1
00:00:00,000 --> 00:00:05,910
Right,

2
00:00:02,830 --> 00:00:05,910
[Music]

3
00:00:10,719 --> 00:00:17,520
welcome back everyone. It is my great

4
00:00:13,599 --> 00:00:20,320
pleasure to introduce Tim and Nikki who

5
00:00:17,520 --> 00:00:22,640
will be talking to us about myths about

6
00:00:20,320 --> 00:00:24,240
open source security that developers

7
00:00:22,640 --> 00:00:28,359
believe.

8
00:00:24,240 --> 00:00:28,359
For now, a warm welcome.

9
00:00:32,320 --> 00:00:37,280
Good day everyone. Welcome to our talk

10
00:00:34,559 --> 00:00:39,520
about a few myths, just a few that open

11
00:00:37,280 --> 00:00:41,760
source developers believe brackets for

12
00:00:39,520 --> 00:00:42,399
now. I'm Nikki. I'm a product manager at

13
00:00:41,760 --> 00:00:43,920
Google.

14
00:00:42,399 --> 00:00:44,320
I'm Tim. I'm a software engineer at

15
00:00:43,920 --> 00:00:46,239
Google.

16
00:00:44,320 --> 00:00:50,879
And we work on open source dependency

17
00:00:46,239 --> 00:00:52,800
analysis in the coolest team, uh, Ghost,

18
00:00:50,879 --> 00:00:54,239
the Google open source security team.

19
00:00:52,800 --> 00:00:56,160
And yes, we have stickers. Come get us

20
00:00:54,239 --> 00:00:58,320
later. Um, so we're going to run through

21
00:00:56,160 --> 00:01:01,359
a bunch of myths. uh that we have

22
00:00:58,320 --> 00:01:02,879
ourselves previously held uh and that

23
00:01:01,359 --> 00:01:05,280
when we're chatting to folks about uh

24
00:01:02,879 --> 00:01:07,680
about open source um that they might

25
00:01:05,280 --> 00:01:10,479
think that's right where the

26
00:01:07,680 --> 00:01:13,439
ghostbusters ghost myth bust. Yeah, I'm

27
00:01:10,479 --> 00:01:16,640
sorry. Okay. Um so myth number one,

28
00:01:13,439 --> 00:01:18,720
installing Python packages is easy and

29
00:01:16,640 --> 00:01:21,360
safe. All right, let's start with the

30
00:01:18,720 --> 00:01:23,280
easy X I You're in the right room. Uh

31
00:01:21,360 --> 00:01:25,920
the easy part. Okay, look. Yeah, that

32
00:01:23,280 --> 00:01:27,920
Okay, that I think easy is correct. Nice

33
00:01:25,920 --> 00:01:30,560
work, US. We're halfway there, at least

34
00:01:27,920 --> 00:01:33,759
in Python. Um, but actually, it's pretty

35
00:01:30,560 --> 00:01:36,240
true of most ecosystems. We're good

36
00:01:33,759 --> 00:01:40,079
here. All right. How about the safe part

37
00:01:36,240 --> 00:01:42,240
of our hypothesis? Like so many things

38
00:01:40,079 --> 00:01:46,159
uh with dependencies, the answer is, of

39
00:01:42,240 --> 00:01:48,720
course, it depends. Um, it depends on

40
00:01:46,159 --> 00:01:51,520
what that thing that we just installed

41
00:01:48,720 --> 00:01:54,320
was and what it contained. uh what you

42
00:01:51,520 --> 00:01:56,560
ended up installing as a free fun bonus

43
00:01:54,320 --> 00:01:59,280
thrown in those transitive dependency

44
00:01:56,560 --> 00:02:02,159
steak knives.

45
00:01:59,280 --> 00:02:04,399
We interrupt this talk for a brief

46
00:02:02,159 --> 00:02:06,960
public safety reminder. It's a beautiful

47
00:02:04,399 --> 00:02:08,720
autumn day. You're in the office. You've

48
00:02:06,960 --> 00:02:11,760
just returned from a meeting when you

49
00:02:08,720 --> 00:02:14,800
notice a USB stick abandoned on the

50
00:02:11,760 --> 00:02:16,640
floor. You look at it. You look at your

51
00:02:14,800 --> 00:02:18,959
laptop.

52
00:02:16,640 --> 00:02:22,640
You leave it untouched. some mysteries

53
00:02:18,959 --> 00:02:25,200
you think weren't meant to be answered.

54
00:02:22,640 --> 00:02:27,280
Okay, pop quiz. Um, what's the

55
00:02:25,200 --> 00:02:29,120
difference between these two things? One

56
00:02:27,280 --> 00:02:31,520
is plugging in a mystery USB stick and

57
00:02:29,120 --> 00:02:33,840
the other is running npm install on a

58
00:02:31,520 --> 00:02:37,920
package. Well, they do have a few

59
00:02:33,840 --> 00:02:41,040
similarities, but actually npm install

60
00:02:37,920 --> 00:02:43,519
gets run potentially as root in your

61
00:02:41,040 --> 00:02:47,360
data center whereas a thumb drive is

62
00:02:43,519 --> 00:02:50,160
just on one developer machine.

63
00:02:47,360 --> 00:02:53,040
Okay, so there have been a few bigname

64
00:02:50,160 --> 00:02:56,319
supply chain attacks over the last few

65
00:02:53,040 --> 00:03:00,720
years uh and lots more that you might

66
00:02:56,319 --> 00:03:03,680
not have heard about. Um in fact just so

67
00:03:00,720 --> 00:03:06,239
so in a few days ago I felt compelled to

68
00:03:03,680 --> 00:03:10,159
update these slides uh with another

69
00:03:06,239 --> 00:03:13,040
headline and another clarification.

70
00:03:10,159 --> 00:03:16,319
Um so

71
00:03:13,040 --> 00:03:18,080
so how risky is it? Um, you can think of

72
00:03:16,319 --> 00:03:20,239
risk in two different dimensions. There

73
00:03:18,080 --> 00:03:22,800
there's the risk of the package itself,

74
00:03:20,239 --> 00:03:24,800
but also how many packages there are.

75
00:03:22,800 --> 00:03:29,120
So, first off, how many? Well, there's

76
00:03:24,800 --> 00:03:31,599
somewhere between zero and a lot. Uh,

77
00:03:29,120 --> 00:03:34,720
how many free as in puppies, transitive

78
00:03:31,599 --> 00:03:36,080
dependencies are we getting? Well, I

79
00:03:34,720 --> 00:03:37,920
don't know. It's hard to say, but on

80
00:03:36,080 --> 00:03:40,560
average, well, there are lots of

81
00:03:37,920 --> 00:03:43,519
different ways to measure average. And

82
00:03:40,560 --> 00:03:46,480
honestly, all of the ways that we tried

83
00:03:43,519 --> 00:03:49,200
for average ended up with the answer, it

84
00:03:46,480 --> 00:03:53,040
depends, but kind of a lot in some

85
00:03:49,200 --> 00:03:55,200
ecosystems more than others. Um, okay.

86
00:03:53,040 --> 00:03:59,760
So, how risky is that dependency? And

87
00:03:55,200 --> 00:04:01,519
again, it depends. One aspect of risk, I

88
00:03:59,760 --> 00:04:04,480
should take this moment to say most of

89
00:04:01,519 --> 00:04:05,840
these images are from emoji kitchen. I I

90
00:04:04,480 --> 00:04:07,200
I was going to mention that at the end

91
00:04:05,840 --> 00:04:09,680
of the talk because now we just lost our

92
00:04:07,200 --> 00:04:12,080
audience. That's fine. But um how how

93
00:04:09,680 --> 00:04:14,560
risky is this package? What is the

94
00:04:12,080 --> 00:04:16,239
impact? If the risk occurred, what's the

95
00:04:14,560 --> 00:04:17,919
probability of it happening? And what's

96
00:04:16,239 --> 00:04:20,239
the impact of it happening? And there

97
00:04:17,919 --> 00:04:22,960
are lots of dimensions to this risk. Is

98
00:04:20,239 --> 00:04:24,720
it malware? Sure. But also, do the

99
00:04:22,960 --> 00:04:26,160
maintainers follow best security

100
00:04:24,720 --> 00:04:27,680
practice or are they burnt out and

101
00:04:26,160 --> 00:04:30,320
susceptible to sock puppet social

102
00:04:27,680 --> 00:04:32,560
engineering attacks? If there are bugs,

103
00:04:30,320 --> 00:04:35,280
are they or security concerns, are they

104
00:04:32,560 --> 00:04:37,759
apt and ideally remediated or is the

105
00:04:35,280 --> 00:04:40,080
package abandoned? Is the package just

106
00:04:37,759 --> 00:04:42,160
new? No one really knows. Maybe it's a

107
00:04:40,080 --> 00:04:43,919
typo squatting attempt. Is it a package

108
00:04:42,160 --> 00:04:46,080
with a solid history, but also with

109
00:04:43,919 --> 00:04:48,880
leaked or fished credentials? The

110
00:04:46,080 --> 00:04:51,120
package is available right now, but will

111
00:04:48,880 --> 00:04:52,960
it be next week? And is there an

112
00:04:51,120 --> 00:04:55,360
attestation available so you can be sure

113
00:04:52,960 --> 00:04:58,240
it's the same package that you think

114
00:04:55,360 --> 00:05:01,440
that it really should be. Some of these

115
00:04:58,240 --> 00:05:04,320
risks change based on the ecosystem.

116
00:05:01,440 --> 00:05:06,240
Packages. Package managers here uh can

117
00:05:04,320 --> 00:05:07,840
be a great help. Shout out to Pippi for

118
00:05:06,240 --> 00:05:11,039
being absolutely on the ball with some

119
00:05:07,840 --> 00:05:14,240
great security features and some others

120
00:05:11,039 --> 00:05:18,479
you can protect against yourself when

121
00:05:14,240 --> 00:05:21,120
those risks matter to you somewhat.

122
00:05:18,479 --> 00:05:23,120
Okay, myth number two. There's only one

123
00:05:21,120 --> 00:05:26,160
dependency graph. You may remember me

124
00:05:23,120 --> 00:05:28,240
from such pyons as 2024. uh we

125
00:05:26,160 --> 00:05:31,840
calculated approximately

126
00:05:28,240 --> 00:05:34,800
this number on the order of 10 to the 81

127
00:05:31,840 --> 00:05:37,840
valid dependency graphs for a single

128
00:05:34,800 --> 00:05:39,199
dependency for D3. Um okay so that's

129
00:05:37,840 --> 00:05:41,759
almost the number of atoms in the

130
00:05:39,199 --> 00:05:43,280
universe. Uh but good news that number's

131
00:05:41,759 --> 00:05:45,440
gone up since then. That was a whole

132
00:05:43,280 --> 00:05:47,840
year ago.

133
00:05:45,440 --> 00:05:49,759
But why? Well there are just more

134
00:05:47,840 --> 00:05:52,479
packages available now which generally

135
00:05:49,759 --> 00:05:54,560
means a lot more options when satisfying

136
00:05:52,479 --> 00:05:56,160
requirements. And what do I mean by

137
00:05:54,560 --> 00:05:57,919
satisfying requirements? Well, you have

138
00:05:56,160 --> 00:05:59,840
a project. Say you want to use Express.

139
00:05:57,919 --> 00:06:01,600
It's a minimalist web framework. It's a

140
00:05:59,840 --> 00:06:03,600
node. But actually, so we've got these

141
00:06:01,600 --> 00:06:05,600
requirements on that package. Maybe you

142
00:06:03,600 --> 00:06:07,120
need it to be version five or above for

143
00:06:05,600 --> 00:06:08,560
those shiny new features. So it turns

144
00:06:07,120 --> 00:06:10,400
out there are three versions that could

145
00:06:08,560 --> 00:06:13,280
satisfy that constraint. Here's our

146
00:06:10,400 --> 00:06:15,600
situation. Your package depends on

147
00:06:13,280 --> 00:06:17,039
Express. It has three valid versions to

148
00:06:15,600 --> 00:06:18,720
pick from. That's not so bad. Three.

149
00:06:17,039 --> 00:06:20,319
Great. I can count that high. But

150
00:06:18,720 --> 00:06:21,919
Express has its own dependencies. For

151
00:06:20,319 --> 00:06:23,280
example, it depends on body paser. And

152
00:06:21,919 --> 00:06:25,039
there are currently around five versions

153
00:06:23,280 --> 00:06:26,240
that would satisfy that constraint. And

154
00:06:25,039 --> 00:06:28,000
that's a very small part of the

155
00:06:26,240 --> 00:06:30,160
dependency tree. So, so there are more

156
00:06:28,000 --> 00:06:31,759
layers and each dependency on each of

157
00:06:30,160 --> 00:06:33,520
those layers on each of those versions

158
00:06:31,759 --> 00:06:36,880
of each of those layers and it stops

159
00:06:33,520 --> 00:06:38,160
becoming a tree. And very quickly, yeah,

160
00:06:36,880 --> 00:06:40,800
it's a mess. And if you were going to

161
00:06:38,160 --> 00:06:43,759
run npm install say a month later, it's

162
00:06:40,800 --> 00:06:45,600
entirely reasonable and expected that a

163
00:06:43,759 --> 00:06:48,720
huge part of this graph is thrown into

164
00:06:45,600 --> 00:06:51,360
chaos and a wildly different one emerges

165
00:06:48,720 --> 00:06:52,800
because which version actually ends up

166
00:06:51,360 --> 00:06:54,479
in your dependency graph depends on a

167
00:06:52,800 --> 00:06:55,600
lot of different things. The build tools

168
00:06:54,479 --> 00:06:57,840
compute the dependency graphs by

169
00:06:55,600 --> 00:06:59,759
evaluating constraints created by

170
00:06:57,840 --> 00:07:02,000
dependency requirements. I prefer

171
00:06:59,759 --> 00:07:03,680
version three or greater. Environmental

172
00:07:02,000 --> 00:07:06,240
constraints. This is a build for

173
00:07:03,680 --> 00:07:08,400
Windows. what kind of build it is. Build

174
00:07:06,240 --> 00:07:10,960
the tests as well as the software and

175
00:07:08,400 --> 00:07:13,440
perhaps build specific considerations.

176
00:07:10,960 --> 00:07:16,000
The binary must be minified. So a new

177
00:07:13,440 --> 00:07:18,160
version showing up can change the graph.

178
00:07:16,000 --> 00:07:19,919
A version being deleted that changes the

179
00:07:18,160 --> 00:07:22,080
graph. Different tooling using UV

180
00:07:19,919 --> 00:07:24,560
instead of pip or yarn instead of npm or

181
00:07:22,080 --> 00:07:27,440
even just a different version of npm.

182
00:07:24,560 --> 00:07:29,440
One quirk I quite like at least with npm

183
00:07:27,440 --> 00:07:33,440
if you rename a package you're

184
00:07:29,440 --> 00:07:36,160
installing think import fu as blah or

185
00:07:33,440 --> 00:07:37,840
arvark that can change your dependency

186
00:07:36,160 --> 00:07:39,840
graph because some implementations of

187
00:07:37,840 --> 00:07:42,639
npm resolve the dependencies in lexical

188
00:07:39,840 --> 00:07:45,840
graphical order which can cause

189
00:07:42,639 --> 00:07:47,680
requirement yeah

190
00:07:45,840 --> 00:07:49,680
requirement conflicts to resolve

191
00:07:47,680 --> 00:07:51,919
differently depending on which one gets

192
00:07:49,680 --> 00:07:53,759
evaluated first.

193
00:07:51,919 --> 00:07:57,199
So

194
00:07:53,759 --> 00:08:00,080
where does that leave us? There is never

195
00:07:57,199 --> 00:08:02,479
the dependency graph. There is only ever

196
00:08:00,080 --> 00:08:05,120
a dependency graph on your particular

197
00:08:02,479 --> 00:08:08,240
machine and at your specific point in

198
00:08:05,120 --> 00:08:12,639
time using this specific package manager

199
00:08:08,240 --> 00:08:14,800
and so on. Okay. Myth number three,

200
00:08:12,639 --> 00:08:16,800
names are unique. Okay. So you've

201
00:08:14,800 --> 00:08:18,479
decided on a dependency, you know you

202
00:08:16,800 --> 00:08:20,400
want to use it. Great. And and you know

203
00:08:18,479 --> 00:08:22,879
its name and what the version that you

204
00:08:20,400 --> 00:08:25,599
want to use is. Go for it. Except what

205
00:08:22,879 --> 00:08:27,280
was it called again? Packages by any

206
00:08:25,599 --> 00:08:29,919
other name might smell as sweet, but

207
00:08:27,280 --> 00:08:34,000
name confusion can certainly be a

208
00:08:29,919 --> 00:08:35,919
thorny. I'm so sorry. Okay. So, um,

209
00:08:34,000 --> 00:08:38,159
names are not unique. Not across

210
00:08:35,919 --> 00:08:39,599
ecosystems. Fair enough. Maybe each

211
00:08:38,159 --> 00:08:41,200
ecosystem really does need a Babel

212
00:08:39,599 --> 00:08:43,599
package, but also names might not be

213
00:08:41,200 --> 00:08:45,040
unique across registries and namespaces.

214
00:08:43,599 --> 00:08:47,040
There could be quite a few reasons for

215
00:08:45,040 --> 00:08:50,000
this, and some of those reasons are more

216
00:08:47,040 --> 00:08:51,839
problematic than others. Name squatting

217
00:08:50,000 --> 00:08:54,320
is essentially when someone comes along

218
00:08:51,839 --> 00:08:56,480
and registers a bunch of common names to

219
00:08:54,320 --> 00:08:59,760
sell later. It's really common with

220
00:08:56,480 --> 00:09:02,240
domains, uh, social media and also

221
00:08:59,760 --> 00:09:03,680
happens with package registries. Uh, I'm

222
00:09:02,240 --> 00:09:05,200
not sure how lucrative it is in case

223
00:09:03,680 --> 00:09:06,880
you're looking for a career change, but

224
00:09:05,200 --> 00:09:09,440
um, you can think of name squatting as a

225
00:09:06,880 --> 00:09:11,600
specific type of typo squatting. The

226
00:09:09,440 --> 00:09:13,519
broader concept here is that some minor

227
00:09:11,600 --> 00:09:15,040
mistake when you're typing the package

228
00:09:13,519 --> 00:09:17,200
means that the user doesn't end up with

229
00:09:15,040 --> 00:09:20,080
a package that they expect. they end up

230
00:09:17,200 --> 00:09:22,320
with your package instead. That mistake

231
00:09:20,080 --> 00:09:24,399
might be a genuine typo, typos targeting

232
00:09:22,320 --> 00:09:26,240
a specific package, or it might be

233
00:09:24,399 --> 00:09:28,560
reusing the names of packages that are

234
00:09:26,240 --> 00:09:31,200
popular in other ecosystems, or it might

235
00:09:28,560 --> 00:09:33,519
just sound right, like these VS Code

236
00:09:31,200 --> 00:09:35,200
extensions that were all uploaded by the

237
00:09:33,519 --> 00:09:37,760
same malicious account and found in

238
00:09:35,200 --> 00:09:41,200
April this year. And of course, that's

239
00:09:37,760 --> 00:09:45,519
the challenge. What is legit? What seems

240
00:09:41,200 --> 00:09:46,959
legit? And what's safe? Two of these are

241
00:09:45,519 --> 00:09:50,160
legitimate offiscation tools that

242
00:09:46,959 --> 00:09:51,839
developers could use. But which two?

243
00:09:50,160 --> 00:09:54,560
It's these ones. But you have no way of

244
00:09:51,839 --> 00:09:58,320
knowing unless you actually know. Um, so

245
00:09:54,560 --> 00:10:00,640
what's a dev to do? Well, uh, maybe Tim

246
00:09:58,320 --> 00:10:03,040
can find some answers.

247
00:10:00,640 --> 00:10:05,120
Maybe.

248
00:10:03,040 --> 00:10:08,480
Be prepared for speaking rate to tone

249
00:10:05,120 --> 00:10:10,000
down to like five.

250
00:10:08,480 --> 00:10:12,240
Yeah. Just just Yeah. If you were

251
00:10:10,000 --> 00:10:15,040
listening to this at 2x uh Well, at

252
00:10:12,240 --> 00:10:16,959
four. Yeah. No. Yeah.

253
00:10:15,040 --> 00:10:19,519
So myth number four, popular things are

254
00:10:16,959 --> 00:10:21,680
more secure.

255
00:10:19,519 --> 00:10:24,399
So you probably know a core tenant of

256
00:10:21,680 --> 00:10:27,680
open source is Linus' law where given

257
00:10:24,399 --> 00:10:30,320
enough eyeballs or bugs are shallow,

258
00:10:27,680 --> 00:10:33,839
which is to roughly say that popular

259
00:10:30,320 --> 00:10:35,920
open source things are safer, right?

260
00:10:33,839 --> 00:10:38,399
Unexpected uh expectedly it's more

261
00:10:35,920 --> 00:10:41,200
complicated than that. Firstly, popular

262
00:10:38,399 --> 00:10:43,440
things tend to be more complex.

263
00:10:41,200 --> 00:10:45,760
Most often, if you're writing Python

264
00:10:43,440 --> 00:10:47,680
packages that are popular, you're likely

265
00:10:45,760 --> 00:10:49,839
to be writing solutions that cater to

266
00:10:47,680 --> 00:10:53,120
thousands or if not millions of distinct

267
00:10:49,839 --> 00:10:54,959
users and use cases.

268
00:10:53,120 --> 00:10:56,880
And it happens that popular things tend

269
00:10:54,959 --> 00:10:59,839
to have more features, more nuanced

270
00:10:56,880 --> 00:11:01,839
optimizations, etc.

271
00:10:59,839 --> 00:11:04,240
Of course, complexity tends to mean

272
00:11:01,839 --> 00:11:06,079
greater attack surface. And this is sort

273
00:11:04,240 --> 00:11:08,640
of the inherent trade-off of using

274
00:11:06,079 --> 00:11:10,640
popular open source software. you get

275
00:11:08,640 --> 00:11:13,440
comprehensive battle tested solutions,

276
00:11:10,640 --> 00:11:15,440
but they're often much more complex and

277
00:11:13,440 --> 00:11:17,519
therefore greater tendency for potential

278
00:11:15,440 --> 00:11:19,120
vulnerabilities.

279
00:11:17,519 --> 00:11:20,640
Secondly,

280
00:11:19,120 --> 00:11:23,519
popular things are often the only thing

281
00:11:20,640 --> 00:11:25,519
that people even care to attack.

282
00:11:23,519 --> 00:11:27,120
So, a tangent.

283
00:11:25,519 --> 00:11:29,360
When I was growing up, I thought there

284
00:11:27,120 --> 00:11:32,720
were only ever two operating systems in

285
00:11:29,360 --> 00:11:34,720
this world, Windows, Mac OS. And then

286
00:11:32,720 --> 00:11:37,680
some day in high school a tragedy

287
00:11:34,720 --> 00:11:39,839
occurred and I heard about Linux for the

288
00:11:37,680 --> 00:11:41,920
first time.

289
00:11:39,839 --> 00:11:44,720
Uh and I genuinely heard this being

290
00:11:41,920 --> 00:11:47,519
lorded as like one of the reasons why a

291
00:11:44,720 --> 00:11:50,959
person ought to consider using Linux

292
00:11:47,519 --> 00:11:53,040
because supposedly it's virus free. And

293
00:11:50,959 --> 00:11:55,200
I recall this was genuinely like a

294
00:11:53,040 --> 00:11:58,240
frequently quoted benefit of using Linux

295
00:11:55,200 --> 00:11:59,600
for personal computing.

296
00:11:58,240 --> 00:12:01,760
As we all know, the statement is

297
00:11:59,600 --> 00:12:04,399
obviously false, but there is some truth

298
00:12:01,760 --> 00:12:06,720
around this idea because Linux

299
00:12:04,399 --> 00:12:09,279
unfortunately never really took off as a

300
00:12:06,720 --> 00:12:11,600
popular operating system for desktop

301
00:12:09,279 --> 00:12:14,079
computing. Uh, in terms of market share,

302
00:12:11,600 --> 00:12:18,399
we're currently sitting at around 2%

303
00:12:14,079 --> 00:12:19,920
roughly. This is 2024 data.

304
00:12:18,399 --> 00:12:21,360
So, if you were an attacker and you

305
00:12:19,920 --> 00:12:23,120
wanted to distribute malicious

306
00:12:21,360 --> 00:12:24,720
executables,

307
00:12:23,120 --> 00:12:26,800
you would probably not be targeting

308
00:12:24,720 --> 00:12:29,920
Linux. And that that happens to be a

309
00:12:26,800 --> 00:12:32,959
huge reason why Linux was just the more

310
00:12:29,920 --> 00:12:35,760
secure choice.

311
00:12:32,959 --> 00:12:38,320
Likewise here, I would not need to tell

312
00:12:35,760 --> 00:12:41,200
you which package would be more likely

313
00:12:38,320 --> 00:12:43,440
uh to be valuable to attack.

314
00:12:41,200 --> 00:12:44,880
Pyramid here, by the way, is another web

315
00:12:43,440 --> 00:12:47,200
server framework that you probably

316
00:12:44,880 --> 00:12:49,440
haven't heard about. It does fewer

317
00:12:47,200 --> 00:12:52,160
things and is expectedly less popular

318
00:12:49,440 --> 00:12:53,519
than Django. And so it just happens

319
00:12:52,160 --> 00:12:56,880
fewer people are going to care to

320
00:12:53,519 --> 00:12:56,880
compromise pyramids.

321
00:12:57,200 --> 00:13:01,360
Okay, here's some really recent news.

322
00:12:59,440 --> 00:13:03,920
This landed like earlier, literally just

323
00:13:01,360 --> 00:13:06,079
this week. Uh, what you're looking at

324
00:13:03,920 --> 00:13:07,600
here is a sample of a few JavaScript

325
00:13:06,079 --> 00:13:11,120
packages that have been released

326
00:13:07,600 --> 00:13:13,200
recently with a malicious change.

327
00:13:11,120 --> 00:13:15,839
That change, by the way, is going to

328
00:13:13,200 --> 00:13:17,680
drain your crypto wallet.

329
00:13:15,839 --> 00:13:20,079
Uh, and here are the weekly downloads

330
00:13:17,680 --> 00:13:23,040
for some of those packages. We're

331
00:13:20,079 --> 00:13:25,120
talking in the ballpark of uh billions

332
00:13:23,040 --> 00:13:29,160
of downloads. And this is in just a

333
00:13:25,120 --> 00:13:29,160
single oneweek period.

334
00:13:29,839 --> 00:13:34,320
So in addition to you as a developer

335
00:13:31,600 --> 00:13:35,920
needing to be security aware, the

336
00:13:34,320 --> 00:13:38,240
ecosystem as a whole has a

337
00:13:35,920 --> 00:13:39,839
responsibility to adopt strong security

338
00:13:38,240 --> 00:13:42,160
practices.

339
00:13:39,839 --> 00:13:44,320
Uh at at a minimum think things like

340
00:13:42,160 --> 00:13:47,440
multiffactor authentication,

341
00:13:44,320 --> 00:13:49,920
package signing, etc.

342
00:13:47,440 --> 00:13:51,839
uh sort of amusingly between ecosystems

343
00:13:49,920 --> 00:13:53,440
there's actually a surprising spectrum

344
00:13:51,839 --> 00:13:56,480
as to how strong their security

345
00:13:53,440 --> 00:13:59,519
practices are. For instance, consider

346
00:13:56,480 --> 00:14:02,639
this. How long is it going to take for

347
00:13:59,519 --> 00:14:05,199
90% of reported malicious packages to be

348
00:14:02,639 --> 00:14:08,320
taken down? In other words, like you

349
00:14:05,199 --> 00:14:09,920
report 100 packages as being malicious.

350
00:14:08,320 --> 00:14:13,279
How long does it take for 90 of them

351
00:14:09,920 --> 00:14:15,839
roughly to be taken down?

352
00:14:13,279 --> 00:14:17,839
For Pippi, the answer is roughly 24 to

353
00:14:15,839 --> 00:14:19,920
36 hours.

354
00:14:17,839 --> 00:14:22,480
And for npm, that figure is like an

355
00:14:19,920 --> 00:14:25,120
order of magnitude higher.

356
00:14:22,480 --> 00:14:27,120
Yeah. And so this is where we give a

357
00:14:25,120 --> 00:14:28,959
special shout out to Pipi for actually

358
00:14:27,120 --> 00:14:32,120
being an outlier with how responsive

359
00:14:28,959 --> 00:14:32,120
they are.

360
00:14:32,560 --> 00:14:38,000
Uh but even though like Pippi tends to

361
00:14:35,040 --> 00:14:40,480
act particularly fast here, things get

362
00:14:38,000 --> 00:14:42,480
downloaded at such a huge scale that we

363
00:14:40,480 --> 00:14:44,959
are still talking about like hundreds of

364
00:14:42,480 --> 00:14:48,639
thousands of downloads for one package

365
00:14:44,959 --> 00:14:51,680
in a single 24-hour period. And that's

366
00:14:48,639 --> 00:14:54,680
still enough to potentially produce huge

367
00:14:51,680 --> 00:14:54,680
impact.

368
00:14:54,959 --> 00:14:59,680
So the lesson here is

369
00:14:57,760 --> 00:15:01,519
the takeaway here is not to stay away

370
00:14:59,680 --> 00:15:04,320
from popular things. Often you don't

371
00:15:01,519 --> 00:15:06,079
have a choice. Um for instance, you

372
00:15:04,320 --> 00:15:10,320
should not be swearing off Python today

373
00:15:06,079 --> 00:15:12,880
and rewriting all your things in Cobalt.

374
00:15:10,320 --> 00:15:15,360
The lesson is just to not blindly use

375
00:15:12,880 --> 00:15:18,240
popular technology expecting that it'll

376
00:15:15,360 --> 00:15:20,480
be more secure.

377
00:15:18,240 --> 00:15:22,160
There are many good good angles to this.

378
00:15:20,480 --> 00:15:23,760
However, like the good thing is if you

379
00:15:22,160 --> 00:15:25,760
do opt to use something secure uh

380
00:15:23,760 --> 00:15:28,000
popular, like generally speaking, if

381
00:15:25,760 --> 00:15:29,440
something bad were to happen, uh you

382
00:15:28,000 --> 00:15:32,800
know, many people care and they're going

383
00:15:29,440 --> 00:15:35,279
to act quickly and remediations tend to

384
00:15:32,800 --> 00:15:37,040
come about quicker.

385
00:15:35,279 --> 00:15:39,760
And as you saw just then with these

386
00:15:37,040 --> 00:15:41,760
malicious JavaScript packages, uh people

387
00:15:39,760 --> 00:15:43,120
noticed it really quickly, they reported

388
00:15:41,760 --> 00:15:46,639
it quickly, and now it's being

389
00:15:43,120 --> 00:15:50,079
remediated quickly.

390
00:15:46,639 --> 00:15:51,759
If you're curious, uh, for actual stats,

391
00:15:50,079 --> 00:15:53,360
it could have been much worse, but, uh,

392
00:15:51,759 --> 00:15:55,199
it turns out those malicious packages

393
00:15:53,360 --> 00:15:57,600
were downloaded roughly on the order of

394
00:15:55,199 --> 00:15:59,279
hundreds of thousands of times before

395
00:15:57,600 --> 00:16:01,360
takedown.

396
00:15:59,279 --> 00:16:03,040
Uh, how disastrous did this end up

397
00:16:01,360 --> 00:16:05,279
being?

398
00:16:03,040 --> 00:16:07,759
Remember, these malicious packages would

399
00:16:05,279 --> 00:16:11,199
drain your crypto wallet. So, how much

400
00:16:07,759 --> 00:16:13,279
did they actually managed to steal?

401
00:16:11,199 --> 00:16:15,360
Uh, current analysis shows that the

402
00:16:13,279 --> 00:16:17,440
attacker actually made off with a

403
00:16:15,360 --> 00:16:20,680
jaw-dropping

404
00:16:17,440 --> 00:16:20,680
$5 US.

405
00:16:21,040 --> 00:16:26,720
Uh, plus plus a bit more in um various

406
00:16:23,360 --> 00:16:29,519
NFTts and meme coins and whatnot.

407
00:16:26,720 --> 00:16:31,360
Anyway, like we were very lucky here,

408
00:16:29,519 --> 00:16:34,160
but this is just another reason why

409
00:16:31,360 --> 00:16:38,600
popular packages get targeted. Like you

410
00:16:34,160 --> 00:16:38,600
need huge volume for huge impact.

411
00:16:39,199 --> 00:16:44,240
onto a different myth. Open source is

412
00:16:41,440 --> 00:16:46,079
immutable.

413
00:16:44,240 --> 00:16:48,000
The packages you install today are going

414
00:16:46,079 --> 00:16:51,120
to be the same as the packages you

415
00:16:48,000 --> 00:16:53,839
install tomorrow, right?

416
00:16:51,120 --> 00:16:55,920
Not necessarily, because unlike other

417
00:16:53,839 --> 00:16:57,839
package managers in other ecosystems,

418
00:16:55,920 --> 00:17:00,560
pip doesn't actually make integrity

419
00:16:57,839 --> 00:17:02,720
checks by default. And by integrity

420
00:17:00,560 --> 00:17:04,640
checks, we just mean pip does not take

421
00:17:02,720 --> 00:17:07,120
the hash of the packages you download

422
00:17:04,640 --> 00:17:10,160
and uh check that in future times you

423
00:17:07,120 --> 00:17:11,520
run pip install it remains the same.

424
00:17:10,160 --> 00:17:14,160
If you want that sort of thing, you

425
00:17:11,520 --> 00:17:17,839
should be running this command instead

426
00:17:14,160 --> 00:17:21,360
with the uh require hashes flag.

427
00:17:17,839 --> 00:17:23,039
And so unless you specifically opt into

428
00:17:21,360 --> 00:17:24,640
doing something like this, you actually

429
00:17:23,039 --> 00:17:26,160
just don't have a real practical

430
00:17:24,640 --> 00:17:28,640
guarantee you're getting the same

431
00:17:26,160 --> 00:17:31,640
contents as the last time you ran pip

432
00:17:28,640 --> 00:17:31,640
install.

433
00:17:31,919 --> 00:17:36,480
And here's another myth. Git tags are

434
00:17:34,240 --> 00:17:39,200
immutable.

435
00:17:36,480 --> 00:17:41,360
To remind you, git tags are very simple.

436
00:17:39,200 --> 00:17:42,880
They're just like aliases for a commit.

437
00:17:41,360 --> 00:17:44,880
That's pretty much it. We use them as a

438
00:17:42,880 --> 00:17:46,799
sort of human readable bookmark for a

439
00:17:44,880 --> 00:17:49,039
specific commit.

440
00:17:46,799 --> 00:17:51,280
The convention

441
00:17:49,039 --> 00:17:53,280
for Python packages as well as packages

442
00:17:51,280 --> 00:17:56,400
in other ecosystems is we tend to use

443
00:17:53,280 --> 00:17:59,360
git tags to point to the specific commit

444
00:17:56,400 --> 00:18:00,799
that a version was built at. Something

445
00:17:59,360 --> 00:18:03,200
that looks like what you're seeing on

446
00:18:00,799 --> 00:18:05,840
the slides here.

447
00:18:03,200 --> 00:18:08,400
Anyway, a big distinction between git

448
00:18:05,840 --> 00:18:10,240
branches and git tags is that tags are

449
00:18:08,400 --> 00:18:12,640
not meant to be reassigned after you

450
00:18:10,240 --> 00:18:15,120
created them.

451
00:18:12,640 --> 00:18:17,360
In other words, by design, or at least

452
00:18:15,120 --> 00:18:19,679
by intention, they're meant to be

453
00:18:17,360 --> 00:18:23,120
immutable.

454
00:18:19,679 --> 00:18:24,720
Except no, you can just do this. Uh

455
00:18:23,120 --> 00:18:26,640
there's no magic here. You really can

456
00:18:24,720 --> 00:18:28,320
just drop and recreate a tag on a

457
00:18:26,640 --> 00:18:31,200
different commit and then push it

458
00:18:28,320 --> 00:18:32,960
upstream. And it feels really trivial to

459
00:18:31,200 --> 00:18:35,600
stand here and point out that you can do

460
00:18:32,960 --> 00:18:37,760
this, but the fact that you can actually

461
00:18:35,600 --> 00:18:41,840
has some very interesting slash dire

462
00:18:37,760 --> 00:18:44,240
consequences to open source security.

463
00:18:41,840 --> 00:18:46,960
Let's put our black hats on. You have a

464
00:18:44,240 --> 00:18:50,640
malicious package fu version 1.0.0 that

465
00:18:46,960 --> 00:18:52,880
you want to publish to users on Pippi.

466
00:18:50,640 --> 00:18:54,480
You can upload this to Pippi with no

467
00:18:52,880 --> 00:18:56,240
issues. Uh but we have an obvious

468
00:18:54,480 --> 00:18:58,480
problem. Anyone who is suspicious of our

469
00:18:56,240 --> 00:18:59,760
malicious uh package can inspect our

470
00:18:58,480 --> 00:19:01,120
source code and they can see our

471
00:18:59,760 --> 00:19:03,520
wrongdoing because it's all in plain

472
00:19:01,120 --> 00:19:06,080
sight. So here's a trick to actually

473
00:19:03,520 --> 00:19:08,400
conceal our malicious code. Suppose we

474
00:19:06,080 --> 00:19:12,240
built and published FU version 1.0.0

475
00:19:08,400 --> 00:19:13,840
zero at the bad commit DF456

476
00:19:12,240 --> 00:19:16,000
and suppose we have a good commit right

477
00:19:13,840 --> 00:19:18,160
after this one's completely innocent the

478
00:19:16,000 --> 00:19:22,240
trick is we can move the version tag

479
00:19:18,160 --> 00:19:23,919
v1.0.0 from bad to good

480
00:19:22,240 --> 00:19:26,080
and the clever thing here is we've moved

481
00:19:23,919 --> 00:19:28,720
the tag from bad to good but Pipi is

482
00:19:26,080 --> 00:19:30,080
actually still publishing the bad one

483
00:19:28,720 --> 00:19:31,760
and this is great because now you can

484
00:19:30,080 --> 00:19:34,160
shake off suspicion very easily like

485
00:19:31,760 --> 00:19:36,799
anyone who suspects foul play they can

486
00:19:34,160 --> 00:19:40,160
just audit your project at version 1.0.0

487
00:19:36,799 --> 00:19:41,840
Z and they'll find nothing wrong.

488
00:19:40,160 --> 00:19:43,440
And it's it really is that easy. And

489
00:19:41,840 --> 00:19:45,600
people really do pull off this kind of

490
00:19:43,440 --> 00:19:48,000
exploit. If you're curious, search up

491
00:19:45,600 --> 00:19:50,240
the Bolt DB incident. People did this

492
00:19:48,000 --> 00:19:52,640
for a package and they actually had

493
00:19:50,240 --> 00:19:56,080
potential remote code execution over

494
00:19:52,640 --> 00:20:00,400
three years before people found out.

495
00:19:56,080 --> 00:20:02,640
Yep. Myth six, Python is dead. Python 2

496
00:20:00,400 --> 00:20:04,160
is dead.

497
00:20:02,640 --> 00:20:05,840
Yeah. Like so far in this conference,

498
00:20:04,160 --> 00:20:06,960
it's been AI this, AI that. I don't

499
00:20:05,840 --> 00:20:09,280
think I've heard a single mention of

500
00:20:06,960 --> 00:20:11,600
Python 2, so probably safe to say it's

501
00:20:09,280 --> 00:20:13,360
dead.

502
00:20:11,600 --> 00:20:16,320
Some background first. Python 2 was

503
00:20:13,360 --> 00:20:19,280
first released in October 2000.

504
00:20:16,320 --> 00:20:22,799
After 20 long years, we finally mark the

505
00:20:19,280 --> 00:20:24,799
end of life of Python 2 in April 2020.

506
00:20:22,799 --> 00:20:26,880
So, it's been a good 5 years since 2020.

507
00:20:24,799 --> 00:20:30,320
Is it dead?

508
00:20:26,880 --> 00:20:32,000
Nah, Python 2 is not dead yet. It has

509
00:20:30,320 --> 00:20:34,480
been 5 years, but people are definitely

510
00:20:32,000 --> 00:20:36,559
still using it. Much less actively for

511
00:20:34,480 --> 00:20:39,360
sure, but definitely still in use. In

512
00:20:36,559 --> 00:20:41,120
fact, the percentage of people

513
00:20:39,360 --> 00:20:43,840
reportedly using Python 2 actually

514
00:20:41,120 --> 00:20:47,360
hasn't changed that much since 2020. It

515
00:20:43,840 --> 00:20:50,640
was sitting at around 6% then and a year

516
00:20:47,360 --> 00:20:53,280
ago was sitting at around 4%.

517
00:20:50,640 --> 00:20:55,120
Anyway, let's look at something more

518
00:20:53,280 --> 00:20:57,919
concrete maybe. Let's look at the actual

519
00:20:55,120 --> 00:21:00,400
download stats from Pippi. This is a

520
00:20:57,919 --> 00:21:02,080
rough count of the number of packages

521
00:21:00,400 --> 00:21:04,799
downloaded in the last 12 months. And

522
00:21:02,080 --> 00:21:08,400
good news, looks like Python 2 is dead.

523
00:21:04,799 --> 00:21:10,640
But uh stretching out the y-axis a bit

524
00:21:08,400 --> 00:21:12,559
more, you can see there is a small but

525
00:21:10,640 --> 00:21:15,520
still persistent sliver remaining for

526
00:21:12,559 --> 00:21:18,159
Python 2. But like in relative terms,

527
00:21:15,520 --> 00:21:20,240
this looks okay, right? Uh the problem

528
00:21:18,159 --> 00:21:22,000
is Python is the world's most popular

529
00:21:20,240 --> 00:21:24,320
programming language at the moment. And

530
00:21:22,000 --> 00:21:26,559
that tiny sliver is actually three

531
00:21:24,320 --> 00:21:29,559
billion package downloads in the past 12

532
00:21:26,559 --> 00:21:29,559
months,

533
00:21:29,840 --> 00:21:35,520
which is to say that nearly 20 million

534
00:21:33,280 --> 00:21:39,360
packages have been downloaded since PyCon

535
00:21:35,520 --> 00:21:41,440
started this Friday at 9:00 a.m.

536
00:21:39,360 --> 00:21:43,919
And in the time it took to deliver that

537
00:21:41,440 --> 00:21:46,400
sentence, another 800 or so were

538
00:21:43,919 --> 00:21:48,720
downloaded.

539
00:21:46,400 --> 00:21:51,840
Yeah. So, Python 2 is not quite dead,

540
00:21:48,720 --> 00:21:53,200
but who cares? in general like why does

541
00:21:51,840 --> 00:21:55,440
it matter that people are using end of

542
00:21:53,200 --> 00:21:56,640
life things?

543
00:21:55,440 --> 00:21:58,400
It matters because when something

544
00:21:56,640 --> 00:22:00,400
becomes end of life it means you no

545
00:21:58,400 --> 00:22:02,400
longer get security patches on that

546
00:22:00,400 --> 00:22:04,000
thing or at least you don't reliably get

547
00:22:02,400 --> 00:22:06,000
them.

548
00:22:04,000 --> 00:22:08,320
In this case it's been like 5 years.

549
00:22:06,000 --> 00:22:10,159
Python 2 and its main libraries have had

550
00:22:08,320 --> 00:22:12,000
ample time for many vulnerabilities to

551
00:22:10,159 --> 00:22:15,440
be discovered and there's only ever

552
00:22:12,000 --> 00:22:17,280
going to be more over time.

553
00:22:15,440 --> 00:22:19,200
So what you're looking at here is just a

554
00:22:17,280 --> 00:22:21,280
very small sample of a few

555
00:22:19,200 --> 00:22:23,440
vulnerabilities affecting the Python 2

556
00:22:21,280 --> 00:22:25,440
standard library and also some

557
00:22:23,440 --> 00:22:28,400
cornerstone libraries like NumPy,

558
00:22:25,440 --> 00:22:30,799
Pillow, Django, etc.

559
00:22:28,400 --> 00:22:32,720
So if something gets marked end of life

560
00:22:30,799 --> 00:22:34,960
and that something actually matters to

561
00:22:32,720 --> 00:22:37,039
you and your organization,

562
00:22:34,960 --> 00:22:39,120
it really is best that you do something

563
00:22:37,039 --> 00:22:41,360
about it.

564
00:22:39,120 --> 00:22:43,520
And gosh, like if only there was some

565
00:22:41,360 --> 00:22:45,120
kind of new industry changing, emerging

566
00:22:43,520 --> 00:22:48,000
technology that could solve all this

567
00:22:45,120 --> 00:22:50,240
without human intervention. Nikki,

568
00:22:48,000 --> 00:22:52,159
surely something along those lines would

569
00:22:50,240 --> 00:22:54,240
come and save us.

570
00:22:52,159 --> 00:22:57,039
Interesting theory. If only we had Oh,

571
00:22:54,240 --> 00:23:01,280
we have a myth for that. Okay, our last

572
00:22:57,039 --> 00:23:03,360
myth. AI will save us. Yay. Um, but

573
00:23:01,280 --> 00:23:05,600
first, nostalgia story time with Nikki.

574
00:23:03,360 --> 00:23:10,000
And so at university I learned that

575
00:23:05,600 --> 00:23:12,080
computers are very very fast. Very dumb

576
00:23:10,000 --> 00:23:14,400
but very very very very fast. So that

577
00:23:12,080 --> 00:23:17,440
means they will do exactly what you tell

578
00:23:14,400 --> 00:23:19,360
them to and not necessarily what you

579
00:23:17,440 --> 00:23:21,760
want them to and they'll do it very

580
00:23:19,360 --> 00:23:24,000
quickly. So now one of those things is

581
00:23:21,760 --> 00:23:28,400
no longer a limiting factor. They no

582
00:23:24,000 --> 00:23:30,320
longer do exactly what you tell them to

583
00:23:28,400 --> 00:23:33,679
but rather they just kind of follow the

584
00:23:30,320 --> 00:23:35,200
vibe of it. LLMs can accelerate the

585
00:23:33,679 --> 00:23:37,360
generation of code. They can automate

586
00:23:35,200 --> 00:23:39,280
the creation of boilerplate. That's a

587
00:23:37,360 --> 00:23:41,039
great thing that they can do. LLMs can

588
00:23:39,280 --> 00:23:43,440
significantly lower the barrier to

589
00:23:41,039 --> 00:23:45,600
entry, maybe help onboard junior

590
00:23:43,440 --> 00:23:48,159
developers or even complete noviceses

591
00:23:45,600 --> 00:23:49,919
much, much faster. And that means that a

592
00:23:48,159 --> 00:23:52,640
lot of the tasks that traditionally

593
00:23:49,919 --> 00:23:56,240
might have required some

594
00:23:52,640 --> 00:23:59,200
critical thought no longer do. That's

595
00:23:56,240 --> 00:24:01,520
great. Like installing dependencies. Oh.

596
00:23:59,200 --> 00:24:04,799
Um, well, okay. So if we're moving from

597
00:24:01,520 --> 00:24:07,200
author to auditor, how closely are folks

598
00:24:04,799 --> 00:24:10,080
actually validating things and do they

599
00:24:07,200 --> 00:24:12,720
understand it? If you are looking, if

600
00:24:10,080 --> 00:24:16,720
they are looking, are they just checking

601
00:24:12,720 --> 00:24:18,880
to see that the vibes seem right? Um

602
00:24:16,720 --> 00:24:21,200
because AI does not solve our open

603
00:24:18,880 --> 00:24:24,559
source problems. It automates and

604
00:24:21,200 --> 00:24:28,480
accelerates human endeavors and human

605
00:24:24,559 --> 00:24:30,559
mistakes at industrial scale. Uh, and

606
00:24:28,480 --> 00:24:32,720
one of those mistakes that we've seen

607
00:24:30,559 --> 00:24:34,799
with our eyes on your dependencies is

608
00:24:32,720 --> 00:24:36,720
slop squatting. Think typos squatting,

609
00:24:34,799 --> 00:24:39,120
but specifically targeting AI generated

610
00:24:36,720 --> 00:24:41,919
code. So, specifically, an LLM model

611
00:24:39,120 --> 00:24:43,760
tendencies to hallucinate a non-existing

612
00:24:41,919 --> 00:24:46,159
package name because it turns out that

613
00:24:43,760 --> 00:24:48,559
LLMs don't know what packages actually

614
00:24:46,159 --> 00:24:52,799
exist. They just go by the vibes and

615
00:24:48,559 --> 00:24:55,679
what seems right. So, LLM make stuff up.

616
00:24:52,799 --> 00:24:57,760
That's what a generative model is. And

617
00:24:55,679 --> 00:24:59,600
according to this linked study, roughly

618
00:24:57,760 --> 00:25:02,080
20% of generated code samples had at

619
00:24:59,600 --> 00:25:05,279
least one recommended package that

620
00:25:02,080 --> 00:25:08,400
didn't exist. And crucially, more than

621
00:25:05,279 --> 00:25:10,960
half of hallucinated packages reappeared

622
00:25:08,400 --> 00:25:13,279
in multiple runs. So a malicious actor

623
00:25:10,960 --> 00:25:15,520
could pretty easily just run an LLM,

624
00:25:13,279 --> 00:25:18,000
find those packages that don't exist,

625
00:25:15,520 --> 00:25:21,679
find register those package names, and

626
00:25:18,000 --> 00:25:25,279
upload some malicious code. Ah, and if

627
00:25:21,679 --> 00:25:26,640
the developers aren't paying a lot of

628
00:25:25,279 --> 00:25:28,159
attention to what they're actually

629
00:25:26,640 --> 00:25:32,480
importing, because remember that those

630
00:25:28,159 --> 00:25:34,960
packages look totally plausible. Well,

631
00:25:32,480 --> 00:25:37,120
that's a bit scary. And there are other

632
00:25:34,960 --> 00:25:38,799
scary potential futures, too. Are LLM's

633
00:25:37,120 --> 00:25:40,960
actually going to be great at finding

634
00:25:38,799 --> 00:25:42,880
zero days in our source code? And how

635
00:25:40,960 --> 00:25:45,440
might we, as a community meet that

636
00:25:42,880 --> 00:25:47,200
challenge? Many maintainers are already

637
00:25:45,440 --> 00:25:49,360
pretty burnt out. They're already

638
00:25:47,200 --> 00:25:50,799
inundated with generated PRs and

639
00:25:49,360 --> 00:25:53,919
security reports generated by

640
00:25:50,799 --> 00:25:56,559
well-meaning people. Does the future

641
00:25:53,919 --> 00:25:59,840
actually look like reducing our reliance

642
00:25:56,559 --> 00:26:01,679
on open source? How long until some CEO

643
00:25:59,840 --> 00:26:03,760
says just vibe that up the whole thing?

644
00:26:01,679 --> 00:26:06,799
Then we won't be relying on that one dev

645
00:26:03,760 --> 00:26:09,120
in Nebraska and our code won't be open

646
00:26:06,799 --> 00:26:11,039
source. So we'll be protected, scare

647
00:26:09,120 --> 00:26:12,559
quotes intended, from those LLMs who are

648
00:26:11,039 --> 00:26:16,960
out there hunting for zero days trying

649
00:26:12,559 --> 00:26:19,840
to hack us. But do we want that future?

650
00:26:16,960 --> 00:26:23,679
Do we want Vibes source instead of open

651
00:26:19,840 --> 00:26:26,400
source? Your being here in this room

652
00:26:23,679 --> 00:26:28,880
demonstrates that you believe in the

653
00:26:26,400 --> 00:26:32,159
value of open source. We know the value

654
00:26:28,880 --> 00:26:34,320
of Eyes on the Code and the myriad other

655
00:26:32,159 --> 00:26:36,640
benefits that come from being part of an

656
00:26:34,320 --> 00:26:38,960
open source community.

657
00:26:36,640 --> 00:26:41,039
Well, no matter which path we go down,

658
00:26:38,960 --> 00:26:43,440
and I'd personally really like to avoid

659
00:26:41,039 --> 00:26:46,320
Vibe Source as a default,

660
00:26:43,440 --> 00:26:48,159
but hey, uh, no matter what future,

661
00:26:46,320 --> 00:26:51,440
scary or not, we end up in, it looks

662
00:26:48,159 --> 00:26:53,039
like there'll be jobs in security.

663
00:26:51,440 --> 00:26:53,679
Uh, so I'm Nikki.

664
00:26:53,039 --> 00:26:55,679
I'm Tim.

665
00:26:53,679 --> 00:26:56,960
And those were some myths about open

666
00:26:55,679 --> 00:26:58,310
source that we hope you no longer

667
00:26:56,960 --> 00:27:08,559
believe.

668
00:26:58,310 --> 00:27:10,799
[Applause]

669
00:27:08,559 --> 00:27:14,400
Thank you uh Nikki and Tim for the very

670
00:27:10,799 --> 00:27:15,120
riveting talk. I

671
00:27:14,400 --> 00:27:16,799
Scary.

672
00:27:15,120 --> 00:27:17,120
Scary. Very scary.

673
00:27:16,799 --> 00:27:18,720
Um

674
00:27:17,120 --> 00:27:20,240
but a little optimistic.

675
00:27:18,720 --> 00:27:21,840
Just a tiny bit.

676
00:27:20,240 --> 00:27:24,559
And you you can all go and check out um

677
00:27:21,840 --> 00:27:28,799
Emoji Kitchen now.

678
00:27:24,559 --> 00:27:31,919
Um please accept these uh speaker gifts

679
00:27:28,799 --> 00:27:35,840
on behalf of the conference. Um,

680
00:27:31,919 --> 00:27:39,320
do we have any questions? We have one

681
00:27:35,840 --> 00:27:39,320
question over there.

682
00:27:47,440 --> 00:27:52,080
Do you have any commentary on the case

683
00:27:49,039 --> 00:27:54,880
of um releasing on Pippi with an SIS and

684
00:27:52,080 --> 00:28:00,080
then at a later point in time adding a

685
00:27:54,880 --> 00:28:01,520
malicious uh wheel to the release? So,

686
00:28:00,080 --> 00:28:04,880
you know, you reinstall the same version

687
00:28:01,520 --> 00:28:06,320
of the package, but now it contains a,

688
00:28:04,880 --> 00:28:07,440
you know, a platform wheel that is

689
00:28:06,320 --> 00:28:11,360
malicious.

690
00:28:07,440 --> 00:28:12,799
Yeah, that seems bad. Um, uh, uh, I I'm

691
00:28:11,360 --> 00:28:16,720
looking at Caleb in the front row

692
00:28:12,799 --> 00:28:20,159
saying, "Yes, we have comments." Um, uh,

693
00:28:16,720 --> 00:28:24,480
come grab us afterwards. Yeah.

694
00:28:20,159 --> 00:28:27,520
Might have time for one short question.

695
00:28:24,480 --> 00:28:29,679
Um, you said that PIP does not grab

696
00:28:27,520 --> 00:28:33,679
hashes by default. Do you know off the

697
00:28:29,679 --> 00:28:35,840
top of your head if UV does?

698
00:28:33,679 --> 00:28:38,320
Yes, it does. Someone knows and they say

699
00:28:35,840 --> 00:28:39,360
yes, it does.

700
00:28:38,320 --> 00:28:41,039
Great.

701
00:28:39,360 --> 00:28:41,760
Any of the ones that use a lock by

702
00:28:41,039 --> 00:28:43,600
default.

703
00:28:41,760 --> 00:28:45,919
Any of the ones that use a lock file by

704
00:28:43,600 --> 00:28:48,720
default to do.

705
00:28:45,919 --> 00:28:53,020
Uh please join me again in thanking

706
00:28:48,720 --> 00:28:57,660
Nikki and Tim for the excellent talk.

707
00:28:53,020 --> 00:28:57,660
[Applause]