1
00:00:00,109 --> 00:00:04,097
So Anthropic just got hacked. Again. And this time not only did the attackers

2
00:00:04,097 --> 00:00:07,687
get access to proprietary information about models, people's salaries,

3
00:00:07,687 --> 00:00:09,157
but we also know who did it.

4
00:00:09,157 --> 00:00:13,627
A one month old startup that hammered their systems with hundreds of agents

5
00:00:13,627 --> 00:00:16,666
24-7 until they found a secret passage in.

6
00:00:16,910 --> 00:00:20,087
But what should concern everyone is how they got access.

7
00:00:20,087 --> 00:00:25,127
It was via a third party platform that Anthropic had zero control over and using

8
00:00:25,127 --> 00:00:29,617
AI models that existed three months ago which means that probably anyone could

9
00:00:29,617 --> 00:00:30,887
have pulled off the same attack.

10
00:00:30,887 --> 00:00:33,967
You see the reason why I want to talk about this today isn't just because this

11
00:00:33,967 --> 00:00:37,883
is another major breach of an AI lab that is concerning in itself

12
00:00:38,168 --> 00:00:41,817
but it's because it's part of a recent growing trend I'm observing where AI

13
00:00:41,817 --> 00:00:44,469
is just capable of attacking and hacking into

14
00:00:44,649 --> 00:00:49,007
any software that exists for a few dollars and this has massive ramifications

15
00:00:49,007 --> 00:00:52,277
around how we protect our bank accounts, our medical records,

16
00:00:52,277 --> 00:00:55,327
our personal information, and even the nuclear codes.

17
00:00:55,327 --> 00:00:59,267
The point is, this affects anything and everything that is software-based,

18
00:00:59,267 --> 00:01:00,840
and it's time to take this seriously.

19
00:01:01,345 --> 00:01:04,926
I have many thoughts, so let's get into it on today's episode brought to you by Qualcomm.

20
00:01:08,881 --> 00:01:12,766
So there's this startup called Cathedral, which is a one-month-old startup run

21
00:01:12,766 --> 00:01:18,946
by ex-Doge employees, but it's co-founded by an ex-Pentagon head of data.

22
00:01:18,946 --> 00:01:24,219
Now, as you know, the history between the Pentagon and Anthropic has been less than ideal.

23
00:01:24,416 --> 00:01:27,736
Pentagon currently has blacklisted Claude and Anthropic after they failed to

24
00:01:27,736 --> 00:01:31,306
comply with a bunch of other initiatives, which I'm not going to get into right now.

25
00:01:31,306 --> 00:01:37,776
But the point is, a one-month-old startup, and they saw that Anthropic was having a public bug bounty.

26
00:01:37,776 --> 00:01:42,396
So typically companies do this when they want ethical hackers or white hat hackers

27
00:01:42,396 --> 00:01:46,466
to stress or test the boundaries of their own systems.

28
00:01:46,466 --> 00:01:50,956
The point is, if they are able to find a bug or a flaw, they can disclose it

29
00:01:50,956 --> 00:01:53,580
to Anthropic or any other company who's running a similar test,

30
00:01:53,824 --> 00:01:58,352
and they'll be able to patch and fix it before a malicious actor is able to do so.

31
00:01:58,613 --> 00:02:03,536
So this startup cathedral took part in this, and they started off by doing the obvious thing.

32
00:02:03,826 --> 00:02:07,306
They tried to go in through the front door. So they tried to attack all of Anthropic

33
00:02:07,306 --> 00:02:11,346
systems, and they did so for several days, but came up with nothing.

34
00:02:11,346 --> 00:02:14,403
They weren't able to penetrate the system. So that's all well and good.

35
00:02:14,902 --> 00:02:19,216
But then they had a bright idea. They thought, hmm, what if we went through

36
00:02:19,216 --> 00:02:22,716
a third-party admin system or a third-party software system?

37
00:02:22,716 --> 00:02:25,350
I wonder how easy it would be to do so.

38
00:02:25,547 --> 00:02:29,696
And so they did so. They started off with targeting a sales and admin platform

39
00:02:29,696 --> 00:02:33,767
known as Salesforce, used by millions of people around the world,

40
00:02:34,062 --> 00:02:39,486
and within 12 hours, they were able to get access to Anthropix Slack.

41
00:02:39,486 --> 00:02:43,866
Now, for context here, Slack might not seem like an important thing.

42
00:02:43,866 --> 00:02:47,036
It's where you kind of exchange messages with your colleagues,

43
00:02:47,036 --> 00:02:48,692
and maybe you share some files here and there.

44
00:02:49,248 --> 00:02:54,222
Anthropic and companies like OpenAI keep everything on Slack.

45
00:02:54,472 --> 00:02:58,432
This is something that the Anthropic researcher that quit a few weeks ago,

46
00:02:58,432 --> 00:03:02,752
Jacob Coxon, that heralded the warnings of all these labs basically

47
00:03:03,132 --> 00:03:06,532
building AI without any necessarily controls and systems, and it's all going

48
00:03:06,532 --> 00:03:09,552
to end up in a cataclysmic end of humanity.

49
00:03:09,552 --> 00:03:14,062
That guy basically said that these companies store everything on Slack.

50
00:03:14,062 --> 00:03:17,392
I'm talking model information about unreleased models.

51
00:03:17,692 --> 00:03:21,552
I'm talking about strategies about how they beat their competitors and pre-release

52
00:03:21,552 --> 00:03:25,475
items for new products. I'm talking personal information like people's salaries.

53
00:03:25,788 --> 00:03:31,593
Anything that is identifiable and valuable is mentioned, discussed, and stored on Slack.

54
00:03:32,000 --> 00:03:36,002
And that's what these attackers got access to via a third-party system.

55
00:03:36,002 --> 00:03:40,472
So it's a very serious breach. Now, Anthropic's response to this was we became

56
00:03:40,472 --> 00:03:44,122
aware of one flaw that they exploited, but we weren't aware of another one,

57
00:03:44,122 --> 00:03:45,542
which they didn't disclose.

58
00:03:45,542 --> 00:03:48,322
And they're trying to claim that a cathedral was basically unethical.

59
00:03:48,322 --> 00:03:52,636
But the point is, these guys got access to very proprietary information.

60
00:03:52,886 --> 00:03:56,342
Now, Anthropic also included in their response that they only got access to

61
00:03:56,342 --> 00:04:02,615
things like people's salaries, some model discussions, but not actually model weights, not actually

62
00:04:02,929 --> 00:04:07,322
proprietary research that could inform an attacker how to build a much better

63
00:04:07,322 --> 00:04:09,662
model or sabotage their own systems.

64
00:04:09,924 --> 00:04:12,994
But personally, I think that's besides the point. They got into the system,

65
00:04:13,150 --> 00:04:16,029
and let's say they got access to other types of Slack channels,

66
00:04:16,244 --> 00:04:19,355
would they have got access to this proprietary information? I don't know.

67
00:04:19,634 --> 00:04:24,032
Now, it's worth spending a moment emphasizing the political landscape around

68
00:04:24,032 --> 00:04:28,982
this whole thing, because as I mentioned earlier, the co-founder and CEO of

69
00:04:28,982 --> 00:04:32,655
Cathedral is the ex-head of data at the Pentagon.

70
00:04:33,015 --> 00:04:36,152
And the Anthropic and the Pentagon have a very storied history.

71
00:04:36,152 --> 00:04:39,441
It started about, I don't know, six months ago, I think in January of,

72
00:04:39,807 --> 00:04:44,729
when the whole war between the US and Iran was going down, and it was revealed

73
00:04:44,729 --> 00:04:50,399
that the Pentagon had secretly used Claude to strategize a certain attack.

74
00:04:50,399 --> 00:04:55,009
Actually, it wasn't about the war. It was about the Maduro capture specifically, right?

75
00:04:55,429 --> 00:05:00,752
So they used Claude to plan and execute the successful capture of Maduro.

76
00:05:01,030 --> 00:05:04,599
And Anthropic had some choice thoughts on that. They said, you shouldn't be

77
00:05:04,599 --> 00:05:05,969
using our models to be able to do that.

78
00:05:05,969 --> 00:05:10,419
We're going to restrict you now, and any use by the Pentagon of Claude to perform

79
00:05:10,419 --> 00:05:14,899
or execute missions should go through us. You need to call us up and we'll give

80
00:05:14,899 --> 00:05:18,009
you the access if we think it's right or wrong.

81
00:05:18,369 --> 00:05:21,729
And obviously the government and the Pentagon thought that this was a massive

82
00:05:21,729 --> 00:05:27,117
breach of trust and could endanger the US as a national security issue if, let's say,

83
00:05:27,430 --> 00:05:31,859
foreign adversaries like China who are known to get access to blacklisted accounts

84
00:05:31,859 --> 00:05:34,989
could do the same thing and use it against them, right?

85
00:05:34,989 --> 00:05:39,000
So I could see valid propositions on both ends, right? But the point is,

86
00:05:39,180 --> 00:05:40,428
they were at odds with each other.

87
00:05:40,637 --> 00:05:45,249
And the result was that Pentagon and the government ended up blacklisting Anthropic

88
00:05:45,249 --> 00:05:46,935
and blacklisting their Claude models.

89
00:05:47,214 --> 00:05:51,749
To this day, it is currently still blacklisted. We've come full circle here,

90
00:05:51,749 --> 00:05:55,199
where an external startup that is only a month old, that is backed by the likes

91
00:05:55,199 --> 00:05:59,569
of Andreessen Horowitz, I believe Mark Andreessen is actually on the board of directors

92
00:05:59,949 --> 00:06:05,436
of this company, was complicit in attacking the company that is currently blacklisted.

93
00:06:05,720 --> 00:06:09,749
Now, of course, within days of this being announced or within hours of this

94
00:06:09,749 --> 00:06:11,965
being announced, the Pentagon reached out to this company and was like.

95
00:06:12,475 --> 00:06:16,518
Hey, we want to offer you a government contract. We want you to work with us

96
00:06:16,518 --> 00:06:19,197
and we want you to help us figure out where

97
00:06:19,522 --> 00:06:23,538
possible attacks may come from, from adversaries and help us harden our own

98
00:06:23,538 --> 00:06:28,079
systems and help us figure out how we prevent attacks like this happening in the future.

99
00:06:28,322 --> 00:06:31,318
So on the one hand, you have the government getting involved.

100
00:06:31,318 --> 00:06:36,258
Now, remember, none of these labs are currently nationalized and maybe this

101
00:06:36,258 --> 00:06:37,796
is a step towards that happening.

102
00:06:37,999 --> 00:06:39,328
The fact that this is being flagged

103
00:06:39,328 --> 00:06:43,305
by both the government and a frontier research lab is a pretty big deal.

104
00:06:43,711 --> 00:06:46,701
And so you have the Pentagon saying, okay, we need to figure this out.

105
00:06:46,945 --> 00:06:51,348
And then you have Anthropic responding and their response has sort of been muted.

106
00:06:51,348 --> 00:06:55,418
I haven't seen really much mention of this on the news channels or mainstream

107
00:06:55,418 --> 00:06:57,104
media channels. But the point is.

108
00:06:57,720 --> 00:07:02,294
They are trying to play it down a little bit. And listen, if the attackers truly

109
00:07:02,294 --> 00:07:07,014
did just get access to some personal information, such as salaries and personal

110
00:07:07,014 --> 00:07:12,462
data, maybe it's being dramatized a bit. And I want to refrain from doing that.

111
00:07:12,741 --> 00:07:18,004
But I'm at odds because what if they did get access to proprietary channels?

112
00:07:18,004 --> 00:07:22,514
And I guess I don't know the extent of which Anthropic specifically keeps their

113
00:07:22,514 --> 00:07:27,004
proprietary information on Slack. All I know is that the rumors are they do.

114
00:07:27,004 --> 00:07:30,255
And so do a lot of other companies. I'm sure it's not just OpenAI or Anthropic.

115
00:07:30,423 --> 00:07:33,574
It could have been any other company, but it's quite prominent in this case

116
00:07:33,574 --> 00:07:36,431
because Anthropic is one of the leading AI research labs out there,

117
00:07:36,808 --> 00:07:40,547
building these models that could potentially cause a lot of harm.

118
00:07:40,831 --> 00:07:45,164
And it confirms exactly what Jacob Cox and the ex-OpenAI and Anthropic alignment

119
00:07:45,164 --> 00:07:49,964
researcher said a few weeks back where he said, listen, these companies are

120
00:07:49,964 --> 00:07:55,334
recklessly racing towards superintelligence at the cost of their own security.

121
00:07:55,334 --> 00:07:59,864
He literally said, they're gambling with our lives. And back then it was just

122
00:07:59,864 --> 00:08:01,381
a headline. It was just hearsay.

123
00:08:01,729 --> 00:08:06,154
And now it's being proven. And many respected voices in the field are raising

124
00:08:06,154 --> 00:08:10,474
some very valid concerns. Will Depew here goes, Anthropic is massively underplaying

125
00:08:10,474 --> 00:08:14,344
the recent security breach. I'm hearing Cathedral achieved full administrator

126
00:08:14,344 --> 00:08:16,136
access of the Slack workspace.

127
00:08:16,432 --> 00:08:21,157
To those unfamiliar, Anthropic and OpenAI keep literally everything in Slack.

128
00:08:21,535 --> 00:08:24,604
We're critical to national security. We must beat China. And yes,

129
00:08:24,604 --> 00:08:27,903
we keep all our crown jewels in Salesforce Slack Messenger.

130
00:08:28,234 --> 00:08:30,306
Yes, we have admin API tokens everywhere.

131
00:08:30,648 --> 00:08:37,484
Why do you ask? Anthropic and OpenAI have openly stated multiple times that

132
00:08:37,484 --> 00:08:40,854
AI models are capable of very dangerous acts. Okay, fair enough.

133
00:08:41,132 --> 00:08:45,464
They also say that if we don't control these systems, if we don't pace the frontier,

134
00:08:45,464 --> 00:08:48,194
remember, there was a letter signed by hundreds and hundreds of very prominent

135
00:08:48,194 --> 00:08:51,419
individuals asking to slow down AI progress and research.

136
00:08:51,668 --> 00:08:55,614
If we don't do this, we could end up with a catastrophic event where someone

137
00:08:55,614 --> 00:08:59,694
builds a bioweapon or there's a massive cybersecurity attack which steals most

138
00:08:59,694 --> 00:09:01,461
of our money from our bank accounts, right?

139
00:09:01,757 --> 00:09:05,814
And last week, we saw Black October unfold over 48 hours where there was a series

140
00:09:05,814 --> 00:09:09,614
of these potential types of attacks. So we're in the mix here and they're making

141
00:09:09,614 --> 00:09:14,289
a very valid point. So then why do you keep a bunch of your important stuff on Slack?

142
00:09:14,446 --> 00:09:17,924
It just doesn't make any sense. He then goes on to make a point that he doesn't

143
00:09:17,924 --> 00:09:21,818
necessarily think that it's the security team's fault at Anthropic,

144
00:09:22,120 --> 00:09:25,344
but more so the higher level C-suite exec, the CEO

145
00:09:25,664 --> 00:09:32,552
and founder's prerogative to build AI models as fast as they can to out-compete their competitors.

146
00:09:32,790 --> 00:09:37,274
And they're doing this at the cost of making sure that their models are one,

147
00:09:37,274 --> 00:09:39,100
aligned, and two, secure.

148
00:09:39,460 --> 00:09:43,164
He's saying that OpenAI and Anthropic and any other model lab are doing these things.

149
00:09:43,164 --> 00:09:46,298
And if that is the truth, then we are going to end up in

150
00:09:46,518 --> 00:09:50,864
crappy situations like this where they're building amazing models that on one

151
00:09:50,864 --> 00:09:52,765
hand are working towards solving cancer

152
00:09:52,985 --> 00:09:56,734
and on the other hand are hacking all of our accounts now there was also a statement

153
00:09:56,734 --> 00:10:00,164
from anthropic's head of red teaming red teaming is basically the team that

154
00:10:00,164 --> 00:10:01,745
stress tests the systems they

155
00:10:01,995 --> 00:10:05,157
are like an internal team that does the bug bounty thing that i described earlier

156
00:10:05,476 --> 00:10:06,986
and his response was listen.

157
00:10:07,536 --> 00:10:11,705
Things aren't as severe as people are making it out to be. They weren't able

158
00:10:11,705 --> 00:10:13,939
to perform the attack in under 12 hours.

159
00:10:14,125 --> 00:10:16,946
We were aware of the initial exploit.

160
00:10:17,184 --> 00:10:21,705
We weren't aware of the second exploit. And he goes on to make a number of excuses,

161
00:10:21,705 --> 00:10:27,424
basically saying this isn't that big of a deal, which I don't necessarily buy, to be honest.

162
00:10:27,714 --> 00:10:32,225
But what concerns me more is what he says near the end of this. He goes, so what to do?

163
00:10:32,585 --> 00:10:37,125
You probably want to red team yourself all the time. current models are very

164
00:10:37,125 --> 00:10:42,405
creative i also think we should redesign software systems so that you're fundamentally

165
00:10:42,405 --> 00:10:45,367
less exposed he then goes in another tweet,

166
00:10:45,826 --> 00:10:51,255
but man do i agree with your overall message extremely extremely important basically

167
00:10:51,255 --> 00:10:55,695
need to reinvent security every year at this point this is coming from one of

168
00:10:55,695 --> 00:10:58,678
the heads of security at one of the leading labs,

169
00:10:59,357 --> 00:11:02,020
that are building all these AI models in the first place.

170
00:11:02,270 --> 00:11:07,181
And the reason why this is concerning is he's basically seeing no way to protect

171
00:11:07,355 --> 00:11:11,396
current systems, current software as it exists today.

172
00:11:11,738 --> 00:11:16,715
He's saying likely anything and everything is flawed, including the bedrock

173
00:11:16,715 --> 00:11:21,455
on which the internet is built on, all the way scaled up to front-end applications

174
00:11:21,455 --> 00:11:23,598
that anyone can make in VibeCode today.

175
00:11:24,033 --> 00:11:27,035
And he's saying we're going to have to rebuild the security infrastructure and

176
00:11:27,035 --> 00:11:28,921
systems around these different things.

177
00:11:29,466 --> 00:11:33,071
One thing immediately pops to mind this is going to be incredibly costly.

178
00:11:33,762 --> 00:11:36,609
This is going to take an incredible amount of time. And three,

179
00:11:36,609 --> 00:11:42,789
and most importantly, this is going to happen at so many different rates of time.

180
00:11:42,789 --> 00:11:47,599
What I mean by that is companies aren't going to suddenly upgrade their systems

181
00:11:47,599 --> 00:11:52,479
to this brand new security parameter every single year, as he's hoping will

182
00:11:52,479 --> 00:11:54,039
happen over the next year.

183
00:11:54,550 --> 00:11:58,059
Companies are going to progress. They're very archaic. They're very bloated.

184
00:11:58,059 --> 00:11:59,168
They're very bureaucratic.

185
00:11:59,546 --> 00:12:01,409
You need policy sign-offs. You need

186
00:12:01,409 --> 00:12:04,393
to kind of convince so many different stakeholders for this to happen.

187
00:12:04,515 --> 00:12:08,649
What that means is there's going to be a long period of time,

188
00:12:08,649 --> 00:12:10,889
which is likely going to be a couple of years at this point,

189
00:12:11,185 --> 00:12:16,559
where anyone and everyone that gets access to cheap old generation models will

190
00:12:16,559 --> 00:12:19,399
be able to probably perform similar attacks like this.

191
00:12:19,683 --> 00:12:22,859
You don't have to take my word for it. You just need to look over our shoulder

192
00:12:22,859 --> 00:12:27,019
at last week, where we saw one of the biggest cybersecurity incidents that happened

193
00:12:27,019 --> 00:12:31,432
over 48 hours that marked the entire month a new nickname, Black October.

194
00:12:31,700 --> 00:12:37,319
The very presumption that we need to now rebuild our entire software stack and

195
00:12:37,319 --> 00:12:42,199
do so every year is bizarre for me to understand, but is, I guess,

196
00:12:42,199 --> 00:12:44,599
necessary to understand this new world.

197
00:12:44,599 --> 00:12:48,309
The immediate analogy that pops to my mind is those world-building games.

198
00:12:48,309 --> 00:12:50,429
You know, the ones that you see on your mobile phone being advertised to you

199
00:12:50,429 --> 00:12:52,934
or like Clash of Clans or whatever that might be, Civilization,

200
00:12:53,253 --> 00:12:57,219
where you have to control multiple things and build a world you have to build

201
00:12:57,219 --> 00:12:59,599
a fence a defense system to prevent

202
00:12:59,599 --> 00:13:02,743
people attacking you but by the time someone's attacking on one side

203
00:13:03,051 --> 00:13:05,509
someone's attacking on the other side and so you have to go there and deploy

204
00:13:05,509 --> 00:13:10,249
resources and it's this whole constant 24 7 thing that's what the new world

205
00:13:10,249 --> 00:13:12,890
is going to look like from a cyber security perspective so

206
00:13:13,047 --> 00:13:17,729
the only way to combat this in my opinion is to spin up other ai agents to defend

207
00:13:17,729 --> 00:13:20,147
against this and that's basically the point he's making he's saying.

208
00:13:20,652 --> 00:13:25,712
Hey, you're probably going to need to pay for access to AI models to red team

209
00:13:25,712 --> 00:13:29,052
your entire system before attackers find exploits themselves.

210
00:13:29,052 --> 00:13:33,492
Now, there's obviously a self-serving reason why he might say this because,

211
00:13:33,492 --> 00:13:37,492
hey, like we have the best models at Anthropic, use our models or use any open

212
00:13:37,492 --> 00:13:40,186
source models or you open air will say the same thing about their models.

213
00:13:40,429 --> 00:13:44,342
So I see that angle, but I fail to see how he's wrong.

214
00:13:44,342 --> 00:13:48,942
We are working ourselves towards a world where at any given moment,

215
00:13:48,942 --> 00:13:50,960
it doesn't matter if you're asleep, it doesn't matter if your

216
00:13:51,232 --> 00:13:56,252
security team is the best of the best, humans have flaws, human handwritten

217
00:13:56,252 --> 00:14:00,606
code, which is 99% of all the code out there also has flaws,

218
00:14:01,042 --> 00:14:03,654
will mean that you will get eventually attacked.

219
00:14:03,869 --> 00:14:07,926
And so you need to have a perpetual defense system that is not only defending,

220
00:14:08,066 --> 00:14:11,961
but is being proactive and plugging any holes in your systems.

221
00:14:12,222 --> 00:14:16,262
This doesn't just apply to Anthropic and their Slack exploit.

222
00:14:16,593 --> 00:14:21,892
This applies to you and i who are consumers of everyday products any kind of

223
00:14:21,892 --> 00:14:26,072
system that you use for leisure or for professional work is liable to this but

224
00:14:26,072 --> 00:14:27,882
it's also applicable to you as a

225
00:14:28,182 --> 00:14:31,642
business owner or a founder who might be building systems or building a new

226
00:14:31,642 --> 00:14:35,542
startup this is something that could potentially happen to you as well so i

227
00:14:35,542 --> 00:14:36,342
guess the reason why this

228
00:14:36,682 --> 00:14:38,716
concerns me and why i wanted to make this episode is that

229
00:14:39,064 --> 00:14:41,852
this has such a large attack surface

230
00:14:41,852 --> 00:14:46,382
area and the cost of these things are going to ultimately be driven.

231
00:14:46,862 --> 00:14:51,122
Super super low a point that was made around this whole thing is that they

232
00:14:51,442 --> 00:14:56,112
likely used older generation models to do this and if i had to guess these older

233
00:14:56,112 --> 00:15:01,877
generation models are at the same capability level of open source models and this is something that.

234
00:15:02,463 --> 00:15:06,735
People like dario mode and sam altman have warned about happening very soon

235
00:15:06,852 --> 00:15:10,913
okay so i want to get into a few final thoughts on today's episode brought to you by qualcomm.

236
00:15:15,001 --> 00:15:19,169
The first is, this was kind of out of Anthropic's control.

237
00:15:19,413 --> 00:15:22,367
They got access via a third-party platform and software. And listen,

238
00:15:22,553 --> 00:15:25,899
I'm not saying Anthropic isn't liable for this. They need to work with these

239
00:15:25,899 --> 00:15:28,739
third parties and companies to make sure that this doesn't happen again.

240
00:15:28,739 --> 00:15:31,650
And from what I hear, they've been able to patch this up.

241
00:15:31,917 --> 00:15:37,049
But the point being is, security is no longer a single siloed thing.

242
00:15:37,049 --> 00:15:40,959
It's no longer just about you and what you can do, how much money you spend

243
00:15:40,959 --> 00:15:44,038
and defense systems that you, walls that you build around your own product.

244
00:15:44,363 --> 00:15:50,597
It's about the products that you use to help make your product and service actually function and work.

245
00:15:50,905 --> 00:15:53,488
And if you look at the majority of software companies out there today,

246
00:15:53,813 --> 00:15:56,079
this is pretty much all of them.

247
00:15:56,079 --> 00:16:01,459
The world internet economy runs on companies being able to work with other companies,

248
00:16:01,459 --> 00:16:07,509
automations, bots, systems that work 24-7 because they're supported by every

249
00:16:07,509 --> 00:16:09,638
other single software system that you need.

250
00:16:10,009 --> 00:16:14,409
And this was the exact flaw that we saw last week when we found out that AI

251
00:16:14,409 --> 00:16:19,119
agents were apparently getting into people's bank accounts here in the US and

252
00:16:19,119 --> 00:16:22,829
moving money around, changing transactions, getting access to personal information.

253
00:16:22,829 --> 00:16:26,569
They were able to get access via a third party for credential information.

254
00:16:26,569 --> 00:16:29,249
This is the same thing. So it's parallels of the same thing,

255
00:16:29,249 --> 00:16:32,179
but happening in two different industries, which means that it's liable to pretty

256
00:16:32,179 --> 00:16:36,029
much anything and everything. Imagine you're walking into the office of the

257
00:16:36,029 --> 00:16:37,581
company that you want to hack into.

258
00:16:37,877 --> 00:16:41,959
You steal the cleaning company, who is a third party's access key.

259
00:16:42,365 --> 00:16:47,169
You get access to the, I don't know, the HR office, and you're able to go through

260
00:16:47,169 --> 00:16:50,889
the filing cabinet and see everyone's salaries or look into the model room and

261
00:16:50,889 --> 00:16:52,239
see what they're cooking up.

262
00:16:52,239 --> 00:16:58,032
Now, the second point I want to mention is just the 24-7 nature of this entire attack.

263
00:16:58,368 --> 00:17:01,063
Humans sleep. Humans are.

264
00:17:01,915 --> 00:17:05,659
Imperfect, and they make mistakes sometimes, even the best of the best.

265
00:17:05,839 --> 00:17:10,390
And also humans, experts specifically, are scarce. It's hard to find.

266
00:17:10,704 --> 00:17:16,180
And so to have AI agents that are not just at the level of the expert humans,

267
00:17:16,180 --> 00:17:20,430
but are starting to beat them, this becomes a problem, especially if it's 24-7.

268
00:17:20,430 --> 00:17:23,886
And the idea here is, it's not just that it's happening all the time,

269
00:17:24,164 --> 00:17:29,012
it means that they're stress testing any and every possible avenue to get in,

270
00:17:29,203 --> 00:17:33,197
which means that the long tail of scenarios, which humans

271
00:17:33,505 --> 00:17:36,190
don't typically spend a lot of time trying to figure out because they're like,

272
00:17:36,190 --> 00:17:38,790
oh, it'll never happen, will actually happen.

273
00:17:38,790 --> 00:17:43,054
And that's a new security vector that most people haven't really considered today.

274
00:17:43,385 --> 00:17:48,545
Third thing is the attacker in this instance is very closely affiliated to a.

275
00:17:48,958 --> 00:17:53,863
Huge government org, in this case being the Pentagon. They signed a new contract with the Pentagon.

276
00:17:54,188 --> 00:17:56,220
They are, I think, maybe have some

277
00:17:56,220 --> 00:17:59,854
very strong affiliation given that the co-founder was an ex-worker there.

278
00:18:00,155 --> 00:18:06,850
And so we are weirdly unofficially one step closer towards nationalizing these labs.

279
00:18:06,850 --> 00:18:10,950
Because if the threat was big enough, let's say it was an adversarial actor,

280
00:18:10,950 --> 00:18:14,230
not a white hat hacker in this instance, and they got access to proprietary

281
00:18:14,230 --> 00:18:18,430
model information that could then be used against the US, you could easily see

282
00:18:18,430 --> 00:18:21,412
the part that leads towards, hey, we need to nationalize this lab.

283
00:18:21,592 --> 00:18:25,940
Otherwise, we're going to end up in a huge pot of hot water and the final thing

284
00:18:25,940 --> 00:18:30,440
i'll say is if you're china and you're reading the headlines today or of this

285
00:18:30,440 --> 00:18:32,958
week of last week seeing all these security flaws.

286
00:18:33,742 --> 00:18:38,212
Maybe you're smiling because not only are you aware that these things are capable,

287
00:18:38,659 --> 00:18:43,332
but you're aware that these things are capable from the average older model.

288
00:18:43,541 --> 00:18:46,963
Remember, you don't need frontier models. You don't need to spend thousands

289
00:18:46,963 --> 00:18:48,382
and thousands of dollars anymore.

290
00:18:48,725 --> 00:18:52,123
You can just spend a couple hundred bucks and get access to some of the most

291
00:18:52,123 --> 00:18:55,812
important and crucial systems that run the world and the future world.

292
00:18:56,387 --> 00:18:58,860
Anyway, those are my thoughts. I hope you enjoyed today's episode.

293
00:18:59,092 --> 00:18:59,943
Thank you so much for listening.

294
00:18:59,943 --> 00:19:05,472
If you're listening to this on YouTube, there are many new subscribers welcome it's great to have you,

295
00:19:06,103 --> 00:19:08,743
if you aren't subscribed or if you haven't turned on notifications please do

296
00:19:08,743 --> 00:19:10,093
so it helps us out massively

297
00:19:10,395 --> 00:19:14,443
also we are on spotify and apple go check us out give us a rating leave us a

298
00:19:14,443 --> 00:19:18,004
comment leave us comments everywhere by the way dm me on x my socials are linked

299
00:19:18,259 --> 00:19:23,576
below and also if you think of a friend colleague or anyone that might find this interesting

300
00:19:23,832 --> 00:19:26,653
please share it with them it helps us out massively and yeah

301
00:19:26,932 --> 00:19:29,613
i think that's it i'll see you on the next one thank you for listening to today's

302
00:19:29,613 --> 00:19:31,230
episode brought to you by qualcomm.