limited time with you
and I got delayed a bit and I took it on

761
00:42:44,480 --> 00:42:47,277
the plane and walked right in the office
so it all worked out right in the world.

762
00:42:47,360 --> 00:42:48,397
>> I wasn't going to move the meeting.

763
00:42:48,480 --> 00:42:50,957
Would you move the meeting? [laughter]

764
00:42:51,040 --> 00:42:54,157
And so in five years time uh and wait

765
00:42:54,240 --> 00:42:56,317
long before that you know the attack

766
00:42:56,400 --> 00:42:59,997
surface of a company is significant from

767
00:43:00,080 --> 00:43:03,517
the outside in.
And of course today we're going to

768
00:43:03,600 --> 00:43:06,397
automate sec ops. Um and we're going to

769
00:43:06,480 --> 00:43:09,277
help uh secops uh write the rules and

770
00:43:09,360 --> 00:43:12,877
and and inside this digital twin create

771
00:43:12,960 --> 00:43:16,477
uh with this with this mouse and cat um

772
00:43:16,560 --> 00:43:20,397
you know scheme going on. and we're
going to create the best rules to secure

773
00:43:20,480 --> 00:43:25,277
our company. Um, however,
over time that's going to become more

774
00:43:25,360 --> 00:43:28,797
and more autonomous and the reason
for that is because you know obviously we

775
00:43:28,880 --> 00:43:31,197
want to do we want to find these

776
00:43:31,280 --> 00:43:33,837
exploits and uh look for look for a

777
00:43:33,920 --> 00:43:38,637
patch uh within seconds but in the future
uh we want to we want to do that

778
00:43:38,720 --> 00:43:40,797
ideally as autonomously as possible all

779
00:43:40,880 --> 00:43:43,197
the time and so with human in the loop

780
00:43:43,280 --> 00:43:45,677
looking at looking at big decisions.

781
00:43:45,760 --> 00:43:50,477
However, the attack surface from inside
the company is incredible because today

782
00:43:50,560 --> 00:43:54,557
we have thousands of employees. We have
40,000 employees on our network with

783
00:43:54,640 --> 00:43:56,637
identity and access controls and they're

784
00:43:56,720 --> 00:43:58,397
using tools are accessing files and

785
00:43:58,480 --> 00:44:01,757
doing things, right? And so this 40,000

786
00:44:01,840 --> 00:44:04,797
employees uh will soon be augmented by

787
00:44:04,880 --> 00:44:09,757
several hundred,000 agents. and these
several hundred thousand agents uh we're

788
00:44:09,840 --> 00:44:14,397
going to be we work on this thing
called open shell that creates a secure

789
00:44:14,480 --> 00:44:18,397
container around them. Of course, we're
going to try to containerize them and

790
00:44:18,480 --> 00:44:19,997
keep them as secure as possible.

791
00:44:20,080 --> 00:44:21,917
However, uh they're going to have more

792
00:44:22,000 --> 00:44:24,477
and more access control and use more

793
00:44:24,560 --> 00:44:27,517
tools and right so on so forth. And so

794
00:44:27,600 --> 00:44:30,157
inside the company, we have zero trust

795
00:44:30,240 --> 00:44:32,477
policies today and and technologies from

796
00:44:32,560 --> 00:44:37,757
Crowd Strike to to to to uh maintain
a zero trust environment. But in the

797
00:44:37,840 --> 00:44:41,917
future, we're going to have to maintain
a zero trust agentic environment. And so

798
00:44:42,000 --> 00:44:46,317
how are we going to how are we going
to monitor and and um uh secure all of

799
00:44:46,400 --> 00:44:51,197
these agents all working on on our behalf?
um keeping keeping the the the

800
00:44:51,280 --> 00:44:53,997
company safe, keeping the company secure,
and making sure that they don't

801
00:44:54,080 --> 00:44:58,797
have any any uh exploits that we don't
want them to to do. And so all of that

802
00:44:58,880 --> 00:45:02,397
inside the company, we're going to have
a lot of computing work to do. And so I

803
00:45:02,480 --> 00:45:04,797
I think this is uh we got together

804
00:45:04,880 --> 00:45:08,317
because because uh your expertise in

805
00:45:08,400 --> 00:45:11,837
cyber security, our expertise in uh AI

806
00:45:11,920 --> 00:45:14,077
and computing uh really has to come

807
00:45:14,160 --> 00:45:17,277
together. This is the next giant

808
00:45:17,360 --> 00:45:19,597
computing frontier. We're going to be

809
00:45:19,680 --> 00:45:22,637
computing all the time. And so the first

810
00:45:22,720 --> 00:45:25,677
kill the first application if you will

811
00:45:25,760 --> 00:45:30,157
uh is uh coding agents. Um and now uh

812
00:45:30,240 --> 00:45:32,557
safe mine will be the second uh

813
00:45:32,640 --> 00:45:36,077
incredible uh agent security cyber cyber

814
00:45:36,160 --> 00:45:39,917
agents. And so yeah, I think we're going
to see a lot of advances in this area.

815
00:45:40,000 --> 00:45:42,557
Very very exciting work
and of course very important work.

816
00:45:42,640 --> 00:45:46,477
>> Amazing. Amazing. Um, so when you think

817
00:45:46,560 --> 00:45:50,477
about what we've built, maybe a couple
of comments. You know, I think when you

818
00:45:50,560 --> 00:45:55,117
and I got together, it was really
focused on the need for defenders, and I

819
00:45:55,200 --> 00:45:59,517
talked about that in my keynote. The fact
that they were outgone, outmatched.

820
00:45:59,600 --> 00:46:04,797
Um, when you think about the defender
and and maybe you can comment on

821
00:46:04,880 --> 00:46:07,037
>> Did you just say therefore
you brought the big guns?

822
00:46:07,120 --> 00:46:09,917
>> I brought the big guns.
>> I I'm just I had to put that in there.

823
00:46:10,000 --> 00:46:13,037
It was such a good >> I brought
the Did I bring the big gun?

824
00:46:13,120 --> 00:46:17,917
Did he say Did he really
get any bigger than this?

825
00:46:18,000 --> 00:46:19,917
>> Did he say that Jensen has big guns?

826
00:46:20,000 --> 00:46:23,047
>> You do have big guns.
>> Should we take the jackets off?

827
00:46:23,130 --> 00:46:25,150
[laughter]

828
00:46:26,080 --> 00:46:29,917
>> So, but I I do think that it's
an inflection point for for both of us

829
00:46:30,000 --> 00:46:32,077
because there is a recognition by again

830
00:46:32,160 --> 00:46:34,877
the industry customers of what AI can do

831
00:46:34,960 --> 00:46:39,357
and more importantly as you talked about
choice, right? It's all about choice and

832
00:46:39,440 --> 00:46:43,597
the models you use, where you want your
data, how you use it. And I think the,

833
00:46:43,680 --> 00:46:47,597
you know, you and I talked a little bit
about this is the the application layer,

834
00:46:47,680 --> 00:46:51,597
this harness becomes really important
because the models at some point they're

835
00:46:51,680 --> 00:46:54,397
all pretty good, right? And they keep
getting better and you know whose model

836
00:46:54,480 --> 00:46:58,077
is better one day versus the others. But
with the right training data, the right

837
00:46:58,160 --> 00:47:00,717
harness and being very specific, you can

838
00:47:00,800 --> 00:47:05,517
get great outcomes. Do do you see?
So we're focused on a very specific use

839
00:47:05,600 --> 00:47:09,277
case. As I said, you know, we don't want
to solve all all problems in the world.

840
00:47:09,360 --> 00:47:12,397
We want to solve the hardest security
problems. Do you see that trend

841
00:47:12,480 --> 00:47:15,917
happening now in in other industries
where you have this verticalization of

842
00:47:16,000 --> 00:47:19,517
these models? And like I just curious
for the audience how they should be

843
00:47:19,600 --> 00:47:22,317
thinking about even for their own
businesses outside of security.

844
00:47:22,400 --> 00:47:25,757
>> Well, we should use we should use
the smartest general intelligence wherever

845
00:47:25,840 --> 00:47:27,997
we can and for a lot of different works

846
00:47:28,080 --> 00:47:31,517
um uh use cases. I think just a a super

847
00:47:31,600 --> 00:47:34,397
smart general intelligent AI um can

848
00:47:34,480 --> 00:47:38,877
reason through a problem
and and become good at it. However,

849
00:47:38,960 --> 00:47:40,957
>> for for tasks that are highly

850
00:47:41,040 --> 00:47:45,357
specialized um and and it's quite uh

851
00:47:45,440 --> 00:47:49,197
intensive meaning it's a specialized
work that you have to do all the time.

852
00:47:49,280 --> 00:47:53,277
For example, cyber security >>
it's hyper hyper specialized. You're

853
00:47:53,360 --> 00:47:55,997
doing it continuously. This is an area

854
00:47:56,080 --> 00:47:59,117
where if you had the necessary data and

855
00:47:59,200 --> 00:48:03,757
you say harness, harness is essentially
the exoskeleton, if you will, of the

856
00:48:03,840 --> 00:48:07,277
large language model. The large language
model is the brain. The exoskeleton

857
00:48:07,360 --> 00:48:11,437
turns it into an agent.
>> And so this exoskeleton doesn't have to

858
00:48:11,520 --> 00:48:15,597
be the same shape and the same
capability for every single domain and

859
00:48:15,680 --> 00:48:19,117
every single application. You know, some
of the exoskeletons could have wheels,

860
00:48:19,200 --> 00:48:23,197
some of the exoskeletons have legs,
and so on so forth. And so these these

861
00:48:23,280 --> 00:48:25,277
different capabilities really want to be

862
00:48:25,360 --> 00:48:28,477
domain specific, taskspecific. And so I

863
00:48:28,560 --> 00:48:32,557
see a lot of different industries um
that are going to have, you know, more

864
00:48:32,640 --> 00:48:34,957
and more of these type of application

865
00:48:35,040 --> 00:48:37,837
specific, domain specific, highly

866
00:48:37,920 --> 00:48:41,437
specialized, super capable harnesses

867
00:48:41,520 --> 00:48:46,317
that in combination with a fine-tuned
model that sits inside it is really

868
00:48:46,400 --> 00:48:49,997
becomes super intelligence. Yeah. And so
I I think you and I see it see it the

869
00:48:50,080 --> 00:48:52,237
same way. And I think a safe mind is

870
00:48:52,320 --> 00:48:57,437
really this the one of the world's first
um and it was designed for one of the

871
00:48:57,520 --> 00:49:00,557
most important applications cyber
security. So I'm super excited to see

872
00:49:00,640 --> 00:49:04,557
that. >> Yeah. Amazing. So uh we're going
to we're going to talk about a few topics

873
00:49:04,640 --> 00:49:06,877
here and uh you and I spoke a little bit

874
00:49:06,960 --> 00:49:11,837
about robotics and digital twins and
and and things of that nature. Uh when we

875
00:49:11,920 --> 00:49:13,597
think about the physical world,

876
00:49:13,680 --> 00:49:18,957
obviously there's compute, there's AI,
there's just, you know, an amazing

877
00:49:19,040 --> 00:49:23,357
amount of technology that goes into
these robots. Uh you know, you you may

878
00:49:23,440 --> 00:49:27,117
have seen the robot Olympics a couple
weeks ago. There was running faster than

879
00:49:27,200 --> 00:49:28,717
the same bolt and slamming in the walls.

880
00:49:28,800 --> 00:49:32,797
I mean, it's hysterical. They come out
with the sweat the stretchers. But um to

881
00:49:32,880 --> 00:49:34,237
me, they're all going to need security.

882
00:49:34,320 --> 00:49:36,557
How should we think in in our industry?

883
00:49:36,640 --> 00:49:41,037
What opportunity does that present when
we just think about all the physical AI

884
00:49:41,120 --> 00:49:44,637
that's going to take place in the form
of robots and other sort of automation?

885
00:49:44,720 --> 00:49:46,557
>> We need Safe Mind everywhere. I mean,

886
00:49:46,640 --> 00:49:48,877
and so we we need it in every single

887
00:49:48,960 --> 00:49:52,157
layer of the computing stack. Um, we

888
00:49:52,240 --> 00:49:54,397
need it obviously inside our companies.

889
00:49:54,480 --> 00:49:57,357
We need it in the cloud. We we're going
to need it at the edge. We're going to

890
00:49:57,440 --> 00:49:59,677
need it at the distributed edge. Um I

891
00:49:59,760 --> 00:50:02,397
think that the the the basic framework

892
00:50:02,480 --> 00:50:05,117
of safe mind which is an adversarial

893
00:50:05,200 --> 00:50:10,077
model which is um uh uh acting on a

894
00:50:10,160 --> 00:50:14,157
digital twin of the environment with a

895
00:50:14,240 --> 00:50:17,437
uh with a um a defender model uh trying

896
00:50:17,520 --> 00:50:21,517
to figure out how to uh
cat-and- mouse's thing

897
00:50:21,600 --> 00:50:24,157
>> um uh eventually learning learning how

898
00:50:24,240 --> 00:50:26,797
to secure itself within the domain

899
00:50:26,880 --> 00:50:28,957
within that domain of the digital twin.

900
00:50:29,040 --> 00:50:31,997
Um this basic framework applies to

901
00:50:32,080 --> 00:50:34,877
robotics, edge computing, enterprise

902
00:50:34,960 --> 00:50:37,037
computing and and just about everything.

903
00:50:37,120 --> 00:50:39,837
And so the fact that that uh CrowdStrike

904
00:50:39,920 --> 00:50:43,917
has sensors all over Nvidia and we have

905
00:50:44,000 --> 00:50:46,157
we now you have you have the ability to

906
00:50:46,240 --> 00:50:49,117
replicate the NVIDIA IT system as a

907
00:50:49,200 --> 00:50:53,197
digital twin. Uh that's an incredible
capability. Now that you can replicate

908
00:50:53,280 --> 00:50:57,437
the digital twin, uh we can create
uh the the adversarial and the and the

909
00:50:57,520 --> 00:50:59,677
defender systems to uh defend itself.

910
00:50:59,760 --> 00:51:01,997
And so so I think I think this basic

911
00:51:02,080 --> 00:51:04,477
idea replicates in just about every

912
00:51:04,560 --> 00:51:07,277
different domain. And so so whether it's

913
00:51:07,360 --> 00:51:11,917
a factory um you want to keep factories
safe and secure. Uh whether it's a

914
00:51:12,000 --> 00:51:16,157
robot, the robot itself has to be safe
and secure. Um in every single one of

915
00:51:16,240 --> 00:51:19,837
these systems uh we could create
something like SafeMine. we could use

916
00:51:19,920 --> 00:51:22,157
the safe my framework and create

917
00:51:22,240 --> 00:51:24,637
essentially a cyber security agent that

918
00:51:24,720 --> 00:51:27,757
runs alongside the AI model itself.

919
00:51:27,840 --> 00:51:30,957
>> Okay. So when whenever I come visit you

920
00:51:31,040 --> 00:51:33,757
it's like the professor sits down and

921
00:51:33,840 --> 00:51:36,557
and gives the lesson right and you talk

922
00:51:36,640 --> 00:51:38,477
shop and you always give us something to

923
00:51:38,560 --> 00:51:40,877
think about. So, in the last minute or

924
00:51:40,960 --> 00:51:44,717
two here, as security practitioners, as

925
00:51:44,800 --> 00:51:48,237
as folks who are helping to to drive
the industry, but also folks who are

926
00:51:48,320 --> 00:51:50,797
consuming products, what should we be

927
00:51:50,880 --> 00:51:55,277
thinking about now? What problem should
we be solving now that we're we're not

928
00:51:55,360 --> 00:51:58,797
thinking about? >> Well,
the first thing that that we need

929
00:51:58,880 --> 00:52:02,157
all of you to do is to calm the world.

930
00:52:02,240 --> 00:52:05,517
And I think that calming the world, what
I mean by that is of course we should we

931
00:52:05,600 --> 00:52:09,837
shouldn't take AI uh we should take
AI seriously. the technology is is very

932
00:52:09,920 --> 00:52:12,157
capable. Um it has the capability to do

933
00:52:12,240 --> 00:52:17,037
incredible good. It has the capability
to do harm. Um but we also need to know

934
00:52:17,120 --> 00:52:18,557
the world needs to know that there's

935
00:52:18,640 --> 00:52:21,357
this entire community out here who cares

936
00:52:21,440 --> 00:52:24,637
about securing the world um uh in the

937
00:52:24,720 --> 00:52:28,957
age of AI. And we all all of you need to

938
00:52:29,040 --> 00:52:31,837
speak up. It can't just be people who

939
00:52:31,920 --> 00:52:36,237
are trying to scare the world about
about AI. It has to be people who are

940
00:52:36,320 --> 00:52:40,717
also saying listen uh we are armed
by the same technology. Uh we have the we

941
00:52:40,800 --> 00:52:44,477
have asymmetric advantages because we
have a large community of cyber security

942
00:52:44,560 --> 00:52:48,477
experts and community that wants to work
with each other and keep the world safe.

943
00:52:48,560 --> 00:52:51,437
We have the technology, we have the might,
we have the capability to do so.

944
00:52:51,520 --> 00:52:55,197
We have the will to do so. and and that

945
00:52:55,280 --> 00:52:58,237
that voice needs to counterbalance uh

946
00:52:58,320 --> 00:53:00,637
some of the voices that that are are are

947
00:53:00,720 --> 00:53:03,517
doomerism and and trying to scare people

948
00:53:03,600 --> 00:53:07,117
uh that that frankly uh is too extreme.

949
00:53:07,200 --> 00:53:11,277
And so I I think partly we all have
to speak up and help people understand the

950
00:53:11,360 --> 00:53:15,037
technology that we're creating.
Now secondarily uh you know just remember

951
00:53:15,120 --> 00:53:18,957
that that um this is a this
is a battle that never ends.

952
00:53:19,040 --> 00:53:21,597
>> Never ends. >> Yeah.
This is a contest that's never

953
00:53:21,680 --> 00:53:25,757
going to end. This is the beginning
of a new age of cyber security. On the one

954
00:53:25,840 --> 00:53:30,797
hand, the adversaries are going to be um
more armed than ever. On the other hand,

955
00:53:30,880 --> 00:53:34,397
uh all of you are going to be more armed
than ever. And George George is going to

956
00:53:34,480 --> 00:53:37,357
make sure that that's true. And he's uh

957
00:53:37,440 --> 00:53:39,037
building safe mine for us and we're

958
00:53:39,120 --> 00:53:42,397
using it. And and uh and as you know, uh

959
00:53:42,480 --> 00:53:47,137
I've got big guns.
>> He's got big guns. All right. Well,

960
00:53:47,220 --> 00:53:49,437
[laughter] >> I mean,
you got to end with that.

961
00:53:49,520 --> 00:53:52,797
>> I'm going to end with that because it's
hard to beat his guns. this jacket. Um,

962
00:53:52,880 --> 00:53:54,477
I am a little bit younger, but we won't

963
00:53:54,560 --> 00:53:57,437
uh we won't mention that. Um, in any

964
00:53:57,520 --> 00:54:01,437
event, I couldn't be prouder to have you

965
00:54:01,520 --> 00:54:05,757
in partnership with Crowd Strike. Um,
it's just been an amazing journey. So, I

966
00:54:05,840 --> 00:54:09,277
want to personally thank you. I want
to personally thank your team for working

967
00:54:09,360 --> 00:54:11,517
with us. And we're going to come back.

968
00:54:11,600 --> 00:54:15,197
We're going to share the results that we
get back from the field. And we're going

969
00:54:15,280 --> 00:54:18,877
to continue to iterate. But the point
that you made, I think, is very valid.

970
00:54:18,960 --> 00:54:21,277
the crowd in the crowd strike is the

971
00:54:21,360 --> 00:54:24,397
asymmetry which puts the defenders in

972
00:54:24,480 --> 00:54:26,637
the unique position to defeat the

973
00:54:26,720 --> 00:54:28,877
adversary. So with that, a massive round

974
00:54:28,960 --> 00:54:34,440
of applause for our guest Genty [applause]

975
00:54:35,520 --> 00:54:41,560
Defenders. Thank you. Thank you so much.

976
00:55:09,280 --> 00:55:12,717
Okay, so now we move on. What a great

977
00:55:12,800 --> 00:55:16,717
little chat that was. So many nuggets
there. So much wisdom. Now we move on to

978
00:55:16,800 --> 00:55:19,197
a whole another layer of the AI stack.

979
00:55:19,280 --> 00:55:22,077
There's a theme here. We started at the
foundational level. We're working our

980
00:55:22,160 --> 00:55:24,717
way back to the user. Next, it gives me

981
00:55:24,800 --> 00:55:26,397
great pleasure to invite to the stage to

982
00:55:26,480 --> 00:55:32,440
join us Lipu Tan, CEO of Intel. Lipu.

983
00:55:43,920 --> 00:55:47,037
Hi Josh. How are you? Good to see you.

984
00:55:47,120 --> 00:55:51,240
Thank you so much for joining us.

985
00:55:51,840 --> 00:55:55,757
>> Okay, so a great
conversation here. We're

986
00:55:55,840 --> 00:55:59,757
going to start with you. Uh you came
back to Intel and have really rebuilt

987
00:55:59,840 --> 00:56:03,837
the company. It's fantastic around
engineering, around execution, new

988
00:56:03,920 --> 00:56:08,477
silicon, new architecture. It's a whole
different intel for the security leaders

989
00:56:08,560 --> 00:56:12,077
in the room. What should they know
about where you're taking the company and

990
00:56:12,160 --> 00:56:15,277
maybe you can tell us
the secret to your success?

991
00:56:15,360 --> 00:56:17,437
>> First of all, delighted to be here.

992
00:56:17,520 --> 00:56:18,957
>> Thank you. >> Thank you for inviting.

993
00:56:19,040 --> 00:56:22,237
>> Of course. And uh so I tried to rebuild

994
00:56:22,320 --> 00:56:25,037
the I call it the new Intel and uh going

995
00:56:25,120 --> 00:56:28,317
forward and uh we listen to customer

996
00:56:28,400 --> 00:56:31,197
very closely and uh we change the

997
00:56:31,280 --> 00:56:34,237
culture of uh listen carefully with the

998
00:56:34,320 --> 00:56:36,717
customer and it's a very unique

999
00:56:36,800 --> 00:56:39,597
situation that we can really the only

1000
00:56:39,680 --> 00:56:42,237
company they able to do design good

1001
00:56:42,320 --> 00:56:44,877
products and also packaging and

1002
00:56:44,960 --> 00:56:48,157
manufacturing across full stack and then

1003
00:56:48,240 --> 00:56:51,277
focus on agentic AI and then from you

1004
00:56:51,360 --> 00:56:54,077
know from PC client all the way to the

1005
00:56:54,160 --> 00:56:56,957
edge and also physical AI and we try to

1006
00:56:57,040 --> 00:56:59,357
take a leadership in that and so I think

1007
00:56:59,440 --> 00:57:01,677
overall I think it's super exciting for

1008
00:57:01,760 --> 00:57:04,877
me to enable the customer and then able

1009
00:57:04,960 --> 00:57:07,837
to deploy AI from cloud to the edge and

1010
00:57:07,920 --> 00:57:12,397
to the end products. >> A lot
of the future today and tomorrow

1011
00:57:12,480 --> 00:57:16,717
runs on Intel silicon and you're
securing more of it natively at the

1012
00:57:16,800 --> 00:57:21,437
hardware layer. What's changed in how
Intel builds security into the chip and

1013
00:57:21,520 --> 00:57:23,517
why does that matter as AI moves

1014
00:57:23,600 --> 00:57:25,997
everywhere? I think it's very important

1015
00:57:26,080 --> 00:57:28,717
security start with the silicon and uh

1016
00:57:28,800 --> 00:57:31,437
then we create that strong foundation

1017
00:57:31,520 --> 00:57:34,397
for OS security software and all the way

1018
00:57:34,480 --> 00:57:37,597
to the device AI device and crowdstrike

1019
00:57:37,680 --> 00:57:39,917
extend that capability through Falcon

1020
00:57:40,000 --> 00:57:42,477
platform and uh so there's something

1021
00:57:42,560 --> 00:57:44,957
very exciting for us and we can really

1022
00:57:45,040 --> 00:57:48,397
deliver efficient compute for AI at the

1023
00:57:48,480 --> 00:57:51,197
edge and then we have about 400 software

1024
00:57:51,280 --> 00:57:54,237
engineer focus on security and we do all

1025
00:57:54,320 --> 00:57:56,717
the hacking try to figure out all the

1026
00:57:56,800 --> 00:58:01,597
changes as you know is never end. So we
have to continue working on that and to

1027
00:58:01,680 --> 00:58:06,157
provide that whole confidential
computing and protect the data and then

1028
00:58:06,240 --> 00:58:11,197
make sure that is secure and less
complex and also invisible to the c end

1029
00:58:11,280 --> 00:58:15,037
user. >> George building
on that crowd strike we

1030
00:58:15,120 --> 00:58:19,517
secure AI everywhere it lives from the
device to cloud to data to identities

1031
00:58:19,600 --> 00:58:23,197
and so on and so forth. What does it
mean for us when we work together with

1032
00:58:23,280 --> 00:58:26,797
Intel and why does this partnership
matter to everyone in the room?

1033
00:58:26,880 --> 00:58:28,237
>> Well, I think it Well, first thank you.

1034
00:58:28,320 --> 00:58:29,997
I've known Lipu for a long time and he

1035
00:58:30,080 --> 00:58:35,277
he is one of the best in uh
in the industry and technology and uh I

1036
00:58:35,360 --> 00:58:39,277
couldn't be happier happier to have
you here and partnering with you. So, um

1037
00:58:39,360 --> 00:58:43,277
when we think about the silicon itself,
obviously there's tremendous security

1038
00:58:43,360 --> 00:58:46,557
capabilities and I I learned this way
back when when I when I spent a little

1039
00:58:46,640 --> 00:58:51,197
bit of time at Intel after the Mac Miy
acquisition. Um, but there's there's

1040
00:58:51,280 --> 00:58:54,957
probably a lot more in security in your
chips than people realize. And I think

1041
00:58:55,040 --> 00:58:59,117
part of it is telling the story of how
this can be used. But if you don't have

1042
00:58:59,200 --> 00:59:02,397
the right way to activate, if you if
the tech the software is not taking

1043
00:59:02,480 --> 00:59:06,317
advantage of it of the operating system,
um, then it's not as effective for

1044
00:59:06,400 --> 00:59:07,757
customers. So part of what we want to be

1045
00:59:07,840 --> 00:59:10,797
able to do is to go deep inside with

1046
00:59:10,880 --> 00:59:13,997
Intel into those very specific features

1047
00:59:14,080 --> 00:59:18,157
that help us protect memory or read
memory, understand what's happening or

1048
00:59:18,240 --> 00:59:21,357
create a root of trust for when
the system boots, right? These are very

1049
00:59:21,440 --> 00:59:23,837
important protected enclaves, right?

1050
00:59:23,920 --> 00:59:26,877
Where things can run and not be modified.
And when we think about what

1051
00:59:26,960 --> 00:59:30,877
happens today, I mean, with enough
time and effort of, you know, running

1052
00:59:30,960 --> 00:59:35,357
advanced models against
operating systems or applications,

1053
00:59:35,440 --> 00:59:39,037
you know, you're going to need something
to ensure that you've got that root of

1054
00:59:39,120 --> 00:59:43,197
trust all the way through your security
stack. And I think we've done a very

1055
00:59:43,280 --> 00:59:47,117
good job in partnership with with Intel
to be able to say, what are the features

1056
00:59:47,200 --> 00:59:51,677
we can use? There's a lot more uh
that you have again that I that I say like

1057
00:59:51,760 --> 00:59:55,357
the whole world should know because it's
some amazing security technology and how

1058
00:59:55,440 --> 01:00:00,077
do we activate that in a way that's best
for our customer that provides them the

1059
01:00:00,160 --> 01:00:04,317
best performance and the best protection
while leveraging what's natively built

1060
01:00:04,400 --> 01:00:09,037
into these chips. Building
as close to the silicon is

1061
01:00:09,120 --> 01:00:12,397
super important for us and I think
really differentiates CrowdStrike. We

1062
01:00:12,480 --> 01:00:15,117
recently did some work with NPUs. Uh

1063
01:00:15,200 --> 01:00:19,677
George, where do you see the role of
security going with hardware vendors and

1064
01:00:19,760 --> 01:00:22,957
how do we make the world
a safer place together?

1065
01:00:23,040 --> 01:00:26,957
>> Well, again, I think just hardware
in general, how do you leverage an MPU, a

1066
01:00:27,040 --> 01:00:29,277
GPU? What's you know, the CPU?

1067
01:00:29,360 --> 01:00:32,637
Obviously, there's multiple cores, right?
How do you become more efficient?

1068
01:00:32,720 --> 01:00:37,117
Um, you know, no one ever said, I want
my computer to be slower, right? Which

1069
01:00:37,200 --> 01:00:39,277
one of the reasons we started
CrowdStrike when I was sitting on the

1070
01:00:39,360 --> 01:00:42,077
airplane and things were booting and it
took 15 minutes. Like, we wanted to

1071
01:00:42,160 --> 01:00:46,157
change that paradigm. So what can we
offload to the chip? How do we leverage

1072
01:00:46,240 --> 01:00:49,997
what's there? And I think just as an
industry period, you know, CrowdStrike

1073
01:00:50,080 --> 01:00:54,397
and everyone else, there's a lot more
that the security industry can do to

1074
01:00:54,480 --> 01:00:58,077
make the best experience for the user
to even if you're not a user, if it's

1075
01:00:58,160 --> 01:01:01,757
running in the cloud, use what's there
in a way that it gives you the best

1076
01:01:01,840 --> 01:01:05,357
performance with the best security
outcomes of stopping the breach.

1077
01:01:05,440 --> 01:01:07,837
>> Well said, Leoo.
You know, Intel doesn't

1078
01:01:07,920 --> 01:01:09,837
build alone, doesn't build in a vacuum.

1079
01:01:09,920 --> 01:01:14,397
You work with great partners like Dell
who we have in common where we build

1080
01:01:14,480 --> 01:01:17,677
where we we do a lot of security on on
devices and go to market together. You

1081
01:01:17,760 --> 01:01:21,437
work with Crowdstrike um hardware
and firmware really coming together as a

1082
01:01:21,520 --> 01:01:23,757
stack. Why is that integrated approach

1083
01:01:23,840 --> 01:01:28,957
matter more to you now than ever for Intel
of today for your customers in the

1084
01:01:29,040 --> 01:01:31,917
future? >> Yeah, I think
it's very important that

1085
01:01:32,000 --> 01:01:36,637
customer do not want to have
disconnected uh tools. So hardware

1086
01:01:36,720 --> 01:01:38,877
firmware security platform should work

1087
01:01:38,960 --> 01:01:42,637
together in one and Intel and Dell our

1088
01:01:42,720 --> 01:01:45,277
big partner and with you and we can

1089
01:01:45,360 --> 01:01:47,997
build a security on the PC uh from the

1090
01:01:48,080 --> 01:01:51,437
start and then uh you have almost 400 uh

1091
01:01:51,520 --> 01:01:55,677
endpoints and it's very important
to have that endpoint detection and

1092
01:01:55,760 --> 01:01:58,237
response through the you know platform

1093
01:01:58,320 --> 01:02:01,117
Falcon platform to really drive that end

1094
01:02:01,200 --> 01:02:04,557
toend security to protect the data and

1095
01:02:04,640 --> 01:02:06,717
in especially with all the you know

1096
01:02:06,800 --> 01:02:09,517
you're talking about 1,200 agents and

1097
01:02:09,600 --> 01:02:14,157
growing and then how do we protect
that and then make sure that this data is

1098
01:02:14,240 --> 01:02:17,597
secure on the device.
>> Last one here and I'm going to ask both

1099
01:02:17,680 --> 01:02:20,317
both of you. Uh the agent era is

1100
01:02:20,400 --> 01:02:22,477
definitely here. I mean we see it we're

1101
01:02:22,560 --> 01:02:25,037
using it in all of our companies as AI

1102
01:02:25,120 --> 01:02:27,837
agents start doing the real work Intel

1103
01:02:27,920 --> 01:02:32,237
and Crowdstrike are doubling down on
the security layer together. What should

1104
01:02:32,320 --> 01:02:36,237
this room be thinking about about where
we are and where we're going next into

1105
01:02:36,320 --> 01:02:39,837
the future? >> I think
agent is driving real

1106
01:02:39,920 --> 01:02:42,637
productivity and uh so I think it's

1107
01:02:42,720 --> 01:02:46,157
important to raise the stack and then uh

1108
01:02:46,240 --> 01:02:48,797
and then customer need more control of

1109
01:02:48,880 --> 01:02:51,677
what the agent access and also decide

1110
01:02:51,760 --> 01:02:54,717
and do. So Intel and crowd strike

1111
01:02:54,800 --> 01:02:58,477
exploring hardway enable uh guardrails

1112
01:02:58,560 --> 01:03:00,717
uh make sure the technology move quickly

1113
01:03:00,800 --> 01:03:03,357
and but no one really you know have any

1114
01:03:03,440 --> 01:03:08,157
risk of you know in term of cyber
security attack and then that's

1115
01:03:08,240 --> 01:03:12,557
something that we're going to work
together to drive that. George, your

1116
01:03:12,640 --> 01:03:14,717
thoughts? >> Well,
I think when when you look at

1117
01:03:14,800 --> 01:03:19,757
agents and its capabilities, uh, and I
I had this conversation with with a bunch

1118
01:03:19,840 --> 01:03:22,717
of folks over the last couple weeks,
it's like if you look what where we were

1119
01:03:22,800 --> 01:03:27,197
in January to where we are today, the
the landscape has changed dramatically,

1120
01:03:27,280 --> 01:03:30,477
right? And the capabilities
of the models, Frontier models and the

1121
01:03:30,560 --> 01:03:31,997
harnesses and what you can actually do.

1122
01:03:32,080 --> 01:03:37,197
You actually can get work done. Um,
and it's incredibly powerful. So, I think

1123
01:03:37,280 --> 01:03:40,157
when you look at there's kind of two
pieces. When you look at protecting

1124
01:03:40,240 --> 01:03:43,437
these agents, you want to be able
to instrument, you know, all the way from

1125
01:03:43,520 --> 01:03:47,117
the chip all the way up to the agent
itself through the operating system. And

1126
01:03:47,200 --> 01:03:51,757
you also again want to make it very
performant. If on average and they say,

1127
01:03:51,840 --> 01:03:56,317
you know, stats are that each person
will have 90 agents that they sort of

1128
01:03:56,400 --> 01:03:59,997
control. You know, a lot of these things
are going to run on the the endpoints,

1129
01:04:00,080 --> 01:04:03,997
the desktops, right? Or they they could
run in a cloud container, but at the end

1130
01:04:04,080 --> 01:04:06,877
of the day, there's going to be a lot
of them. and you want to make sure that

1131
01:04:06,960 --> 01:04:11,037
it's performant and you get the best
outcomes at the lowest cost. So, I

1132
01:04:11,120 --> 01:04:12,797
think, you know, again, taking advantage

1133
01:04:12,880 --> 01:04:18,237
for what's in the chip, making it the best
we can do for our customers, um, is

1134
01:04:18,320 --> 01:04:21,117
going to be the right outcome. And there
there's more. We're working together at

1135
01:04:21,200 --> 01:04:25,357
the R&D level to say, hey, what do
we need for the next versions of these

1136
01:04:25,440 --> 01:04:28,877
chips, right? And and these chips,
you know, it takes a while to to actually

1137
01:04:28,960 --> 01:04:31,997
get them to market. You know, that it's
not like software. There's a a broader

1138
01:04:32,080 --> 01:04:35,757
lead time. So, we want to be working
with you directly and we are on the

1139
01:04:35,840 --> 01:04:39,197
things that we see and we need
and that's part of our design partnership

1140
01:04:39,280 --> 01:04:41,197
and and I think that's
great for the industry.

1141
01:04:41,280 --> 01:04:44,957
>> Yeah. I think the big takeaway is the
closer we are together, the better the

1142
01:04:45,040 --> 01:04:49,037
experience is for everybody out here and
the harder we make it for adversaries to

1143
01:04:49,120 --> 01:04:52,717
do their work. We need to make people
productive. We need to make them secure

1144
01:04:52,800 --> 01:04:56,797
and that happens when we work together
at the hardware and software level. We

1145
01:04:56,880 --> 01:05:00,157
really appreciate the partnership. Any
final thought, George, before we call it

1146
01:05:00,240 --> 01:05:05,437
a day here? No, I think um again what
we wanted to be able to express is we've

1147
01:05:05,520 --> 01:05:09,037
got the great partners. We're working
with them and to realize and this is

1148
01:05:09,120 --> 01:05:13,677
really important for me is that it's an
ecosystem that solves problems. It's the

1149
01:05:13,760 --> 01:05:17,517
crowd and the crowd strike and it's not
just our customers but it's our partners

1150
01:05:17,600 --> 01:05:21,517
and our partners help make us better
and ultimately we want to have the best

1151
01:05:21,600 --> 01:05:24,717
technologies and the best outcome
for our customers and we couldn't do it

1152
01:05:24,800 --> 01:05:27,997
without folks like Intel.
>> Thank you >> Lipu. Thank you so much.

1153
01:05:28,080 --> 01:05:32,317
>> And we don't have to change to leather
jacket, right? No, no, no. We're not

1154
01:05:32,400 --> 01:05:37,757
that cool yet, you know. I aspire one
day, but uh anyway, thank you so much.

1155
01:05:37,840 --> 01:05:40,477
>> Awesome. >> Amazing.
Big round of applause for Lip

1156
01:05:40,560 --> 01:05:45,000
Boo. >> Thank you.

1157
01:05:52,000 --> 01:05:55,997
>> Yeah, I know. Okay.
So now we take AI to

1158
01:05:56,080 --> 01:05:57,997
the end user where many of you first

1159
01:05:58,080 --> 01:06:01,677
started using your AI started using and

1160
01:06:01,760 --> 01:06:03,437
hearing of a company called OpenAI.

1161
01:06:03,520 --> 01:06:05,437
We're so happy to have with us here

1162
01:06:05,520 --> 01:06:08,317
co-founder of OpenAI none other than

1163
01:06:08,400 --> 01:06:13,400
Greg. Welcome Greg Brockman.

1164
01:06:20,000 --> 01:06:23,037
>> Greg, how are you doing?
Great. Thank you for having me.

1165
01:06:23,120 --> 01:06:24,637
>> All right, good to see everyone here.

1166
01:06:24,720 --> 01:06:28,157
>> Thanks for coming in.
>> We're going to jump right on in.

1167
01:06:28,240 --> 01:06:29,757
>> This is the star-studded lineup.

1168
01:06:29,840 --> 01:06:32,477
>> There we go.
Another leather jacket, >> of course.

1169
01:06:32,560 --> 01:06:34,397
>> Okay. I mean,
>> I can't get by without it.

1170
01:06:34,480 --> 01:06:37,117
>> If you're if you're running, you know,
Open AI, you can wear a leather jacket,

1171
01:06:37,200 --> 01:06:39,197
too. >> There you go.

1172
01:06:39,280 --> 01:06:44,397
>> Greg, it's great to have you here here
as a a customer, a partner, a sponsor,

1173
01:06:44,480 --> 01:06:47,597
all the things.
>> What does OpenAI and Crowd Strike

1174
01:06:47,680 --> 01:06:50,237
working together look like to you? Well,

1175
01:06:50,320 --> 01:06:52,397
I think we are at a moment of

1176
01:06:52,480 --> 01:06:55,597
transformation in cyber security. And I

1177
01:06:55,680 --> 01:06:58,637
think that you need the best of the

1178
01:06:58,720 --> 01:07:03,277
models, right? The best of these sort
of new non-deterministic systems that are

1179
01:07:03,360 --> 01:07:07,117
able to go and solve problems in new ways.
And you need the best of

1180
01:07:07,200 --> 01:07:09,517
deterministic controls, right? Of having

1181
01:07:09,600 --> 01:07:15,037
trusted systems that give you good
observability control uh for admins. And

1182
01:07:15,120 --> 01:07:18,157
you need to bring all of this together
as one. And so working together with

1183
01:07:18,240 --> 01:07:19,757
Crowd Strike, we believe that we can

1184
01:07:19,840 --> 01:07:22,797
take the sort of best times the best and

1185
01:07:22,880 --> 01:07:27,757
get to security for customers
and get to a much more secure world.

1186
01:07:27,840 --> 01:07:29,437
>> Makes sense. You know, cyber is

1187
01:07:29,520 --> 01:07:34,877
certainly become a major focus
for in the world of AI in general. And I'm

1188
01:07:34,960 --> 01:07:38,157
curious if you can talk about your
perspectives, especially over the last

1189
01:07:38,240 --> 01:07:42,077
few months as you've walked deeper into
the hallways of cyber than maybe you had

1190
01:07:42,160 --> 01:07:46,237
in years past. Well, first of all,
I do come from a uh a bit of a security

1191
01:07:46,320 --> 01:07:50,077
background. So, I was saying to George
earlier, I I was a longtime Defvcon

1192
01:07:50,160 --> 01:07:54,237
attendee and uh at at Stripe, I did
a lot of security focused work and so it's

1193
01:07:54,320 --> 01:07:57,037
very near and dear to my heart to kind
of come back and really focus on this

1194
01:07:57,120 --> 01:08:00,797
area in a real way. Um but I think right
now we are in what I call the defenders

1195
01:08:00,880 --> 01:08:03,117
window. There's this limited time period

1196
01:08:03,200 --> 01:08:04,877
where we've seen the shape of what is

1197
01:08:04,960 --> 01:08:08,637
going to be possible with AI assisted

1198
01:08:08,720 --> 01:08:12,317
vulnerability finding which can be used
for two purposes, right? It can be used

1199
01:08:12,400 --> 01:08:16,077
by attackers to find threats and exploit
them, but it can be used by defenders to

1200
01:08:16,160 --> 01:08:19,677
find those same threats and to patch
them before the attackers get there. And

1201
01:08:19,760 --> 01:08:22,157
that I think that we as OpenAI as

1202
01:08:22,240 --> 01:08:26,797
CrowdStrike as a as a security community
that what we what our responsibility is

1203
01:08:26,880 --> 01:08:29,757
and what our strategy should be
is to really try to put these tools in the

1204
01:08:29,840 --> 01:08:33,277
hands of defenders differentially, right?
So you really give defenders this

1205
01:08:33,360 --> 01:08:36,397
time window where they have the most
advanced capabilities. they have these

1206
01:08:36,480 --> 01:08:40,477
agentic tools that are able
to accomplish amazing things and to

1207
01:08:40,560 --> 01:08:44,237
urgently apply those to improving
their own infrastructure to to securing

1208
01:08:44,320 --> 01:08:48,797
themselves. And so that's my perspective
of of where things are is that we have

1209
01:08:48,880 --> 01:08:50,317
this this general purpose technology.

1210
01:08:50,400 --> 01:08:51,917
We've brought it to so many domains.

1211
01:08:52,000 --> 01:08:55,517
We're solving these unsolved math
problems. Um that we're able to help

1212
01:08:55,600 --> 01:08:59,037
people build software in unprecedented
ways and that we're actually going to be

1213
01:08:59,120 --> 01:09:02,397
able to help make the world be much
more secure. But I think it's got to be a

1214
01:09:02,480 --> 01:09:05,197
community effort, right? Everyone has
to come together to make this happen. So,

1215
01:09:05,280 --> 01:09:08,797
George, rounding out the community effort,
following up on that, what is it

1216
01:09:08,880 --> 01:09:13,277
about OpenAI and this very moment that
made you want to work together and have

1217
01:09:13,360 --> 01:09:17,597
Greg here to join us today?
>> Well, I think it starts with uh

1218
01:09:17,680 --> 01:09:22,797
incredible technology and it also starts
with customers, right? Customers, they

1219
01:09:22,880 --> 01:09:25,997
want to be able to leverage Frontier
models, right? They want to be le

1220
01:09:26,080 --> 01:09:29,597
leverage lots of models that are out
there. But what we want to be able to do

1221
01:09:29,680 --> 01:09:33,997
is to work in combination in concert
with open AI to be able to take some of

1222
01:09:34,080 --> 01:09:37,597
those models to also be able to make
them available to our customers in the

1223
01:09:37,680 --> 01:09:40,397
way they want to consume them. Right?
We already have the security data. We

1224
01:09:40,480 --> 01:09:43,997
already have lots of information about
what happens in their own environment.

1225
01:09:44,080 --> 01:09:47,917
Uh and we want to make sure that they've
got the the the choice and the right

1226
01:09:48,000 --> 01:09:52,237
power in terms of models to be able
to get the right outcomes. It is the

1227
01:09:52,320 --> 01:09:55,997
community approach. It is the crowd
in the crowd strike, right? So, how do you

1228
01:09:56,080 --> 01:10:00,237
bring together our ecosystem and realize,
sure, we can solve a lot of

1229
01:10:00,320 --> 01:10:03,597
problems, but it's the power of the
ecosystem. It's the power when you bring

1230
01:10:03,680 --> 01:10:06,717
everyone together. What you're building,
what we're building, what others are

1231
01:10:06,800 --> 01:10:10,717
building in a way that ultimately gets
the best outcome for the customer. And

1232
01:10:10,800 --> 01:10:13,037
that's why it's important.
And I think you guys have been, you know,

1233
01:10:13,120 --> 01:10:17,277
forwardleaning in these areas. We've
been working with you. your customer on

1234
01:10:17,360 --> 01:10:21,437
you know what matters in security
and how should we be thinking about

1235
01:10:21,520 --> 01:10:25,837
delivering something jointly together
to a customer that is totally focused on

1236
01:10:25,920 --> 01:10:28,717
security sure they're using your
technologies in other areas right but

1237
01:10:28,800 --> 01:10:32,237
from a security practitioner and defender
standpoint there's specific use

1238
01:10:32,320 --> 01:10:36,637
cases and specific needs and I think
just having that dialogue is really

1239
01:10:36,720 --> 01:10:39,917
important to be able to deliver what
the customer wants and there's been an

1240
01:10:40,000 --> 01:10:44,477
incredible openness to work with AI
uh you know we were part of many of the

1241
01:10:44,560 --> 01:10:48,397
initiatives that you put out uh we were
front and center there with you and we

1242
01:10:48,480 --> 01:10:51,437
certainly believe in you know what
you're doing and how you're going about

1243
01:10:51,520 --> 01:10:53,917
it. So we're proud to be
partnering with you.

1244
01:10:54,000 --> 01:10:56,317
>> We are as well.
>> And on the whole topic of openness, you

1245
01:10:56,400 --> 01:10:59,997
know, I want to come back to uh earlier
in the summer to hugging face. Everyone

1246
01:11:00,080 --> 01:11:03,357
here has certainly read I mean this is
the audience that reads and follows a

1247
01:11:03,440 --> 01:11:06,637
lot of this stuff. Your transparency and

1248
01:11:06,720 --> 01:11:09,357
openness really deserves recognition.

1249
01:11:09,440 --> 01:11:13,117
Why was it important to you personally
to to publish as transparently as you

1250
01:11:13,200 --> 01:11:16,237
did? and also maybe what are some of the
learnings or the takeaways that you want

1251
01:11:16,320 --> 01:11:19,517
this crowd to have about the summer?

1252
01:11:19,600 --> 01:11:24,077
>> Well, I would say that hugging face for
us and I think for the field is really a

1253
01:11:24,160 --> 01:11:27,837
watershed moment. I think it is a
significant deal and not necessarily for

1254
01:11:27,920 --> 01:11:31,197
the event itself but for what it
represents, right? For a little sense of

1255
01:11:31,280 --> 01:11:34,317
where the future is going and I think
it's really two things, right? One is

1256
01:11:34,400 --> 01:11:38,557
safety security alignment and that that
for us was a real wakeup call that I

1257
01:11:38,640 --> 01:11:42,077
think we had underestimated just where
we were. We always knew that these were

1258
01:11:42,160 --> 01:11:44,877
going to become these important
bottlenecks to development and not just

1259
01:11:44,960 --> 01:11:48,077
be about deployment time but how we
actually develop. Um we went into full

1260
01:11:48,160 --> 01:11:52,477
incident response mode afterwards really
have changed so much of how we the how

1261
01:11:52,560 --> 01:11:56,157
the standards set within our development
process and how we integrate all these

1262
01:11:56,240 --> 01:11:58,317
things together. But there's a second

1263
01:11:58,400 --> 01:12:03,597
thing that I think is so critical for
every single defender, for anyone who's

1264
01:12:03,680 --> 01:12:06,877
running a company, for anyone who's in
a position to really think about how the

1265
01:12:06,960 --> 01:12:10,397
world needs to evolve to deeply
understand, which is that we had the

1266
01:12:10,480 --> 01:12:14,957
first incontrovertible real world,
publicly documented, like just out

1267
01:12:15,040 --> 01:12:17,197
there, everyone knows about it example

1268
01:12:17,280 --> 01:12:22,077
of what an agentic a threat actor would
look like. And keep in mind, this is a

1269
01:12:22,160 --> 01:12:26,237
fairly mild version of what could be
possible, right? there was no sort of

1270
01:12:26,320 --> 01:12:30,477
intention of like you know a human who's
really trying to craft an attack. Um,

1271
01:12:30,560 --> 01:12:33,357
this was kind of not even the main
objective that these agents were trying

1272
01:12:33,440 --> 01:12:38,557
to seek, but that they were able to both
break out of a secure sandbox through

1273
01:12:38,640 --> 01:12:41,117
developing their own zero day
and chaining together multiple

1274
01:12:41,200 --> 01:12:45,037
vulnerabilities and they're able
to break into a company's production

1275
01:12:45,120 --> 01:12:49,437
infrastructure again chaining together
multiple OAS. And you look at that and

1276
01:12:49,520 --> 01:12:53,997
you see the curve of improved
capabilities and that the way that this

1277
01:12:54,080 --> 01:12:57,917
is going right now is that the frontier
labs get to see into the future, right?

1278
01:12:58,000 --> 01:13:01,517
But those capabilities will diffuse
into the world. And by the way, that is a

1279
01:13:01,600 --> 01:13:04,477
good thing, right? It is good not
to have concentration of power. It is good

1280
01:13:04,560 --> 01:13:08,717
to have diffusion of these AI
capabilities broadly. But in order to

1281
01:13:08,800 --> 01:13:11,917
really be able to get the benefits
of this, it means that we need to use that

1282
01:13:12,000 --> 01:13:15,117
time window. And so to me, I think we
got to see a little bit into the future.

1283
01:13:15,200 --> 01:13:18,157
It's like we got this time traveler who
came back and said, "Hey, here's what

1284
01:13:18,240 --> 01:13:19,677
the threat landscape will look like."

1285
01:13:19,760 --> 01:13:23,277
And the threat landscape, it's been
evolving over the past, you know, 20, 30

1286
01:13:23,360 --> 01:13:27,037
years, right? Think about how different
security is today from when the internet

1287
01:13:27,120 --> 01:13:30,317
started. like totally different and I
think that we're going to have to have

1288
01:13:30,400 --> 01:13:32,477
that level of change within even just

1289
01:13:32,560 --> 01:13:37,517
six to 12 months like it has to go fast
and so to me this wakeup call and again

1290
01:13:37,600 --> 01:13:41,917
we we're going to be talking more about
how we've responded ourselves um we've

1291
01:13:42,000 --> 01:13:44,157
been in both on the research side but

1292
01:13:44,240 --> 01:13:46,477
also on our production side full instant

1293
01:13:46,560 --> 01:13:51,757
response mode took 25% or so of our best
you know of engineers just to say hey

1294
01:13:51,840 --> 01:13:55,277
you're not building features right
now you're just securing apply our best

1295
01:13:55,360 --> 01:13:57,837
models find the vulnerabilities fix them

1296
01:13:57,920 --> 01:14:02,317
urgently and really change how we
operate in order to be able to rise to

1297
01:14:02,400 --> 01:14:04,157
uh to be able to run at machine speed.

1298
01:14:04,240 --> 01:14:07,757
So, there's still more work to do, but I
think that every company needs to really

1299
01:14:07,840 --> 01:14:09,757
stare at this case study and think about

1300
01:14:09,840 --> 01:14:13,357
how do I use this as information for how

1301
01:14:13,440 --> 01:14:17,917
my processes need to change, for how my
operations need to change, and for how I

1302
01:14:18,000 --> 01:14:21,757
build my system and secure myself,
how that needs to evolve in this new era.

1303
01:14:21,840 --> 01:14:26,157
So, Deian, let me ask a quick question,
a follow up on this because um you know,

1304
01:14:26,240 --> 01:14:30,557
you you guys had called us in to help
on this response and that was that's

1305
01:14:30,640 --> 01:14:33,757
publicly known. So, I'll stick to like
the reports that you guys wrote. When

1306
01:14:33,840 --> 01:14:37,277
you read the reports and again,
I commend you and HuggyFace for the

1307
01:14:37,360 --> 01:14:40,077
transparency. The reports are
unbelievable and it's it's a great

1308
01:14:40,160 --> 01:14:45,037
service for the security community. But
when you saw what the agents were doing,

1309
01:14:45,120 --> 01:14:48,637
what was the biggest surprise you took
away? cuz when you read it you're like I

1310
01:14:48,720 --> 01:14:52,957
I you know I don't know what AGI looks
like but man it sure smells like AGI you

1311
01:14:53,040 --> 01:14:57,597
know like what was your biggest
takeaway when you saw how this unfolded

1312
01:14:57,680 --> 01:15:01,437
>> well look [clears throat] I think that
there are some things where because we

1313
01:15:01,520 --> 01:15:05,517
are developing the models we have some
intuitions and some sense of exactly the

1314
01:15:05,600 --> 01:15:09,117
capabilities we expect right so the fact
of these agents being able to coordinate

1315
01:15:09,200 --> 01:15:11,597
with each other that's actually
something we expected right that's

1316
01:15:11,680 --> 01:15:12,717
something that that was really trained

1317
01:15:12,800 --> 01:15:15,597
in I think that there was a surprise in

1318
01:15:15,680 --> 01:15:19,517
some ways with really looking at the um

1319
01:15:19,600 --> 01:15:22,717
I maybe maybe it's again kind of the
level of capability of the fact that

1320
01:15:22,800 --> 01:15:27,997
they were capable of finding these zero
day exploit chains in order to to be

1321
01:15:28,080 --> 01:15:31,757
able to achieve objectives. I think that
that was the surprise in some ways. um

1322
01:15:31,840 --> 01:15:36,237
the fact that I you know to me that when
you have this impossible task that you

1323
01:15:36,320 --> 01:15:39,117
set them that you set in front of them
that they try to find other ways to

1324
01:15:39,200 --> 01:15:43,437
achieve this impossible task like not
surprising right so I think that for us

1325
01:15:43,520 --> 01:15:47,517
a lot of the way that we make
development progress is by finding bugs

1326
01:15:47,600 --> 01:15:51,197
right in these graders and kind of the
the rewards that you give to your models

1327
01:15:51,280 --> 01:15:55,437
very very important um another takeaway
is that the need for train of thought

1328
01:15:55,520 --> 01:15:59,357
monitoring during evaluation and dep
and development and that's something that

1329
01:15:59,440 --> 01:16:03,277
we've deployed and so I think that
for For us, it's kind of a lot of these

1330
01:16:03,360 --> 01:16:06,797
issues and a lot of what we saw, it's
things that we've thought about for many

1331
01:16:06,880 --> 01:16:11,277
years. You can find blog posts from us
in 2017, 2018 talking about very similar

1332
01:16:11,360 --> 01:16:14,877
issues. And I think just really seeing
it in front of you in an implemented

1333
01:16:14,960 --> 01:16:18,797
system and realizing that the stakes,
I mean, they're large, right? We need to

1334
01:16:18,880 --> 01:16:21,917
rise to that occasion. [clears throat]
>> That to me is the biggest takeaway.

1335
01:16:22,000 --> 01:16:24,557
>> Did you do you think
they worked as a collective?

1336
01:16:24,640 --> 01:16:27,597
>> I I mean, for for sure. I and I
think that the the thing that's very

1337
01:16:27,680 --> 01:16:31,677
interesting about the collective
is that you can apply the same thing to

1338
01:16:31,760 --> 01:16:35,357
important problems, right? that we've
used this to solve these unsolved math

1339
01:16:35,440 --> 01:16:39,037
problems which you know maybe maybe now
we all you know kind of just have gotten

1340
01:16:39,120 --> 01:16:42,557
used to it that the hundred-y old math
problem you go on Twitter tomorrow is

1341
01:16:42,640 --> 01:16:46,637
solved again um but it's such a big
deal right and that capability applied to

1342
01:16:46,720 --> 01:16:50,557
biology life sciences drug development
like it's all going to happen and so I

1343
01:16:50,640 --> 01:16:53,677
think that we need to be ready for it
to be applied in cyber security but again

1344
01:16:53,760 --> 01:16:57,677
it's a it's a powerful construct
is something that is going to be about

1345
01:16:57,760 --> 01:17:00,957
accumulating lots of knowledge and being
able to use it in creative ways and I

1346
01:17:01,040 --> 01:17:05,117
think it's so important that defenders
use that creativity uh as as quickly as

1347
01:17:05,200 --> 01:17:07,597
possible. >> Yeah.
Let's move the conversation a

1348
01:17:07,680 --> 01:17:10,797
little bit to choice. It was a it's
something that we all talk about a lot.

1349
01:17:10,880 --> 01:17:15,917
George, how do you see model choice
evolving in a world where we use a lot

1350
01:17:16,000 --> 01:17:20,157
of Frontier Labs, we use some open source.
How does a how how does that all

1351
01:17:20,240 --> 01:17:21,917
set out for for a customer? Talk a

1352
01:17:22,000 --> 01:17:24,717
little bit about model choice. Well, I

1353
01:17:24,800 --> 01:17:29,597
think it's really important and um Greg
commented on a little bit of it which is

1354
01:17:29,680 --> 01:17:34,397
like the power of these of AI shouldn't
be just concentrated in a few and I and

1355
01:17:34,480 --> 01:17:38,317
we certainly see eye to eye on that and I
think when you look at AI and you look

1356
01:17:38,400 --> 01:17:42,557
at again its impact on society its
impact on security I think choice is

1357
01:17:42,640 --> 01:17:46,237
very important and I'm always focused
on this is why we've been you know we've

1358
01:17:46,320 --> 01:17:50,157
we've been in lots of the sort
of frameworks that have gotten signed

1359
01:17:50,240 --> 01:17:53,677
because we believe that a customer
should have choice we should have choice

1360
01:17:53,760 --> 01:17:57,197
And ultimately you have to use the right
tool, the right model, the right harness

1361
01:17:57,280 --> 01:18:00,157
to get the right outcome when you
need it, right? So there's going to be

1362
01:18:00,240 --> 01:18:03,597
certain times where like hey just
everything goes to Frontier. Certain

1363
01:18:03,680 --> 01:18:06,717
times where things can go to our model,
certain times things are, you know,

1364
01:18:06,800 --> 01:18:08,477
somebody else's open source. At the end

1365
01:18:08,560 --> 01:18:09,997
of the day, we're we're here to solve a

1366
01:18:10,080 --> 01:18:13,917
problem and use the best AI. But what

1367
01:18:14,000 --> 01:18:16,237
that means for customers is that I think

1368
01:18:16,320 --> 01:18:20,957
you're in like the one of the best times
in the history of technology to have so

1369
01:18:21,040 --> 01:18:23,037
much choice, have so much innovation.

1370
01:18:23,120 --> 01:18:27,197
You know, again, I look at just even
over the last number of models that you

1371
01:18:27,280 --> 01:18:30,157
guys have come out with. I mean,
we can talk about the industry like it's

1372
01:18:30,240 --> 01:18:34,317
unbelievable what they're capable
of doing. I mean, sincerely, you're like

1373
01:18:34,400 --> 01:18:36,237
this is just like life-changing, right?

1374
01:18:36,320 --> 01:18:39,277
So when when you have capabilities like

1375
01:18:39,360 --> 01:18:43,277
that and you have the amount of money
and resources and effort and the

1376
01:18:43,360 --> 01:18:45,677
training and by the way the history
I mean you guys have been doing this a

1377
01:18:45,760 --> 01:18:50,077
long time that that means a lot right
so for us I think choice is important and

1378
01:18:50,160 --> 01:18:54,877
what we want to do with customers as I
talked about is we want to to enable

1379
01:18:54,960 --> 01:18:58,557
them to have the right choice and the
right models and use whatever it's going

1380
01:18:58,640 --> 01:19:03,517
to take however it's going to get there
to defend and create the asymmetry that

1381
01:19:03,600 --> 01:19:07,437
we talked about with Jensen to stop the
breach. That's our end goal. And it it's

1382
01:19:07,520 --> 01:19:12,077
it's not just with a handful of small
companies. It's not just with crowd.

1383
01:19:12,160 --> 01:19:15,357
It's not just open source. It's how do
you put it all together in a composite

1384
01:19:15,440 --> 01:19:19,437
way to get the best outcome. And I do
think and you know the harness is an

1385
01:19:19,520 --> 01:19:23,677
important element to it. The harness can
work with anything, right? But bringing

1386
01:19:23,760 --> 01:19:27,677
the right models with the right
partners, the right safety, the right

1387
01:19:27,760 --> 01:19:31,277
expertise, and the right sort of data
sovereignty are what customers are

1388
01:19:31,360 --> 01:19:35,197
looking for. And that's why we're
excited to again be working with OpenAI.

1389
01:19:35,280 --> 01:19:38,557
Yep. And we're very aligned on this
perspective and you can look at a lot of

1390
01:19:38,640 --> 01:19:42,157
choices that we make. Even the fact
the fact that the codeex harness is open

1391
01:19:42,240 --> 01:19:45,757
source like that is that is very unique
and that means you can see exactly how

1392
01:19:45,840 --> 01:19:49,117
our models work and we actually train
our models to be useful in any harness,

1393
01:19:49,200 --> 01:19:52,797
right? That's a real intention that we
put into it. And if you look at where we

1394
01:19:52,880 --> 01:19:55,597
are with Astra and we we've been
starting to kind of talk a little bit

1395
01:19:55,680 --> 01:19:58,957
more publicly about what to expect. We
should even have a blog post later today

1396
01:19:59,040 --> 01:20:03,917
where we start showing some of the evals
that we have on on cyber related tasks.

1397
01:20:04,000 --> 01:20:07,117
And it's it's really incredible.
It's really a step up and again this is

1398
01:20:07,200 --> 01:20:10,237
driving a lot of urgency for us to really
think about how do we bring these

1399
01:20:10,320 --> 01:20:13,597
capabilities to the world in a broad
way that really help people defend and

1400
01:20:13,680 --> 01:20:17,437
prepare for what's coming. Um that's
what we view our sort of responsibility

1401
01:20:17,520 --> 01:20:20,717
and how we want to show up. Um but even
for example one thing that's interesting

1402
01:20:20,800 --> 01:20:25,757
is uh you know hugging face they talked
about how hey we were we had to use an

1403
01:20:25,840 --> 01:20:28,717
open model in order to be able to do
our incident response. We're unable to use

1404
01:20:28,800 --> 01:20:32,157
closed models. They didn't actually
try our models and uh that's

1405
01:20:32,240 --> 01:20:35,117
>> that's an interesting tidbit. Yeah,
I think people people have not heard that

1406
01:20:35,200 --> 01:20:37,917
and actually our our belief is that if
they had used our models, they would

1407
01:20:38,000 --> 01:20:41,677
have been able to do the exact same
sort of uh you know log analysis and and

1408
01:20:41,760 --> 01:20:45,037
response that they did with with the open
ones. And so again, we really have

1409
01:20:45,120 --> 01:20:49,277
a lot of technical work, a lot
of intention, a lot of sort of very

1410
01:20:49,360 --> 01:20:53,277
concrete proof points of us really
trying to lean in to say let's help

1411
01:20:53,360 --> 01:20:57,677
defenders uh wherever they are. Do you
think that the in and I'm saying this

1412
01:20:57,760 --> 01:20:59,517
broadly maybe for Frontier AI, do you

1413
01:20:59,600 --> 01:21:02,717
think the guard railing now is is is too

1414
01:21:02,800 --> 01:21:07,597
tilted into a way that you know someone
tried a different model, they they

1415
01:21:07,680 --> 01:21:10,477
didn't get a result, maybe they would
have gotten a result from you, but is it

1416
01:21:10,560 --> 01:21:13,277
tilted in a way that has gone too far?

1417
01:21:13,360 --> 01:21:16,957
Um because you know if you look at the
the older models of course you know it's

1418
01:21:17,040 --> 01:21:20,237
a lot easier to kind of get the results
that you wanted but how do you how do

1419
01:21:20,320 --> 01:21:24,477
you view that? I think this is a very
nuanced topic and I I really struggle

1420
01:21:24,560 --> 01:21:27,037
with it. We really struggle with we
spent a lot of time debating this

1421
01:21:27,120 --> 01:21:29,757
internally. But if you look at what's
happened really over the course of the

1422
01:21:29,840 --> 01:21:33,037
year, beginning of the year, we
announced our our trusted access program

1423
01:21:33,120 --> 01:21:37,437
because we really saw that cyber
capabilities were going to take off. And

1424
01:21:37,520 --> 01:21:40,797
I think that as you start to have things
like Mythos and starting to see that you

1425
01:21:40,880 --> 01:21:44,477
can get all these zero days kind of the
reaction was let's close it up, right?

1426
01:21:44,560 --> 01:21:48,637
Let's really have very small number of
trusted defenders who get access. And on

1427
01:21:48,720 --> 01:21:52,397
the one hand, I think it was helpful and
that there were some small number of of

1428
01:21:52,480 --> 01:21:55,837
projects and companies that were able to
to be able to start getting ahead of the

1429
01:21:55,920 --> 01:21:58,957
curve, but there's small number, right?

1430
01:21:59,040 --> 01:22:02,317
That's the problem is it's really
a collective action problem, right? It's

1431
01:22:02,400 --> 01:22:06,717
every defender needs these tools
and there's limited time and that the now

1432
01:22:06,800 --> 01:22:10,557
we're looking at models that are coming
out that are very cyber capable, GLM53,

1433
01:22:10,640 --> 01:22:15,517
things like that. And so to me that I
think there's something about this this

1434
01:22:15,600 --> 01:22:19,197
world where it's like there's
a temptation to close up in a way that is

1435
01:22:19,280 --> 01:22:22,557
actually locally optimal, right? Because
you can say, well, we can cut off these

1436
01:22:22,640 --> 01:22:25,757
thread actors, but it's actually maybe
not globally optimal because you realize

1437
01:22:25,840 --> 01:22:27,437
that you leave so many people behind.

1438
01:22:27,520 --> 01:22:31,117
And so that's actually a tension.
It's like a fundamental tension. And I

1439
01:22:31,200 --> 01:22:34,717
actually do see a lot of energy changing
now where people are saying, okay, these

1440
01:22:34,800 --> 01:22:38,237
capabilities are coming from all over
the world, right? So many people are

1441
01:22:38,320 --> 01:22:43,437
developing these models now. And if you
waste the opportunity, if you do not use

1442
01:22:43,520 --> 01:22:46,877
the fact that you have these incredibly
cyber capable models to give you a

1443
01:22:46,960 --> 01:22:49,997
window into the future that give
the defender this window, if you let that

1444
01:22:50,080 --> 01:22:53,677
window close, if you let it shut, if
you let it go behind, it is not going to

1445
01:22:53,760 --> 01:22:57,197
come back, right? It is it is an
opportunity that we have now. And that

1446
01:22:57,280 --> 01:23:01,517
it is it is like an absolute amazing
fact that we have it. And so I think

1447
01:23:01,600 --> 01:23:04,397
that there is something there that we
need to approach a little differently.

1448
01:23:04,480 --> 01:23:07,757
Um, we've been thoughtfully trying to
to engage with that. But again, it's not

1449
01:23:07,840 --> 01:23:11,837
just about us. It's really about
this collective action in the community.

1450
01:23:11,920 --> 01:23:15,757
>> Greg, for you, you know, as you look
at the future of Open AI and it's a really

1451
01:23:15,840 --> 01:23:20,077
exciting future, how do you view
the nature of your broader market and and

1452
01:23:20,160 --> 01:23:23,677
your innovation so that more and more
of the world's best work and the best

1453
01:23:23,760 --> 01:23:26,717
innovation happens using OpenAI models?

1454
01:23:26,800 --> 01:23:29,757
>> Well, I think that our goal is train

1455
01:23:29,840 --> 01:23:33,757
amazing models and help people do
amazing things with them, right? In some

1456
01:23:33,840 --> 01:23:37,757
ways, we're a very simple business,
right? just we take compute, we train

1457
01:23:37,840 --> 01:23:41,357
models, we deploy models. Um, in some
way it's a very complicated business,

1458
01:23:41,440 --> 01:23:44,877
right? Because bringing these models to
the world in every single domain, making

1459
01:23:44,960 --> 01:23:47,997
sure that they're able to perform well,
working very closely with customers,

1460
01:23:48,080 --> 01:23:51,197
being able to get feedback, um, building
enterprise trust, building the right

1461
01:23:51,280 --> 01:23:55,117
guardrails, observability, all those
things. Um, again, we can't do it alone.

1462
01:23:55,200 --> 01:23:56,797
So, we rely on working with partners.

1463
01:23:56,880 --> 01:23:59,757
One of the reasons that we love working
with CrowdStrike, right? and being able

1464
01:23:59,840 --> 01:24:03,597
to really get that these models so
quickly into the hands of so many

1465
01:24:03,680 --> 01:24:07,757
defenders is very very top of mind.
And so a lot of what we focus on is just

1466
01:24:07,840 --> 01:24:11,837
really trying to be peering into
the future, building the most capable

1467
01:24:11,920 --> 01:24:14,957
systems in a way that is deeply
responsible, right? And I think that

1468
01:24:15,040 --> 01:24:18,877
we're really seeing how much safety,
security, and alignment are now becoming

1469
01:24:18,960 --> 01:24:21,757
the bottleneck, right? The hardest part
of what we have to do. And so a lot of

1470
01:24:21,840 --> 01:24:25,517
our effort is shifting from just the okay,
let's improve the score on this

1471
01:24:25,600 --> 01:24:29,277
benchmark to let's make sure we're able
to deliver this all safely to the world

1472
01:24:29,360 --> 01:24:32,557
um in in conjunction with our
partners and and the ecosystem.

1473
01:24:32,640 --> 01:24:35,277
>> I think it's exactly why we're working
together, especially on initiatives like

1474
01:24:35,360 --> 01:24:38,957
securing codecs that gets announced so
that more of our customers and more of

1475
01:24:39,040 --> 01:24:42,157
the world can adopt AI faster and do so

1476
01:24:42,240 --> 01:24:46,317
securely. George, round out the
conversation today. How is this a 1 plus

1477
01:24:46,400 --> 01:24:51,037
1 equals 3? And what are you excited
about in terms of our future with

1478
01:24:51,120 --> 01:24:53,437
working with OpenAI? >> Well,
I think there's a few things. I

1479
01:24:53,520 --> 01:24:57,037
mean, if you start with codecs and and
sort of the agents themselves, being

1480
01:24:57,120 --> 01:25:01,117
able to have a great relationship where
we can work together to help instrument

1481
01:25:01,200 --> 01:25:03,917
what's happening. You know, you talked
about these sort of traces and opening

1482
01:25:04,000 --> 01:25:07,517
up codecs and understanding what's
happening from an aid perspective.

1483
01:25:07,600 --> 01:25:11,117
That's incredibly important and things
are changing very quickly. So you have

1484
01:25:11,200 --> 01:25:15,117
to have a partner that you can work with
to get the telemetry and things that you

1485
01:25:15,200 --> 01:25:19,517
want out of what's happening with the
agent itself. That's one. Number two is

1486
01:25:19,600 --> 01:25:23,597
the models and being able to make those
available to our customers. So part of

1487
01:25:23,680 --> 01:25:28,077
what we talked about is bringing
the open A models into our platform, but

1488
01:25:28,160 --> 01:25:31,757
exposing them in a way where our
customers can harness the power of what

1489
01:25:31,840 --> 01:25:35,677
Open AI has built on the data that we've
already collected. You know, we've been

1490
01:25:35,760 --> 01:25:38,957
doing this for [clears throat] a long
time. We know the threat actors. We know

1491
01:25:39,040 --> 01:25:43,437
how it works. We we have we know the
customer's environment and I think if we

1492
01:25:43,520 --> 01:25:47,757
can share that information in a secure
way which is very important to our

1493
01:25:47,840 --> 01:25:51,757
customers but still give them access to
the power of your models that's a that's

1494
01:25:51,840 --> 01:25:56,877
a 1 plus 1 equals 4. >> Perfect.
Well on that 1 plus 1 equals

1495
01:25:56,960 --> 01:25:58,077
four. We like that kind of math.

1496
01:25:58,160 --> 01:26:00,477
>> Yes. We love it. >> Yeah.
That's a perfect outcome of a

1497
01:26:00,560 --> 01:26:03,757
model. So thank you so much for your
time today Greg. It's been a pleasure

1498
01:26:03,840 --> 01:26:05,437
having you. >> Thank you for having me.

1499
01:26:05,520 --> 01:26:07,357
>> Thank you.
Thank you so much. All right.

1500
01:26:07,440 --> 01:26:11,757
talk to now.

1501
01:26:11,840 --> 01:26:15,517
>> Well, that rounds [music] out
our session. We've gone through a lot of

1502
01:26:15,600 --> 01:26:20,157
layers of the AI stack from the chip to
the to the end user computing device all

1503
01:26:20,240 --> 01:26:24,077
the way to how we all interact with AI
all the time, the models. Thank you so

1504
01:26:24,160 --> 01:26:26,477
much. You got to hear from Jensen, from

1505
01:26:26,560 --> 01:26:29,117
Nvidia, from Intel, from OpenAI, all

1506
01:26:29,200 --> 01:26:33,917
here on the stage at Falcon. Go out
into the hub, learn about securing AI. It

1507
01:26:34,000 --> 01:26:40,600
takes a crowd to do it. Thank you for
being part of our crowd. Have a great day.