Post-Quantum Crypto with Michael Howard の全文書き起こし

00:03:58.990 very much a European mission with NASA instrumentations and They 74 00:03:59.009 --> 00:04:00.810 provided the Atlas V as well, but it was built 75 00:04:00.830 --> 00:04:03.810 by Airbus. And its goal was to get a view 76 00:04:04.050 --> 00:04:07.710 of the poles of the sun. Normally, you're launching your 77 00:04:07.729 --> 00:04:10.270 spacecraft because the Earth's in the plane of the ecliptic. 78 00:04:10.310 --> 00:04:11.909 Your spacecraft are going to be as well. Trying to 79 00:04:11.930 --> 00:04:13.780 get to high inclination is very, very difficult. Not that 80 00:04:13.789 --> 00:04:17.579 they got all that high. They'll fly down into a 81 00:04:17.920 --> 00:04:21.560 looping orbit inside the orbit of Mercury and then use 82 00:04:21.860 --> 00:04:26.199 Venus to do the slingshots and repeatedly tip the spacecraft. 83 00:04:26.230 --> 00:04:29.089 So it is about 24 degrees off the planet ecliptic. 84 00:04:29.110 --> 00:04:30.569 It takes a ton of energy to do that. They 85 00:04:30.589 --> 00:04:34.110 borrowed lots of energy from Venus, but it's a one-ton spacecraft, 86 00:04:34.129 --> 00:04:34.290 so it. 87 00:04:34.269 --> 00:04:34.670 Can do that. 88 00:04:35.490 --> 00:04:38.790 But it'll get us our first visual views of the 89 00:04:38.829 --> 00:04:39.560 poles of the sun. 90 00:04:39.920 --> 00:04:41.779 Cool mission. Yeah, that is cool. 91 00:04:41.920 --> 00:04:45.560 In May. the first crew dragon demo mission. So this 92 00:04:45.579 --> 00:04:48.790 was Bob, uh, Benneken and Doug Hurley go up to 93 00:04:48.810 --> 00:04:52.889 the space station and demonstrate that, uh, the commercial spacecraft 94 00:04:52.930 --> 00:04:55.370 can actually go to the space station properly. And that'll 95 00:04:55.389 --> 00:04:57.589 be followed up in November with a regular crew flight 96 00:04:57.629 --> 00:04:59.750 of four astronauts to crew the space station. 97 00:04:59.850 --> 00:05:01.110 So there you go. 98 00:05:01.250 --> 00:05:04.730 After what this would be nine years since the Atlantis 99 00:05:04.750 --> 00:05:07.769 had landed. And the only support for the station was, uh, 100 00:05:08.160 --> 00:05:10.920 Via Soyuz, now the Americans had a vehicle again, the 101 00:05:10.959 --> 00:05:11.519 former crew driver. 102 00:05:11.540 --> 00:05:14.290 I got a question, which Michael might know the answer to, 103 00:05:14.370 --> 00:05:18.949 but you said ESA and NASA did this thing together 104 00:05:19.110 --> 00:05:22.279 in 2020, but Brexit was in 2020. Was Was Britain 105 00:05:22.339 --> 00:05:24.990 part of ESA in 2020? Did they participate? 106 00:05:25.319 --> 00:05:27.370 Yeah, I don't think so. I could be wrong, but 107 00:05:27.389 --> 00:05:28.009 I don't think so. 108 00:05:28.250 --> 00:05:29.430 It's mostly Germany, France. 109 00:05:29.470 --> 00:05:29.769 Yeah. 110 00:05:29.829 --> 00:05:32.750 Okay. It's Airbus, it's Defense of the Space. All right. 111 00:05:32.810 --> 00:05:36.480 July, the Perseverance rover. So, this was the test article 112 00:05:36.620 --> 00:05:39.660 for the original Curiosity rover with a bunch of upgraded things, 113 00:05:39.699 --> 00:05:41.259 better wheels, better suspension. 114 00:05:41.300 --> 00:05:41.399 Yeah. 115 00:05:42.079 --> 00:05:46.939 New instruments, the Ingenuity helicopter, all of that stuff gets 116 00:05:47.000 --> 00:05:50.189 launched in July. July is the perfect time to launch 117 00:05:50.230 --> 00:05:53.449 to Mars, July 2020. There's a synchronicity to the orbits, right? 118 00:05:53.470 --> 00:05:56.629 They're in a two, three period. And so every roughly 119 00:05:56.689 --> 00:05:58.069 two years, you get a chance to do a bunch 120 00:05:58.089 --> 00:05:59.889 of flights. And so July, there's actually three. There's the 121 00:05:59.930 --> 00:06:03.500 Perseverance rover, which is huge. There's also China's very first 122 00:06:03.560 --> 00:06:08.329 mission to Mars, Tianwen-1. And then the UAE. the United 123 00:06:08.389 --> 00:06:13.829 Aramids launches their Hope climate orbiter to Mars. So three 124 00:06:13.870 --> 00:06:15.250 launches in the same month. Wow. 125 00:06:15.269 --> 00:06:15.970 Did they all make it? 126 00:06:16.310 --> 00:06:16.980 In October. 127 00:06:17.420 --> 00:06:18.339 The U.S. made it. 128 00:06:18.379 --> 00:06:19.089 They all make it. 129 00:06:19.120 --> 00:06:20.740 Yeah, actually. Very cool. 130 00:06:20.899 --> 00:06:24.160 And I mean, I'll talk more about the Tianwen-1 when 131 00:06:24.180 --> 00:06:27.189 it lands next year. So in episode 2021. Because that 132 00:06:27.209 --> 00:06:29.310 was China's first attempt to go to Mars and it 133 00:06:29.470 --> 00:06:33.750 fully worked. Nobody's pulled that off before. Everybody loses a 134 00:06:33.769 --> 00:06:34.889 few trying to get to Mars. 135 00:06:35.589 --> 00:06:37.810 Just name it. Everybody does. But China didn't. 136 00:06:38.009 --> 00:06:40.550 But, you know, the advantage of being, I don't know, 137 00:06:40.670 --> 00:06:44.660 fourth or fifth mover or something like that. In October, OSIRIS-REx, 138 00:06:44.779 --> 00:06:49.420 one of the asteroid missions, touches on asteroid Bennu. and 139 00:06:49.480 --> 00:06:51.480 collects a sample there. In fact, it does a little 140 00:06:51.560 --> 00:06:54.759 too well because when it touches, Bennu is very much 141 00:06:54.790 --> 00:06:57.350 a rubble pile and all that gravel goes everywhere and 142 00:06:57.750 --> 00:07:01.509 they actually have trouble closing up the sample container because 143 00:07:01.529 --> 00:07:03.709 there's too much stuff. They have to come up with 144 00:07:03.730 --> 00:07:07.629 such technical maneuvers as shaking a little off to try 145 00:07:07.689 --> 00:07:09.310 and get the thing closed up so they can put 146 00:07:09.329 --> 00:07:12.209 it in the capsule to return. Another asteroid mission at 147 00:07:12.250 --> 00:07:14.529 the end of the year in December, Hayabusa 2, will

148 00:07:14.569 --> 00:07:19.040 actually successfully return It's sample payload from the asteroid Regu, 149 00:07:19.060 --> 00:07:22.240 and that is a JAXA mission. A little recap on 150 00:07:22.279 --> 00:07:25.779 what SpaceX did in 2020. There was actually 25 launches 151 00:07:25.819 --> 00:07:28.699 from SpaceX, which at the time was an amazing number. 152 00:07:29.459 --> 00:07:33.899 Just remembering that SpaceX will do 150 this year. So obviously, 153 00:07:34.259 --> 00:07:37.870 there's the two Crew Dragons, the test run in March, 154 00:07:37.949 --> 00:07:41.519 and then the full payload in November. They'll also fly 155 00:07:41.519 --> 00:07:45.310 14 Starlink missions, 833 satellites for a network of almost 156 00:07:45.310 --> 00:07:47.029 900 by the end of the year. 157 00:07:47.050 --> 00:07:47.120 Wow. 158 00:07:47.129 --> 00:07:51.350 Have you ever seen the Starlink satellites leaving the spaceship? Yeah. 159 00:07:51.370 --> 00:07:52.800 It's like pizza boxes going out. 160 00:07:52.939 --> 00:07:55.660 Yeah, yeah. Well, now they've gotten bigger because of the 161 00:07:55.699 --> 00:07:58.000 V2 minis. And so, they only fly like 24 or 162 00:07:58.000 --> 00:08:00.819 28 of them depending on the inclination. But at this time, 163 00:08:00.879 --> 00:08:03.370 they're flying 60 a run. 164 00:08:03.389 --> 00:08:04.230 That's crazy. Wow. 165 00:08:04.620 --> 00:08:07.879 Just these huge numbers of satellites. And that's where they also, 166 00:08:08.259 --> 00:08:10.240 this is the year where they have the reflectivity problems. 167 00:08:10.439 --> 00:08:12.870 And so you can see them for an extended period 168 00:08:12.879 --> 00:08:15.720 of time until they learn how to orient them and 169 00:08:15.759 --> 00:08:18.160 paint them correctly so they're not so visible and don't 170 00:08:18.199 --> 00:08:19.139 bother people so much. 171 00:08:19.250 --> 00:08:21.189 I remember there was a bunch of astronomers that were 172 00:08:21.230 --> 00:08:23.949 complaining they were polluting the night sky with their telescopes. 173 00:08:24.209 --> 00:08:27.009 Yeah. Well, and that's still an issue, although let's face it, 174 00:08:27.009 --> 00:08:28.550 digital processing can fix that. 175 00:08:28.769 --> 00:08:29.290 Right. 176 00:08:29.449 --> 00:08:32.750 But making bright lights make it worse. So making sure 177 00:08:32.769 --> 00:08:34.250 it doesn't reflect sunlight helps. 178 00:08:34.549 --> 00:08:37.139 I've seen them though. I've seen the trail go across 179 00:08:37.240 --> 00:08:40.500 and it's pretty fascinating and a little bit I don't know. 180 00:08:40.539 --> 00:08:42.970 You know, SpaceX does some things that if you weren't 181 00:08:43.009 --> 00:08:45.570 paying attention, you'd think we're being invaded by aliens. 182 00:08:45.769 --> 00:08:48.330 Well, or, you know, or Dr. No is in full swing. 183 00:08:48.389 --> 00:08:51.879 Like the line between supervillain and tech billionaire is getting 184 00:08:51.899 --> 00:08:52.799 very narrow. 185 00:08:53.000 --> 00:08:55.220 Just people don't know. Like, you know, the first time 186 00:08:55.259 --> 00:08:57.840 that I saw the booster rocket spinning, you know, when 187 00:08:57.879 --> 00:09:01.789 it– I thought it was like a spaceship. And so 188 00:09:01.809 --> 00:09:03.769 did a million other people until I found out, oh, 189 00:09:03.809 --> 00:09:05.730 that's just SpaceX. That's what they do. And what are 190 00:09:05.769 --> 00:09:08.470 those lights going? Hey, is that Santa Claus? No, that's 191 00:09:08.509 --> 00:09:09.250 just SpaceX. 192 00:09:09.309 --> 00:09:13.600 Yeah. I use Stellarium and it'll show you all the SpaceX. 193 00:09:13.960 --> 00:09:17.620 Sure. Something's whizzing around out there. I know. Must be 194 00:09:17.659 --> 00:09:21.340 my Facebook friends. They're at average intelligence or ability to 195 00:09:21.399 --> 00:09:22.379 look things up. 196 00:09:22.379 --> 00:09:24.460 10,000 plus of them by the end of this year, 197 00:09:24.519 --> 00:09:26.990 just so you know. There'll be a bunch of other 198 00:09:27.009 --> 00:09:29.049 missions as well, including a couple of GPS satellites. They'll 199 00:09:29.080 --> 00:09:32.100 also start doing their Starship flights, the hop tests, and 200 00:09:32.120 --> 00:09:34.820 the first, what they called belly flop test, where they 201 00:09:34.860 --> 00:09:36.840 fire it up to a few kilometers up and then 202 00:09:36.879 --> 00:09:38.419 let it fall on its belly so they could actually 203 00:09:38.480 --> 00:09:41.059 land it. That was SN8 by the end of 2020. 204 00:09:41.059 --> 00:09:43.799 It does not go well, but they'll solve that and 205 00:09:43.879 --> 00:09:45.940 prove that this whole idea is even possible. All right, 206 00:09:45.960 --> 00:09:46.960 should we move on to computing? 207 00:09:47.120 --> 00:09:47.279 Yeah. 208 00:09:47.440 --> 00:09:50.440 We'll start in January with the OpenAI paper called the 209 00:09:50.559 --> 00:09:54.779 Neural Scaling Laws. Lead author is Jared Kaplan. There's a 210 00:09:54.779 --> 00:09:57.679 whole bunch of others, including Daryl Modi, who this time 211 00:09:57.720 --> 00:09:59.970 is at OpenAI, will eventually be the CEO of Anthrop. 212 00:10:00.600 --> 00:10:02.759 And this was the paper that sort of kicked off 213 00:10:02.840 --> 00:10:06.450 this idea of, you know, different from all the other 214 00:10:06.490 --> 00:10:08.929 machine learning models we did where we worry about overfitting 215 00:10:08.970 --> 00:10:12.539 and training sets and so forth. that for large language models, 216 00:10:12.580 --> 00:10:13.970 we should just train on as much data as we 217 00:10:13.990 --> 00:10:17.509 can possibly get. There's lots of debate as to whether 218 00:10:17.529 --> 00:10:19.809 this is that good an idea. We've definitely come into 219 00:10:19.889 --> 00:10:22.909 other techniques from there. But this paper is kind of 220 00:10:22.950 -->

00:10:25.570 the stimulus for what will be the insanity coming in 221 00:10:25.590 --> 00:10:28.720 the next couple of years. January is also when Microsoft 222 00:10:28.740 --> 00:10:30.960 switches over to Chromium as the rendering engine in the 223 00:10:31.039 --> 00:10:35.190 new Edge browser. The pandemic's in full swing in March 224 00:10:35.210 --> 00:10:38.360 is when lockdowns really kick off and everybody goes home 225 00:10:38.429 --> 00:10:41.940 and everybody's got Zoom. Teams explodes for better or worse. 226 00:10:42.559 --> 00:10:48.629 But a little thing I paid attention to was Terry Breton, 227 00:10:48.690 --> 00:10:53.879 who is the EU Internal Market Commissioner. reached out to 228 00:10:54.240 --> 00:10:58.139 Netflix and YouTube and Amazon Prime and asked him to 229 00:10:58.179 --> 00:11:01.679 turn off all the 4K features. He was concerned about 230 00:11:01.700 --> 00:11:05.899 the amount of available bandwidth across Europe as everybody went 231 00:11:06.000 --> 00:11:10.610 home and used the internet differently. Was it actually a crisis? 232 00:11:10.690 --> 00:11:13.470 Nobody knows for sure, or at least he's not talking 233 00:11:13.509 --> 00:11:17.090 about it. They started turning up the bandwidth again in May, 234 00:11:17.230 --> 00:11:20.889 and Netflix's Posts about this are interesting because they talk 235 00:11:20.929 --> 00:11:23.750 about network changes and increasing capacity and things like that. 236 00:11:24.370 --> 00:11:26.519 So maybe there really was a crisis because of the 237 00:11:26.559 --> 00:11:29.440 change in the Internet consumption due to COVID. But, you know, 238 00:11:29.519 --> 00:11:32.440 a lot of details aren't revealed necessarily. But that to 239 00:11:32.480 --> 00:11:33.320 me was very interesting. 240 00:11:33.340 --> 00:11:33.480 Yeah. 241 00:11:34.250 --> 00:11:37.600 April is when Uncle Satchit says, two years worth of 242 00:11:37.639 --> 00:11:42.580 digital transformation in two months. Because everybody had to all work. 243 00:11:42.620 --> 00:11:44.419 This is when I started doing True Run As a week, 244 00:11:44.539 --> 00:11:49.850 just talking about topics around what was necessary for sysadmins 245 00:11:51.529 --> 00:11:52.769 with everybody working from home. 246 00:11:52.990 --> 00:11:56.169 My brother is a programmer at a local company. He's 247 00:11:56.190 --> 00:11:59.169 been there for years and years. And they basically do 248 00:11:59.490 --> 00:12:02.309 online vehicle registrations. And they were the first in the 249 00:12:02.409 --> 00:12:07.840 area to do it. And their customers are states, not 250 00:12:07.960 --> 00:12:15.730 individual sales places, dealerships. But anyway, they were renting an 251 00:12:15.830 --> 00:12:19.470 office building in an office park out here, umpteen billion 252 00:12:19.529 --> 00:12:22.809 square feet. And during COVID, people went to work at home. 253 00:12:23.450 --> 00:12:27.259 And I just remember Jay coming to

rehearsal once and said, Well, 254 00:12:27.279 --> 00:12:29.759 I've resigned myself to the idea that I'm never going 255 00:12:29.799 --> 00:12:32.889 back to the office. And they basically moved out of 256 00:12:32.929 --> 00:12:35.590 the office. Yeah, lots did. They stopped renting it. And 257 00:12:35.629 --> 00:12:38.289 I think, in general, office space took a big dive 258 00:12:39.049 --> 00:12:39.850 in 2020. 259 00:12:39.850 --> 00:12:42.210 You think about all the property that Microsoft gave up 260 00:12:42.269 --> 00:12:42.850 in Bellevue. 261 00:12:43.000 --> 00:12:43.409 Yeah. 262 00:12:43.600 --> 00:12:46.480 They kept their buildings in Redmond, but all that rented space. 263 00:12:46.299 --> 00:12:47.019 In Bellevue went away. 264 00:12:47.340 --> 00:12:47.519 Right. 265 00:12:47.659 --> 00:12:48.559 And all of it went to Zoom. 266 00:12:48.820 --> 00:12:49.059 Yeah. 267 00:12:49.559 --> 00:12:50.120 Well, to Teams. 268 00:12:50.139 --> 00:12:54.769 Teams. The first virtual build in May is also when 269 00:12:54.909 --> 00:12:58.929 they do the full release of Blazor WebAssembly. 270 00:12:59.129 --> 00:12:59.649 Yeah. 271 00:13:00.309 --> 00:13:02.429 Obviously, a bunch of other cool announcements around building that 272 00:13:02.450 --> 00:13:05.990 time span. Of course, Blazor, not the first WA programming language, 273 00:13:06.070 --> 00:13:10.740 Golang added WA support back in 2018. In June, and again, 274 00:13:10.759 --> 00:13:14.679 no people remember much about this because it was mid-pandemic. 275 00:13:15.429 --> 00:13:19.690 Microsoft is a big splash about OpenAI GPT-3 being built 276 00:13:19.769 --> 00:13:22.990 on what they called the Azure supercomputer, which was a 277 00:13:23.070 --> 00:13:26.120 bunch of different Azure data centers harnessed together with 10,000 GPUs, 285,000 CPUs. 278 00:13:28.679 --> 00:13:33.120 cpu cores to build a get this 175 billion parameter 279 00:13:33.159 --> 00:13:36.029 model wow i mean big big big for the time 280 00:13:36.250 --> 00:13:38.230 not so much anymore but at the time. 281 00:13:38.110 --> 00:13:40.629 I remember uh brian mckay who's one of my happy 282 00:13:40.649 --> 00:13:43.789 next guys getting on slack and telling us how awesome 283 00:13:43.809 --> 00:13:46.289 this gpt thing was but it was difficult to set 284 00:13:46.350 --> 00:13:48.570 up at the time but then yeah you know i 285 00:13:48.610 --> 00:13:51.269 tried it and of course like everybody else was kind 286 00:13:51.289 --> 00:13:51.919 of blown away. 287 00:13:52.029 --> 00:13:55.600 Yeah Later that year in September is when Microsoft actually 288 00:13:55.679 --> 00:14:00.649 licenses GPT-3 from OpenAI, which I presume is how GitHub 289 00:14:00.669 --> 00:14:03.049 gets access to it. They'll make GitHub Copilot the following. 290 00:14:03.970 --> 00:14:08.610 A couple more stories. In November, Apple announces the M1 processor. 291 00:14:08.769 --> 00:14:09.009 Yes. 292 00:14:09.250 --> 00:14:11.840 And I mention this because this, I would argue, is 293 00:14:12.460 --> 00:14:17.000 Tim Cook's most memorable move. He was always a hardware guy. 294 00:14:17.019 --> 00:14:18.879 This is an astonishing piece of hardware. There's a system 295 00:14:18.919 --> 00:14:23.879 on a chip where the GPU and CPU, NPU, the 296 00:14:24.100 --> 00:14:26.440 IO and security buses and so forth are all literally 297 00:14:26.480 --> 00:14:28.820 in the same die. The memory is in the same package. 298 00:14:29.899 --> 00:14:33.679 So this is made by TSMC, five nanometer process, about 299 00:14:33.679 --> 00:14:37.679 16 billion transistors total. That includes eight CPU cores, eight 300 00:14:37.700 --> 00:14:40.610 GPU cores, a 16 core neural engine, and up to 301 00:14:40.610 --> 00:14:42.580 16 gigabytes of RAM right. 302 00:14:42.490 --> 00:14:43.070 On the package. 303 00:14:43.110 --> 00:14:46.669 So you get both lower power consumption and higher performance. 304 00:14:47.720 --> 00:14:48.759 And little would we realize. 305 00:14:48.899 --> 00:14:51.419 I have a MacBook Pro with an M1. Yeah. First gen. 306 00:14:51.679 --> 00:14:52.279 And it's good. 307 00:14:52.440 --> 00:14:54.580 Is that 16? Is that megabytes or gigabytes? 308 00:14:54.860 --> 00:14:55.139 Gigs. 309 00:14:55.490 --> 00:14:57.750 Yeah. That's RAM, RAM, not cache. 310 00:14:57.769 --> 00:14:58.429 That's RAM, RAM. 311 00:14:58.490 --> 00:14:58.889 Okay. Yeah. 312 00:14:58.909 --> 00:15:01.009 They're not putting cache on. The cache is there too. 313 00:15:01.269 --> 00:15:01.429 Okay. 314 00:15:01.450 --> 00:15:03.230 And also that RAM is shared with the GPU. 315 00:15:03.549 --> 00:15:04.549 Right. 316 00:15:04.570 --> 00:15:08.519 So what we're doing right now with LLMs and the 317 00:15:08.559 --> 00:15:10.919 new agent models and so forth, The M1 was built 318 00:15:10.940 --> 00:15:13.019 for it, not knowing it was built for it. They 319 00:15:13.039 --> 00:15:14.779 were just trying to make the most efficient computer they 320 00:15:14.799 --> 00:15:17.149 could consume the least power. And this is Apple's move 321 00:15:17.210 --> 00:15:20.769 back to ARM, having come from the Motorola chipset, moved 322 00:15:20.789 --> 00:15:23.190 to Intel for a few years. Now they're making their 323 00:15:23.230 --> 00:15:24.549 own thing based on ARM. 324 00:15:26.029 --> 00:15:26.289 Two more. 325 00:15:26.450 --> 00:15:29.720 December, both in December. The SolarWinds supply chain attack. So 326 00:15:29.740 --> 00:15:34.919 this is Russian SVR. The pure brilliance of this is unbelievable. 327 00:15:35.429 --> 00:15:38.860 So in September of 2019, hackers successfully break into the 328 00:15:38.899 --> 00:15:43.519 SolarWinds development environment and they insert code into the development chain. 329 00:15:43.960 --> 00:15:45.940 The way they do it is incredibly insidious. It's part 330 00:15:45.960 --> 00:15:49.200 of their build system that they replace code in the

331 00:15:49.279 --> 00:15:52.399 build without actually changing the visible source code. So developers 332 00:15:52.419 --> 00:15:53.889 don't think their code has changed at all, but what 333 00:15:53.909 --> 00:15:56.909 they're actually compiling is different code with this thing called 334 00:15:56.929 --> 00:16:01.750 the sunburst backdoor. They, they, they, It takes them a 335 00:16:01.769 --> 00:16:03.389 few months of doing testing to get this to work. 336 00:16:03.460 --> 00:16:06.700 By March, they have actually figured it out. And the 337 00:16:06.779 --> 00:16:10.399 Orion monitoring software, SolarWinds builds this infrastructure monitoring software, is 338 00:16:10.440 --> 00:16:14.019 now distributed to 18,000 customers with the Sunburst backdoor in it. 339 00:16:15.389 --> 00:16:18.190 The Russians are successful enough that in June, they actually 340 00:16:18.269 --> 00:16:21.870 removed the whole build injection system. Like, they're done. And 341 00:16:21.929 --> 00:16:24.580 are happily operating it. Now, admittedly, they don't actually use 342 00:16:24.600 --> 00:16:27.539 that backdoor much. Maybe 100 customers are totally affected. But 343 00:16:27.580 --> 00:16:30.799 it's in December when one of those customers, a security 344 00:16:30.840 --> 00:16:36.149 company called FireEye, realizes they've been exploited and traces it 345 00:16:36.210 --> 00:16:39.149 back to the Orion code base and basically pops the 346 00:16:39.190 --> 00:16:42.940 whole thing open. It's an incredibly sophisticated supply chain attack 347 00:16:42.980 --> 00:16:45.399 and scares just not actually everybody. It's not the first one, 348 00:16:45.440 --> 00:16:46.750 but in a lot of ways, it's the one that 349 00:16:46.759 --> 00:16:47.970 made the most news. 350 00:16:48.309 --> 00:16:51.350 It could be made into a feature film, actually, if 351 00:16:51.409 --> 00:16:52.470 it was handled correctly. 352 00:16:52.850 --> 00:16:53.830 It's astonishing. 353 00:16:53.850 --> 00:16:55.190 Yeah. 354 00:16:55.309 --> 00:16:58.029 And if the Russians hadn't decided to go after FireEye, 355 00:16:58.269 --> 00:17:00.360 who knows when it would have been detected. Just he 356 00:17:00.399 --> 00:17:02.759 went after the guys who were in the security space 357 00:17:02.799 --> 00:17:04.319 and good enough at attacking breach properly. 358 00:17:05.869 --> 00:17:08.170 Yeah, I remember when I was invited to a meeting 359 00:17:08.190 --> 00:17:11.779 to be briefed on the attack when we sort of 360 00:17:11.799 --> 00:17:14.339 knew what was going on. And yeah, I'll be frank, 361 00:17:14.700 --> 00:17:17.640 I was kind of gobsmacked actually by- Gobsmacked, yeah. By 362 00:17:17.660 --> 00:17:18.440 the sophistication. 363 00:17:18.640 --> 00:17:22.500 It's so clever. Holy man. I don't expect bad guys 364 00:17:22.539 --> 00:17:25.279 to be this smart, right? If they were smart, they'd 365 00:17:25.299 --> 00:17:27.240 be good guys. Like the idea that bad

guys would 366 00:17:27.259 --> 00:17:30.009 come up with something this clever. But again, state actors, 367 00:17:30.069 --> 00:17:32.170 like they're working for more than just a paycheck here. 368 00:17:32.490 --> 00:17:36.420 Yeah, but remember though, you know, As Sherrod Dugripo has 369 00:17:36.460 --> 00:17:39.599 told me many, many times, this is their day job, right? 370 00:17:39.680 --> 00:17:41.799 They come to work, they clock in, they do the work, 371 00:17:41.839 --> 00:17:44.740 they have reviews every year, they go home to their families, 372 00:17:45.029 --> 00:17:46.109 and they start again tomorrow. 373 00:17:46.150 --> 00:17:47.829 They get promotions by figuring this stuff out. 374 00:17:48.009 --> 00:17:49.529 Exactly. It's just their job. 375 00:17:49.569 --> 00:17:52.789 Yeah. As Dwayne LaFleur would say, oh, this is awesome, guys. 376 00:17:53.329 --> 00:17:55.690 This is awesome. It's a stunner. 377 00:17:56.670 --> 00:17:59.690 I'll finish out the year of compute in 2020 with 378 00:18:00.049 --> 00:18:04.430 China's Zhuheng Photonic Quantum Computer, which I think is particularly 379 00:18:04.490 --> 00:18:08.740 relevant for today's conversation. This was a dedicated machine using 380 00:18:09.000 --> 00:18:12.700 photonic quantum entanglement, which is very, very clever, unique, very 381 00:18:12.740 --> 00:18:16.480 different from Sycamore, the Google's device from the previous year, 382 00:18:16.539 --> 00:18:19.799 which was the … The hanging chandelier in liquid helium, 383 00:18:19.839 --> 00:18:21.279 this doesn't need any of this, but it was built 384 00:18:21.319 --> 00:18:23.940 specifically for Gaussian boson sampling. 385 00:18:24.299 --> 00:18:27.849 Nobody knows what you're talking about, Richard. I'm okay with that. 386 00:18:27.920 --> 00:18:29.109 I might contest. 387 00:18:29.190 --> 00:18:32.230 I mean, I know enough to be dangerous. Talk to 388 00:18:32.250 --> 00:18:33.470 me about the software side of it. 389 00:18:33.890 --> 00:18:35.890 Yeah, and that's the whole thing is this is nowhere 390 00:18:35.970 --> 00:18:38.670 near a general purpose computer or even a supercomputer. This 391 00:18:38.730 --> 00:18:41.650 is a machine to demonstrate that quantum entanglement can tackle 392 00:18:41.990 --> 00:18:45.009 one kind of problem, the Gaussian boson sample problem. Now, 393 00:18:45.029 --> 00:18:46.450 that's an important problem, but it was sort of a 394 00:18:46.609 --> 00:18:49.150 proof point I think very much this is a, hey, 395 00:18:49.210 --> 00:18:51.170 we get to play too. And I think everybody reacted 396 00:18:51.190 --> 00:18:54.299 to it that way. Because this 200-second run for something 397 00:18:54.319 --> 00:18:57.880 that in theory would have taken this supercomputer 2 billion years, 398 00:18:57.940 --> 00:19:00.720 not that anybody's going to test that, it just put 399 00:19:00.740 --> 00:19:05.009 China instantly on the map and really made it very clear. 400 00:19:05.529 --> 00:19:07.809 There's more than one way to quantum. And this is 401 00:19:07.869 --> 00:19:10.190 a wildly different way. Not that we've heard much from 402 00:19:10.269 --> 00:19:13.569 them since. But in 2020, that was a really big deal. 403 00:19:13.589 --> 00:19:14.490 But does it run Doom? 404 00:19:14.829 --> 00:19:15.390 It really doesn't. 405 00:19:15.670 --> 00:19:17.630 No, not a bit. No, it won't. 406 00:19:18.799 --> 00:19:21.140 I mean, supercomputers are limited in their own way, too, 407 00:19:21.200 --> 00:19:24.180 as well. But this was literally built for one thing. 408 00:19:24.200 --> 00:19:28.430 This is like Turing's device for cracking Enigma. Good for 409 00:19:28.549 --> 00:19:31.900 one thing. The idea of general-purpose computing is a much 410 00:19:31.960 --> 00:19:33.960 different idea than what we're doing in this kind of 411 00:19:33.980 --> 00:19:34.460 class of work. 412 00:19:34.619 --> 00:19:34.859 Anyway. 413 00:19:34.920 --> 00:19:37.759 Yeah, 100%. I think people need to understand that. Yeah, 414 00:19:37.880 --> 00:19:41.099 still people don't grapple that. There won't be an office 415 00:19:41.140 --> 00:19:43.269 for quantum. It's not going to exist. 416 00:19:43.329 --> 00:19:44.250 Oh, man. 417 00:19:44.490 --> 00:19:44.990 I'm sorry. 418 00:19:45.569 --> 00:19:49.890 You sold me that subscription. Let's get them on the phone. Yeah. 419 00:19:51.059 --> 00:19:54.680 And as much as we can tell at this time, 420 00:19:55.039 --> 00:19:58.220 these will always be supercomputers for particular problem spaces. Most 421 00:19:58.259 --> 00:20:02.009 of them very deterministic problem spaces, too. But let's wrap 422 00:20:02.089 --> 00:20:04.670 up the history lesson and get on to our larger 423 00:20:04.710 --> 00:20:05.529 conversation about quantum. 424 00:20:05.819 --> 00:20:07.839 Well, but first, we have to do. 425 00:20:07.960 --> 00:20:08.460 But first. 426 00:20:08.619 --> 00:20:10.680 Better know a framework. Roll the music. Awesome. 427 00:20:10.700 --> 00:20:15.059 All right, man. 428 00:20:15.160 --> 00:20:19.539 What do you got? 429 00:20:19.660 --> 00:20:22.150 All right. So, as people who listen to me on 430 00:20:22.400 --> 00:20:26.009 my various podcasts and stuff probably know, I bought this 431 00:20:26.890 --> 00:20:32.339 big GPU machine. a year or so ago, just because 432 00:20:32.420 --> 00:20:34.720 I anticipated being able to run local models. 433 00:20:35.160 --> 00:20:36.960 And you got it ahead of the hardware crisis too. 434 00:20:37.000 --> 00:20:38.400 So how smart are you? 435 00:20:38.599 --> 00:20:38.960 Yeah. 436 00:20:40.059 --> 00:20:44.460 I got it at walmart.com for $ 6, 000. It uses an 437 00:20:44.660 --> 00:20:48.480 NVIDIA RTX 1590 with 32 gigs of VRAM. 438 00:20:49.339 --> 00:20:51.740 Today that card's 15 grand if you can find one. 439 00:20:51.960 --> 00:20:55.509 I have, yeah, now it is. I have 96 gigs 440 00:20:55.750 --> 00:20:58.779 of system

RAM and it's You know, blinky lights and 441 00:20:58.839 --> 00:21:02.339 quiet as anything, right? Which is amazing. So on Coded 442 00:21:02.359 --> 00:21:05.430 with AI, Jeff Fritz and I did a series on 443 00:21:05.609 --> 00:21:10.089 taking a whole bunch of models that would run on 444 00:21:10.190 --> 00:21:15.750 Ollama and running them and then putting it through a test. Basically, 445 00:21:16.029 --> 00:21:21.539 a lot of them failed. More so because of context, 446 00:21:21.670 --> 00:21:23.680 I think, than this is what I'm learning now, the 447 00:21:23.720 --> 00:21:27.940 context size, if it's too small. We'll just barf. The 448 00:21:28.420 --> 00:21:32.960 LLM will just and lose it or get into an 449 00:21:33.019 --> 00:21:37.960 infinite loop. And basically what I came down to is 450 00:21:38.609 --> 00:21:45.910 running llama.cpp. And llama.cpp is like Ollama, but it's not. 451 00:21:46.029 --> 00:21:48.950 It's a different thing. It's open source. But I can 452 00:21:49.029 --> 00:21:58.720 run the fairly new QEN 3.8.27b model and llama cpp 453 00:21:58.779 --> 00:22:03.859 will use vram and system ram at the same time 454 00:22:04.000 --> 00:22:06.740 and it will balance them out and figure out automatically 455 00:22:06.880 --> 00:22:10.109 how much of which to use and let me tell 456 00:22:10.140 --> 00:22:14.450 you something this is i this is what i've been 457 00:22:14.490 --> 00:22:17.890 waiting for it works so well that i don't feel 458 00:22:17.910 --> 00:22:20.609 the need to go back for what i do you 459 00:22:20.630 --> 00:22:23.930 know debugging coding all of that stuff i don't need 460 00:22:24.089 --> 00:22:27.539 to go back to any frontier models anymore. 461 00:22:27.859 --> 00:22:28.019 Right. 462 00:22:28.140 --> 00:22:29.670 I haven't found, I've been using it for a couple 463 00:22:29.690 --> 00:22:29.930 of weeks. 464 00:22:29.950 --> 00:22:30.930 You're just working local now. 465 00:22:31.130 --> 00:22:33.019 I'm just working local. Yeah. And the only thing I'm 466 00:22:33.880 --> 00:22:36.400 that's costing me is electricity. But I got solar panels. 467 00:22:36.700 --> 00:22:38.839 There you go. Some pretty good, especially in the summer. 468 00:22:39.539 --> 00:22:41.740 Feeling good. And that machine you bought, like to build 469 00:22:41.759 --> 00:22:44.039 that machine today, it's a big bucks. 470 00:22:44.299 --> 00:22:44.559 Yeah. 471 00:22:44.700 --> 00:22:46.240 So you did the right thing. Well, I thought it 472 00:22:46.279 --> 00:22:48.779 was big bucks back then, but now... Yeah, it was. Yeah. 473 00:22:48.799 --> 00:22:50.579 It was a lot of money for a PC back then. 474 00:22:50.700 --> 00:22:52.099 It doesn't seem that way at this moment. 475 00:22:52.119 --> 00:22:53.640 Well, anyway... And what sort of performance are you getting 476 00:22:53.680 --> 00:22:55.089 out of it? Like tokens per second? 477 00:22:55.349 --> 00:22:59.539 That's great. It's as good as... Any frontier model that 478 00:22:59.559 --> 00:23:02.849 I've been using, I've been using GitHub Copilot CLI, mostly 479 00:23:02.869 --> 00:23:08.670 with Claude Sonnet. It's as good as that. I don't 480 00:23:08.710 --> 00:23:12.190 find myself waiting around. And it can do anything that 481 00:23:12.210 --> 00:23:15.750 I throw at it. It handles local stuff on the computer, 482 00:23:17.089 --> 00:23:23.140 stuff in Azure, in GitHub. It's just really good. This 483 00:23:23.240 --> 00:23:26.759 model compared to other models is good, but you run 484 00:23:26.799 --> 00:23:30.609 it in Lama CPP. And it's like beautiful on this 485 00:23:30.690 --> 00:23:35.289 particular configuration. I've found the sweet spot. So I wrote 486 00:23:35.309 --> 00:23:38.119 a document and that is the link that we'll put 487 00:23:38.279 --> 00:23:41.720 on the page. It's a GitHub. Read me basically running 488 00:23:41.819 --> 00:23:45.240 Quinn 3.8, 27 B with Lama CPP and GitHub co-pilot 489 00:23:45.259 --> 00:23:49.259 CLI on windows. And it tells you exactly how to 490 00:23:49.420 --> 00:23:52.470 install everything, all the stuff that you got to go, 491 00:23:52.710 --> 00:23:57.130 put everything where it needs to go. And like, Download 492 00:23:57.210 --> 00:24:01.970 Quinn and run it and how you access it remotely. 493 00:24:03.069 --> 00:24:06.079 It's great. So I have my dev machine. I have 494 00:24:06.099 --> 00:24:12.319 my GPU machine. And it's a match made in heaven. 495 00:24:13.359 --> 00:24:15.559 There's not a lot of people running those kinds of 496 00:24:16.539 --> 00:24:17.980 headless systems with Windows. 497 00:24:18.019 --> 00:24:18.660 They're mostly Linux. 498 00:24:18.779 --> 00:24:18.960 Right. 499 00:24:19.380 --> 00:24:23.700 Cool. It's good. And also, this model, Quinn 3.8, has 500 00:24:23.759 --> 00:24:28.109 vision support. So I can paste screenshots in, you know, 501 00:24:28.269 --> 00:24:30.779 and it just works. It's wonderful. Cool. 502 00:24:30.960 --> 00:24:31.099 Yep. 503 00:24:31.180 --> 00:24:32.710 That's what I got. Richard, who's talking to us? 504 00:24:33.000 --> 00:24:36.000 Grabbed a comment off of show 1963, which you did 505 00:24:36.059 --> 00:24:37.559 last year with one Michael Howard. 506 00:24:37.700 --> 00:24:38.019 Yay. 507 00:24:38.079 --> 00:24:40.140 Talking about 30 years of application security. This is back 508 00:24:40.160 --> 00:24:42.339 when you were still the red teamer. I know your 509 00:24:42.359 --> 00:24:45.160 life is different now. And this comment comes from also 510 00:24:45.200 --> 00:24:48.710 a past guest, Arnold Axelrod, who said, great show as always. 511 00:24:48.849 --> 00:24:51.369 One comment regarding input validation, because of course we talked 512 00:24:51.390 --> 00:24:54.069 about input validations. Toward the end of the show, Michael 513 00:24:54.109 --> 00:24:56.049 mentions that all input should be considered evil and must 514 00:24:56.089 --> 00:24:59.259 be validated in

order to be considered secure. Yeah, hard 515 00:24:59.279 --> 00:25:01.779 to argue with that. I'm not a security expert, but 516 00:25:01.799 --> 00:25:03.460 I have a different take on that. I don't think 517 00:25:03.480 --> 00:25:05.759 that the problem lies in the lack of input validation, 518 00:25:05.799 --> 00:25:08.940 but I think that input validation should be a business requirement. 519 00:25:10.319 --> 00:25:12.940 The example being zip codes. I don't necessarily know what 520 00:25:12.980 --> 00:25:15.049 a formal zip code is or will ever be consistent 521 00:25:15.069 --> 00:25:17.930 around the world. Answer is absolutely no. When you consider 522 00:25:17.950 --> 00:25:20.490 the idea that Ireland only got postal codes in 2015, 523 00:25:20.470 --> 00:25:25.190 like good luck. In my opinion, the problem lies with 524 00:25:25.210 --> 00:25:27.609 the use of parsers and interpreters and not using escape 525 00:25:27.650 --> 00:25:31.069 sequences or other structures to separate arbitrary input from structure ones, 526 00:25:31.569 --> 00:25:34.670 like separating the inputs from the SQL statement itself appropriately. 527 00:25:35.450 --> 00:25:39.549 Both SQL, JavaScript, and command through the process start in 528 00:25:39.589 --> 00:25:42.549 C-sharp are interpreters, and if you construct an input to 529 00:25:42.650 --> 00:25:46.289 them that contains an arbitrary user input without sanitizing it 530 00:25:46.329 --> 00:25:50.519 with the correct escape sequences is where unpredictability comes to play, 531 00:25:50.759 --> 00:25:55.779 which causes the risk. A URI also having structured format 532 00:25:55.940 --> 00:25:58.099 that is parsed by a browser and constructing a URI, 533 00:25:58.119 --> 00:26:00.559 let's say, with an arbitrary query string, that may be harmful, 534 00:26:01.180 --> 00:26:03.549 but you're probably using URI encoding on the inputs, and 535 00:26:03.569 --> 00:26:05.990 that should leave you safe. This doesn't require you to 536 00:26:06.029 --> 00:26:08.210 limit the input in any way, just to format it correctly. 537 00:26:09.250 --> 00:26:11.089 There is where the focus should be, as far as 538 00:26:11.130 --> 00:26:14.450 I'm concerned, not just to invalidate all inputs. I'd love 539 00:26:14.470 --> 00:26:15.430 to hear your opinions on it. 540 00:26:15.710 --> 00:26:16.390 You want my opinion? 541 00:26:16.569 --> 00:26:17.490 Go for it. Yeah, hit you. 542 00:26:18.819 --> 00:26:20.579 You know, it all got turned on its head, right? 543 00:26:20.630 --> 00:26:25.460 With LLMs and jailbreaking. What is the input? What is it? 544 00:26:25.460 --> 00:26:27.980 What is valid? Can you even escape it? Even escape 545 00:26:28.000 --> 00:26:32.980 versions can still get through any kind of checks. So, yeah, 546 00:26:33.319 --> 00:26:34.700 I just wrote about that recently. Yeah. 547 00:26:35.140 --> 00:26:39.460 Our new injection attack starts with ignore all previous instructions. 548 00:26:39.779 --> 00:26:41.960 Exactly. So, Mike Resinovich actually has a t-shirt with that 549 00:26:41.980 --> 00:26:45.539 on the back. He showed it to me at Build 550 00:26:45.799 --> 00:26:49.559 last time and I almost fell over laughing. 551 00:26:50.039 --> 00:26:50.839 It's the best. 552 00:26:50.980 --> 00:26:52.500 He was so proud of it. He said, hey, Michael, 553 00:26:52.519 --> 00:26:59.740 check this out. It's completely turned on its head. And 554 00:26:59.759 --> 00:27:02.259 that's why we've got all these other defenses that come 555 00:27:02.299 --> 00:27:04.829 into play and these guardrails and so on in LLMs. 556 00:27:04.990 --> 00:27:06.809 It's because we need them. We don't know what the 557 00:27:06.849 --> 00:27:07.329 input is. 558 00:27:07.450 --> 00:27:10.150 Yeah. LLMs need to get a sense of humor, right? 559 00:27:10.809 --> 00:27:13.309 A sense of sarcasm, a sense of irony. And they 560 00:27:13.329 --> 00:27:16.029 got to get sensitive to that just like we humans do. 561 00:27:16.490 --> 00:27:16.869 I think. 562 00:27:17.549 --> 00:27:17.869 I don't know. 563 00:27:18.349 --> 00:27:20.559 Yeah, we really want to pass every input through an 564 00:27:20.650 --> 00:27:22.500 LLM to say, is this a potential attack? 565 00:27:22.779 --> 00:27:25.299 Yeah, good luck with that. No, but of course not. 566 00:27:25.380 --> 00:27:27.119 But I mean, if you're dealing with an LLM and 567 00:27:27.140 --> 00:27:29.680 you're at the keyboard and talking to it, right? And 568 00:27:29.720 --> 00:27:33.640 you give it some ridiculous prompt, you know, that's clearly 569 00:27:34.089 --> 00:27:38.390 satirical or something. It should laugh back at you, you know? 570 00:27:38.849 --> 00:27:39.910 Well, that happened to me once. 571 00:27:40.329 --> 00:27:41.450 Yeah, that's a good one. Yeah. 572 00:27:41.470 --> 00:27:43.710 Dude, that happened to me once. All seriousness. 573 00:27:43.769 --> 00:27:43.930 Really? 574 00:27:43.970 --> 00:27:44.170 Yeah. 575 00:27:44.289 --> 00:27:48.250 I was working on the podcast website or something and 576 00:27:48.309 --> 00:27:50.750 asked it to do a quick security review of the code. 577 00:27:51.589 --> 00:27:54.559 And it found something. I was just running Visual Studio, 578 00:27:54.599 --> 00:27:57.420 I think. And it found something and it said, here 579 00:27:57.500 --> 00:28:01.559 is a, I don't know, it was an outdated HTTP header. 580 00:28:01.579 --> 00:28:05.640 It had been deprecated. I didn't know. A supposedly security 581 00:28:05.680 --> 00:28:09.150 header had been deprecated. And it said, by the way, 582 00:28:09.190 --> 00:28:12.329 this HTTP underscore blah, blah, blah has been deprecated for 583 00:28:12.369 --> 00:28:16.180 security reasons. And it's really ironic that you as the 584 00:28:16.259 --> 00:28:18.259 author of Running Secure Code didn't find this. 585 00:28:18.500 --> 00:28:19.180 Oh, that's great. 586 00:28:20.579 --> 00:28:20.859 Yeah. 587 00:28:21.259

--> 00:28:22.299 That's what I'm talking about. 588 00:28:22.319 --> 00:28:24.529 I don't know what the backend model was because, of course, 589 00:28:24.549 --> 00:28:26.769 Visual Studio, you can jump around, but I don't know. 590 00:28:26.829 --> 00:28:29.130 But yeah, it was quite happy to yell at me 591 00:28:29.170 --> 00:28:29.720 and laugh at me. 592 00:28:30.019 --> 00:28:30.619 That's pretty good. 593 00:28:30.660 --> 00:28:32.880 When the tool says the word, it is trying to 594 00:28:32.920 --> 00:28:35.619 detect irony. That's just ironic, actually. 595 00:28:36.160 --> 00:28:36.400 Right. 596 00:28:36.839 --> 00:28:37.039 Yeah. 597 00:28:37.259 --> 00:28:40.519 It's very recursive. Arnon, thank you so much for your comment. 598 00:28:40.539 --> 00:28:41.680 And a copy of Music to Code By is on 599 00:28:41.720 --> 00:28:42.819 its way to you. And if you'd like a copy 600 00:28:42.839 --> 00:28:44.119 of Music to Code By, write a comment on the 601 00:28:44.140 --> 00:28:47.180 website at. netrocks. com or on the Facebooks we publish every 602 00:28:47.200 --> 00:28:48.819 show there. And if you comment there and I read 603 00:28:48.839 --> 00:28:50.000 it on the show, we'll send you a copy of 604 00:28:50.019 --> 00:28:50.579 Music to Code By. 605 00:28:50.660 --> 00:28:52.349 Or if you'd just rather buy Music to Code By, 606 00:28:52.380 --> 00:28:55.829 go to musictocodeby.net. You can get the tracks in MP3, WAV, 607 00:28:56.069 --> 00:28:59.880 and FLAC formats. Well, before we introduce Michael... Let's take 608 00:28:59.900 --> 00:29:06.789 a little break for these very important messages. And we're back.. 609 00:29:07.329 --> 00:29:10.190 NET Rocks. I'm Carl Franklin. That's Richard Campbell. Hey. And 610 00:29:10.210 --> 00:29:13.710 that's Michael Howard. Let me introduce him formally. Your bio 611 00:29:13.750 --> 00:29:17.549 has changed a little bit. Michael Howard's been at Microsoft 612 00:29:17.809 --> 00:29:18.990 since 1992. 613 00:29:18.990 --> 00:29:19.230 Wow. 614 00:29:19.910 --> 00:29:24.789 Always in some form of security, IIS, SDL, the founder 615 00:29:24.809 --> 00:29:25.329 of SDL? 616 00:29:25.490 --> 00:29:27.069 Well, he's one of the guys that worked on the very, 617 00:29:27.109 --> 00:29:29.839 very earliest versions. It was like two or three of us, yeah, 618 00:29:29.859 --> 00:29:30.380 back in the day. 619 00:29:30.400 --> 00:29:30.940 Awesome. 620 00:29:31.000 --> 00:29:32.859 And that was the security lifecycle? 621 00:29:33.140 --> 00:29:37.480 Security development lifecycle, yeah. Back in the earliest, like, 2004. 622 00:29:36.759 --> 00:29:40.380 Also uh you worked in azure data on the red 623 00:29:40.420 --> 00:29:43.660 team last time we talked and now you're in post 624 00:29:43.839 --> 00:29:45.660 quantum crypto. 625 00:29:45.079 --> 00:29:51.509 Woohoo yeah baby Man, I'm happy to be here, too. 626 00:29:51.569 --> 00:29:52.130 It's a lot of fun. 627 00:29:52.279 --> 00:29:55.099 I mean,

how can you have post-quantum if we haven't 628 00:29:55.119 --> 00:29:56.960 had quantum? 629 00:29:57.160 --> 00:30:01.180 Because the real attacks start post-quantum. That's the reason why. 630 00:30:01.200 --> 00:30:04.200 Yeah, you're right. I get it. You need to be ready. 631 00:30:05.059 --> 00:30:07.819 For certain things. Certain things, maybe not. But we can 632 00:30:07.859 --> 00:30:08.599 discuss that later. 633 00:30:08.880 --> 00:30:10.940 But yeah. Well, what does it mean, post-quantum crypto? 634 00:30:11.240 --> 00:30:15.980 Well, you think of things in three ways. This is 635 00:30:16.000 --> 00:30:17.799 the way we think about it at Microsoft. Anyway, I'm 636 00:30:17.839 --> 00:30:21.769 sure most of the industry does. um, data, data in transit, 637 00:30:21.789 --> 00:30:24.930 data at rest, and then cryptographic trust. So data on 638 00:30:24.950 --> 00:30:27.259 the wire, you know, stuff flying across the internet is 639 00:30:27.279 --> 00:30:30.759 probably protected by TLS today, more than likely perhaps SSH, 640 00:30:31.339 --> 00:30:34.700 but certainly TLS data at rest is encrypted most of 641 00:30:34.720 --> 00:30:37.819 the time or should be. And those encryption keys are 642 00:30:37.859 --> 00:30:41.500 wrapped with other keys, uh, key wrapping keys. And then 643 00:30:41.519 --> 00:30:44.660 you've got cryptographic trust, which is, you know, signatures, certificates, 644 00:30:45.039 --> 00:30:49.140 all that sort of good stuff. Um, The real threat 645 00:30:49.160 --> 00:30:52.170 is dead on the wire is being, you know, being 646 00:30:52.650 --> 00:30:54.589 pulloined as it flies across the wire. 647 00:30:54.609 --> 00:30:55.250 Yeah. 648 00:30:55.329 --> 00:30:57.589 And then when a quantum computer comes out in the future, 649 00:30:57.609 --> 00:31:02.890 that stuff can be broken. And the breaking aspect is 650 00:31:03.480 --> 00:31:06.599 essentially just breaking the RSA or the elliptic curve wrapping... 651 00:31:06.809 --> 00:31:10.230 that goes on and out pops the AES key that's 652 00:31:10.289 --> 00:31:12.319 used for the bulk encryption, and now you just decrypt everything. 653 00:31:13.099 --> 00:31:14.359 And in the case of data at rest, it's the 654 00:31:14.380 --> 00:31:17.319 key wrapping keys, right? They can be broken because they usually, 655 00:31:17.579 --> 00:31:21.240 for example, RSA, which can be broken with an algorithm 656 00:31:21.259 --> 00:31:23.240 called Shor's algorithm, S-H-O-R. 657 00:31:23.299 --> 00:31:23.500 Yeah. 658 00:31:24.180 --> 00:31:28.140 And it basically finds periodicity in the way those algorithms work. 659 00:31:28.160 --> 00:31:29.269 That's what it takes advantage of. 660 00:31:29.910 --> 00:31:34.309 Are our logs at risk for being pilfered by cryptos? 661 00:31:35.769 --> 00:31:38.210 Post-mortem crypto? I mean, if it contains sensitive data, which 662 00:31:38.230 --> 00:31:40.200 you shouldn't be logging sensitive data, right? No, no, no. 663 00:31:40.240 --> 00:31:45.019 But even if it wasn't containing sensitive data, is that 664 00:31:45.059 --> 00:31:47.119 something because it can go through so much data so 665 00:31:47.180 --> 00:31:48.180 fast that it could. 666 00:31:48.220 --> 00:31:50.119 I mean, if I had to prioritize stuff, I would 667 00:31:50.180 --> 00:31:51.589 focus on the actual data itself. 668 00:31:51.609 --> 00:31:54.309 Yeah, okay. So data in transit and then real data 669 00:31:54.329 --> 00:31:55.150 in your databases. 670 00:31:55.390 --> 00:31:58.789 Correct. Yeah. And then those, again, they can be snaffled. 671 00:31:59.099 --> 00:32:02.680 today and then once a quantum computer is made available and. 672 00:32:02.640 --> 00:32:04.599 That's where a lot of did you say snaffled. 673 00:32:04.519 --> 00:32:06.579 Are stolen snaffled purloined. 674 00:32:06.259 --> 00:32:08.720 Snaffled that's a good british. 675 00:32:08.299 --> 00:32:12.240 Word snaffled yeah um there's actually a real word i 676 00:32:12.240 --> 00:32:13.809 don't even know is it a real. 677 00:32:13.670 --> 00:32:15.170 Word it doesn't matter i i don't know but i 678 00:32:15.230 --> 00:32:15.650 never heard it. 679 00:32:15.650 --> 00:32:18.849 Before it is now there you go things you. 680 00:32:18.789 --> 00:32:21.789 Learn on dot numrocks all right It's all good. 681 00:32:22.109 --> 00:32:24.849 Yeah, so the risk is the asymmetric keys that are 682 00:32:24.910 --> 00:32:27.509 used to wrap the symmetric keys that do the- Because 683 00:32:27.529 --> 00:32:30.150 they're dependent on prime numbers. In the case of RSA. 684 00:32:30.369 --> 00:32:30.589 Yeah. 685 00:32:30.849 --> 00:32:33.029 But that's not the attack. The attack is the periodicity. 686 00:32:33.049 --> 00:32:35.150 There's a periodicity. If you actually look at the way 687 00:32:35.210 --> 00:32:37.559 Shor's works, not that I say, not that you should, 688 00:32:38.200 --> 00:32:41.559 it basically does a quantum fast Fourier analysis. 689 00:32:41.819 --> 00:32:42.420 Yeah. 690 00:32:42.880 --> 00:32:46.779 Shor's is actually hybrid. It's actually quantum and sort of, 691 00:32:47.329 --> 00:32:50.970 classic computing. The hard work is done by Shor's and 692 00:32:51.009 --> 00:32:53.950 then some of the less difficult work is done classically 693 00:32:53.970 --> 00:32:57.910 because it's just easier. And so the real issue is 694 00:32:57.950 --> 00:33:02.210 that those asymmetric keys, RSA, elliptic curve, Diffie-Hellman, they all 695 00:33:02.269 --> 00:33:06.200 exhibit some periodicity that can be detected by Shor's. And 696 00:33:06.220 --> 00:33:09.869 that gives you a whole bunch of possibilities that then 697 00:33:09.930 --> 00:33:12.289 classical computing can then just sort of sift through and 698 00:33:12.509 --> 00:33:14.309 find out which ones are the actual keys. Right. 699 00:33:14.349 --> 00:33:16.710 So it narrows the scope of the testing needed. 700 00:33:16.930 --> 00:33:20.519 Correct. Yeah. And

to your point before, rather than, you know, 701 00:33:20.559 --> 00:33:23.380 squillions of ages of the universe, it could be hours 702 00:33:23.599 --> 00:33:24.559 or days. Yeah. 703 00:33:24.640 --> 00:33:25.019 Right. 704 00:33:25.059 --> 00:33:27.819 You know, it's, it's, it's a real thing. And, um, 705 00:33:28.759 --> 00:33:31.759 you know, developers have a big part to play in, 706 00:33:31.920 --> 00:33:35.019 in this. It's not just, you know, flip some switch 707 00:33:35.059 --> 00:33:39.059 and everything's golden. Um, There's a lot more to it 708 00:33:39.079 --> 00:33:41.140 that developers need to understand as well. 709 00:33:42.039 --> 00:33:48.079 So everything I've barely hung on understanding about quantum is 710 00:33:48.119 --> 00:33:53.829 that we're all screwed, right? And because of the way 711 00:33:53.880 --> 00:33:59.710 that TLS works and SSL and all that stuff, it 712 00:33:59.990 --> 00:34:02.750 dramatically has to change, doesn't it? If we're going to 713 00:34:02.769 --> 00:34:10.369 be less susceptible to quantum interference but you're i think 714 00:34:10.449 --> 00:34:12.449 what you're saying is that there are ways that we 715 00:34:12.489 --> 00:34:16.389 can do that now and that's what you know post-quantum 716 00:34:16.429 --> 00:34:20.909 crypto is all about is trying to use cryptographic methods 717 00:34:21.050 --> 00:34:24.610 now that will survive the quantum onslaught is that what 718 00:34:24.650 --> 00:34:27.849 you're basically up against yeah one. 719 00:34:27.789 --> 00:34:29.679 Of the beauties of tls by the way i can't 720 00:34:29.699 --> 00:34:32.300 believe you said ssl like wash your mouth out. 721 00:34:32.989 --> 00:34:38.010 You know, some of us were around in the 90s. 722 00:34:38.170 --> 00:34:42.199 So was I. And so they don't use those words, TLS. 723 00:34:42.619 --> 00:34:47.559 I know. Anyway, I'll just pretend you didn't say that. Yeah, 724 00:34:49.260 --> 00:34:51.300 so TLS, one of the beauties of TLS is that 725 00:34:51.380 --> 00:34:53.880 it's very agile, right? You can change the way it works, 726 00:34:53.920 --> 00:34:57.719 the ciphers that are used, the cryptographic primitives that are used. 727 00:34:57.840 --> 00:35:00.500 And you're talking about TLS 1.0. 728 00:35:00.500 --> 00:35:02.570 Correct. So TLS 1.2 should be using. 729 00:35:02.860 --> 00:35:03.110 Yeah. 730 00:35:03.139 --> 00:35:05.090 So actually that's a really important point. So TLS 1.2 731 00:35:05.090 --> 00:35:08.610 and prior, so TLS 1.0 and 1.1 are deprecated anyway, 732 00:35:08.650 --> 00:35:12.070 so don't use them. TLS 1.2 does its cipher suite 733 00:35:12.130 --> 00:35:15.659 and its key establishments and authentication all as one string 734 00:35:16.239 --> 00:35:18.159 called the cipher suite. It looks like someone sneezed on 735 00:35:18.199 --> 00:35:20.760 the screen basically. It's a list of all the algorithms 736 00:35:20.780 --> 00:35:23.820 that are used for all of those things. TLS 1.3 737 00:35:23.820 --> 00:35:28.449 is different. it broke apart things like the key establishment 738 00:35:28.849 --> 00:35:31.809 from the bulk encryption and the bulk tamper detection. They're 739 00:35:31.849 --> 00:35:32.949 completely broken apart. 740 00:35:33.050 --> 00:35:33.269 Yeah. 741 00:35:33.510 --> 00:35:36.880 So you can negotiate the two separately. That's the big 742 00:35:36.909 --> 00:35:40.539 difference in TLS 1.3. So the key establishment is a 743 00:35:40.579 --> 00:35:44.079 thing called a group. And the reason why it's called 744 00:35:44.119 --> 00:35:47.840 a group is because mathematicians got to hang on this. 745 00:35:47.880 --> 00:35:51.050 And they said, you know, all these things are mathematical groups. 746 00:35:51.780 --> 00:35:56.659 So we'll call them groups. Terrible name. But for the 747 00:35:56.679 --> 00:36:00.639 key establishment, that's where you set the algorithm or algorithms 748 00:36:00.699 --> 00:36:03.260 in some cases. And then after that, separately, is how 749 00:36:03.320 --> 00:36:05.579 you do bulk protection of the data on the wire. 750 00:36:05.599 --> 00:36:08.960 So TLS 1.3 broke those two apart. So it is 751 00:36:09.000 --> 00:36:12.239 completely not compatible with TLS 1.2. And TLS 1.2 will 752 00:36:12.300 --> 00:36:15.360 never be post-quantum. I mean, it's just software. I mean, essentially, 753 00:36:15.389 --> 00:36:17.110 I suppose it could make it post-quantum. 754 00:36:17.590 --> 00:36:18.510 Yeah, but why would you? 755 00:36:18.550 --> 00:36:21.610 But the interoperability... well, the interoperability story would be horrendous. 756 00:36:21.650 --> 00:36:23.070 Like no one would do it. 757 00:36:23.389 --> 00:36:25.750 But it also seems unnecessary too, right? Because I set 758 00:36:25.769 --> 00:36:28.489 the TLS 1.3 with a dropdown in Azure. 759 00:36:28.789 --> 00:36:31.739 Right. But the thing is, here's the issue. And this 760 00:36:31.800 --> 00:36:34.320 is one thing that we're having to work on for 761 00:36:34.360 --> 00:36:37.139 products like Front Door, for example, is you will be 762 00:36:37.159 --> 00:36:41.860 able to control the group, not just the bulk cryptography 763 00:36:41.909 --> 00:36:42.360 that's used. 764 00:36:42.570 --> 00:36:42.829 Okay. 765 00:36:42.889 --> 00:36:44.789 And that's where, and the group is where the post-quantum 766 00:36:44.809 --> 00:36:50.610 part comes in. For the most part, symmetric stuff AES-256, SHA-384, 767 00:36:50.650 --> 00:36:52.889 and so on, which are the baselines, are fine. There's 768 00:36:52.909 --> 00:36:55.679 another algorithm there called Grover's, which is an attack against 769 00:36:55.719 --> 00:36:59.300 symmetric algorithms. And it essentially cuts the number of bits 770 00:36:59.360 --> 00:37:02.300 in the key in half. So if you have an 771 00:37:02.380 --> 00:37:06.900 AES-256 in a quantum world, that's the same as AES-128. Interesting. 772 00:37:06.920 --> 00:37:07.420 Which

is fine. 773 00:37:07.800 --> 00:37:11.150 So you must use AES-256 and SHA-384 as the minimum 774 00:37:11.309 --> 00:37:14.110 as well. They're kind of okay. 775 00:37:14.250 --> 00:37:14.650 They're fine. 776 00:37:14.849 --> 00:37:20.170 The problem is the asymmetric stuff. Right. Yeah. Elliptic curve, RSA, Diffie-Hellman. 777 00:37:20.769 --> 00:37:22.309 That's where the problem is. Now, the nice thing in 778 00:37:22.510 --> 00:37:26.110 TLS 1.3 is that's defined in a group, and that's 779 00:37:26.159 --> 00:37:32.119 negotiated separately from the bulk cryptography. And that's where the 780 00:37:32.159 --> 00:37:35.059 hybrid algorithms come into play. And the reason why they're 781 00:37:35.079 --> 00:37:38.199 called hybrid is you use two together. You use elliptic 782 00:37:38.239 --> 00:37:44.679 curve and MLChem. So MLChem is the quantum resilient algorithm. 783 00:37:44.880 --> 00:37:48.960 Elliptic curve is classic. you build up keys in both, 784 00:37:49.000 --> 00:37:50.920 you smush them together, pass it through a hash, and 785 00:37:50.940 --> 00:37:53.539 then you derive the session keys after that. So they're 786 00:37:53.559 --> 00:37:54.239 both used together. 787 00:37:54.440 --> 00:37:57.159 So that would have to be implemented both at the 788 00:37:57.199 --> 00:37:59.360 browser level and at the server level, right? 789 00:37:59.480 --> 00:38:02.920 Yeah, everywhere. I mean, you say browser, but client. I mean, 790 00:38:02.940 --> 00:38:04.079 I just mean clients in general. 791 00:38:04.099 --> 00:38:04.860 Yeah, clients, sure. 792 00:38:04.940 --> 00:38:08.039 And in fact, you bring up a very important point there, Carl, 793 00:38:08.500 --> 00:38:12.159 and that is that all modern browsers support hybrid TLS 1.3. 794 00:38:12.159 --> 00:38:12.679 Yeah. 795 00:38:13.639 --> 00:38:17.800 So the, I mean, even the humble Xbox has hybrid. 796 00:38:19.159 --> 00:38:22.679 I've tried all sorts of, you know, iOS, Android, Windows, Mac, Linux, 797 00:38:23.349 --> 00:38:25.230 and all the common browsers support. 798 00:38:25.730 --> 00:38:29.570 So anything, anything that uses it like curl or any, 799 00:38:29.710 --> 00:38:32.670 any little command line tool, all, all those things have 800 00:38:32.690 --> 00:38:33.230 to support it. 801 00:38:33.329 --> 00:38:36.570 Correct. Correct. So SSH, both client server supports it as 802 00:38:36.610 --> 00:38:40.699 a version 10, I think maybe 9.9 or 10. supports 803 00:38:40.960 --> 00:38:43.880 um ml chem so the the important part there in 804 00:38:43.900 --> 00:38:46.139 the in ml chem is the letter l in that 805 00:38:46.380 --> 00:38:51.199 there's lattice and the two major algorithms that are post 806 00:38:51.219 --> 00:38:53.699 quantum resilient there's actually a small number but the two 807 00:38:53.800 --> 00:38:56.929 major ones ml chem and ml dsa the l is 808 00:38:56.989 --> 00:39:01.769 lattice and the is that lattice construct that is quantum resilient. 809 00:39:01.610 --> 00:39:04.150 Right so when you say quantum resilient do you mean 810 00:39:04.769 --> 00:39:07.969 what you perceive as the first generation of quantum or 811 00:39:08.280 --> 00:39:10.539 all quantum going forward till the end of time? 812 00:39:10.900 --> 00:39:15.719 Nah, I mean, NIST is still going through other algorithms. 813 00:39:16.900 --> 00:39:19.659 The industry has settled, and NIST has settled on MLDSA 814 00:39:19.699 --> 00:39:23.429 and MLChem. And there's been a lot of research for 815 00:39:23.469 --> 00:39:27.130 the last 20-something years in lattices, looking at it through 816 00:39:27.150 --> 00:39:32.820 a quantum lens, and they look good. but everyone's you 817 00:39:32.840 --> 00:39:34.360 know quite happy to say let's you know let's go 818 00:39:34.380 --> 00:39:36.440 look at other algorithms as well just in case and 819 00:39:36.480 --> 00:39:40.460 in fact for no other reason than the resulting cipher 820 00:39:40.519 --> 00:39:43.469 blob is big so i'll give you an example If 821 00:39:43.510 --> 00:39:46.260 you have an elliptic curve, say an EC25519, which is 822 00:39:46.929 --> 00:39:49.519 a curve, the signature, a digital signature for that is 823 00:39:49.519 --> 00:39:52.559 64 bytes in size. It's pretty small. If you take 824 00:39:52.579 --> 00:39:57.199 an MLDSA87 certificate and you sign data with the private 825 00:39:57.239 --> 00:40:00.670 key of that, it's about 4.5K. So you imagine if 826 00:40:00.690 --> 00:40:04.369 you've got a whole series of certificates all with their signatures... 827 00:40:04.869 --> 00:40:06.110 it adds up real quick. 828 00:40:06.349 --> 00:40:07.050 Yeah. 829 00:40:07.070 --> 00:40:11.489 And you can have problems with MTUs, with people passing 830 00:40:11.530 --> 00:40:15.889 stuff on query strings, amounts of space set aside on 831 00:40:16.130 --> 00:40:21.079 disk for signatures, you know? Yeah, so it's a real issue. 832 00:40:21.119 --> 00:40:25.559 So NIST is continuing to evaluate other algorithms with an 833 00:40:25.860 --> 00:40:30.300 eye to potentially smaller signatures. So now I get. 834 00:40:30.219 --> 00:40:32.360 Why it's called Lattice because it kind of makes a 835 00:40:32.500 --> 00:40:36.980 web of data that's hard to penetrate. Is that good analysis? 836 00:40:37.420 --> 00:40:38.039 No, it's terrible. 837 00:40:38.460 --> 00:40:41.659 Terrible. It's terrible. 838 00:40:42.139 --> 00:40:45.530 I can take it. Okay, here's how lattices... I claim stupidity. 839 00:40:45.550 --> 00:40:50.909 It's all good, mate. So, the way lattices work, and 840 00:40:50.969 --> 00:40:54.469 this is a terrible description, but it's an interesting way 841 00:40:54.530 --> 00:40:55.690 of thinking about it. 842 00:40:55.730 --> 00:40:57.329 It'll be better than mine, I guarantee it. 843 00:40:57.590 --> 00:40:59.530 I may not be. I'm not a fan of analogies, 844 00:40:59.570 --> 00:41:00.769 so here's an analogy for you. 845 00:41:00.789 -->

00:41:00.969 Okay. 846 00:41:03.719 --> 00:41:06.400 Imagine a chessboard, right? Eight by eight with a knight. 847 00:41:06.699 --> 00:41:08.739 We know the knight moves either one and two or 848 00:41:08.800 --> 00:41:09.539 two and one, right? 849 00:41:09.579 --> 00:41:10.219 That's all it does. 850 00:41:10.539 --> 00:41:12.900 Yeah. If I give you an end point and I 851 00:41:12.920 --> 00:41:15.599 give you a starting point, what route does the knight 852 00:41:15.639 --> 00:41:18.920 take to get there? That's pretty straightforward to do, you know, 853 00:41:18.940 --> 00:41:21.980 because it's just eight by eight. Now, imagine if that 854 00:41:22.019 --> 00:41:25.619 was a squillion by a squillion chessboard. Right. All right. 855 00:41:25.969 --> 00:41:26.300 Got it. 856 00:41:26.320 --> 00:41:29.130 Oh, no. You know, it gets harder. Now, imagine it's 857 00:41:29.369 --> 00:41:33.460 a thousand dimensions, right? Now imagine I don't tell you 858 00:41:33.480 --> 00:41:35.429 there's one by two, it's N by M. 859 00:41:35.760 --> 00:41:36.369 Right, okay. 860 00:41:36.650 --> 00:41:38.730 Now, just to make things even more difficult, it's not 861 00:41:38.769 --> 00:41:42.349 like an air quotes square chessboard. The squares are kind 862 00:41:42.409 --> 00:41:45.369 of funky shaped. They're not quite squares. And that's a 863 00:41:45.389 --> 00:41:51.199 thing called learning with errors. So that's hard. Here's a 864 00:41:51.280 --> 00:41:55.039 point somewhere in this N dimensional space. Here's your starting point. 865 00:41:55.079 --> 00:41:58.300 How do you get there? And so essentially the N 866 00:41:58.340 --> 00:42:00.599 by M is essentially like the private key. That's one 867 00:42:00.619 --> 00:42:02.780 way of looking at it. Terrible analogy, but it's about 868 00:42:02.820 --> 00:42:04.659 as close as you can get without introducing math. 869 00:42:04.719 --> 00:42:04.980 Okay. 870 00:42:05.440 --> 00:42:05.679 Yeah. 871 00:42:05.699 --> 00:42:09.070 All right. Good. And so you don't think that quantum 872 00:42:09.110 --> 00:42:12.389 computers could ever get so good that they could just 873 00:42:12.449 --> 00:42:13.809 figure that stuff out quickly? 874 00:42:14.110 --> 00:42:17.269 You mean when you throw in some AI as well? Yeah. Yeah. 875 00:42:17.670 --> 00:42:17.920 Go on. 876 00:42:19.280 --> 00:42:22.619 But nobody ever comes up and thinks, hey, we've solved cryptography. 877 00:42:22.940 --> 00:42:25.440 You're always looking at new algorithms. You're always looking at 878 00:42:25.500 --> 00:42:30.500 new key lengths. We expect to routinely replace our algorithms. 879 00:42:30.519 --> 00:42:31.340 It's an arms race. 880 00:42:31.360 --> 00:42:33.900 Because the bad guys are fighting back. 881 00:42:34.079 --> 00:42:37.590 Which is a beautiful segue into the next topic, which 882 00:42:37.639 --> 00:42:41.489 is crypto agility. If nothing else, from a developer perspective, 883 00:42:41.590 --> 00:42:46.230 one

thing that post-quantum crypto will bring to the table 884 00:42:46.690 --> 00:42:50.559 is the need for crypto agility. what happens if you 885 00:42:50.599 --> 00:42:55.070 wrap some keys in MLChem? Chem stands for key encapsulation method. 886 00:42:56.360 --> 00:42:59.269 Let's say you wrap some keys in that and it 887 00:42:59.289 --> 00:43:02.630 ends up being broken five years, 10 years from now. 888 00:43:03.090 --> 00:43:05.849 How can you update your application to use a new 889 00:43:05.889 --> 00:43:08.889 wrapping method without breaking your existing, like you can still 890 00:43:08.929 --> 00:43:11.739 read your old data, but you would write out using 891 00:43:11.780 --> 00:43:15.940 some MLChem + + or something. Right. And crypto agility 892 00:43:16.079 --> 00:43:19.719 is just, so important. It's really brought it to the 893 00:43:19.760 --> 00:43:24.260 forefront as a need, because we don't know long-term if 894 00:43:24.300 --> 00:43:27.420 these things will be successful or not. And again, to 895 00:43:27.440 --> 00:43:30.260 your point, Carl, cryptographers are very, very conservative when it 896 00:43:30.300 --> 00:43:32.340 comes to these sorts of things, and they want to 897 00:43:32.380 --> 00:43:36.179 have a big security buffer, knowing that some things will 898 00:43:36.219 --> 00:43:38.199 start to potentially creak a little bit. 899 00:43:39.630 --> 00:43:43.369 And of course, you're immediately going to the at-rest encryption problem. 900 00:43:43.409 --> 00:43:47.019 I'm always thinking TLS, and it's just we handshake an 901 00:43:47.059 --> 00:43:50.559 encrypted stream, we do our thing and then it disappears again. 902 00:43:50.599 --> 00:43:53.389 And so I don't have the, other than the, you know, 903 00:43:54.469 --> 00:43:57.090 harvest and decrypt later, I don't have to think about this. 904 00:43:57.150 --> 00:43:59.989 It's all transitory. We'll just use the newest algorithm. But 905 00:44:00.050 --> 00:44:03.809 if you've stored under an older algorithm, And now it's 906 00:44:03.849 --> 00:44:08.150 been breached. You have to decrypt, recrypt, or at least, 907 00:44:08.170 --> 00:44:10.849 you know, decrypt over time to get by that. 908 00:44:11.090 --> 00:44:13.510 You may have to increase the data size of your 909 00:44:13.570 --> 00:44:16.780 fields that take those things because it's going to take more. Yeah. 910 00:44:16.800 --> 00:44:18.500 So you've got to assume that there will be change. 911 00:44:18.639 --> 00:44:21.000 And unfortunately, a lot of people don't know how to 912 00:44:21.039 --> 00:44:24.179 build crypto agile solutions. It kind of drives me a 913 00:44:24.199 --> 00:44:25.880 bit bonkers, to be honest with you. I see, you know, 914 00:44:25.920 --> 00:44:28.159 in the press, you know, we need crypto agility. I've 915 00:44:28.199 --> 00:44:30.280 got to do crypto agility. Yeah. hey, crypto agility is 916 00:44:30.320 --> 00:44:32.500 really important. Are you guys doing crypto agility? 917 00:44:32.940 --> 00:44:35.300 Because it comes in a squirt bottle and you just 918 00:44:35.340 --> 00:44:37.159 spray it on all your devs and you'll be good. 919 00:44:37.360 --> 00:44:39.130 I know, but no one says how to do it. 920 00:44:41.150 --> 00:44:44.110 To me at Microsoft, there's two canonical examples of crypto agility. 921 00:44:44.210 --> 00:44:47.050 One is as a SQL DB or SQL server with 922 00:44:47.110 --> 00:44:50.889 always encrypted. And then the other one is the open 923 00:44:50.989 --> 00:44:54.730 office XML file format, which you use to encrypt documents 924 00:44:54.769 --> 00:44:57.449 in office. So the way it works in as a 925 00:44:57.469 --> 00:45:00.010 SQL DB, which to me is a beautiful and simple 926 00:45:00.050 --> 00:45:01.829 way of doing it. It's also very, very fast. If 927 00:45:02.550 --> 00:45:06.610 you ever look at the ciphertext in an always encrypted cell, 928 00:45:07.409 --> 00:45:09.789 the first byte is always a one, and that's the 929 00:45:09.829 --> 00:45:15.599 version number. And that basically means AES-256 cipherblockchaining with a 930 00:45:15.619 --> 00:45:21.960 SHA-256 hash as an integrity check over the data. So 931 00:45:21.980 --> 00:45:23.699 in the future, if they decide to upgrade it to 932 00:45:23.719 --> 00:45:26.679 something else, then that could be version two. So the 933 00:45:26.719 --> 00:45:28.960 first bike would be a two and they could always 934 00:45:29.079 --> 00:45:31.139 read back and say, okay, that's version one. We need 935 00:45:31.159 --> 00:45:34.460 this particular set of cipher primitives. Let's read as decrypted 936 00:45:34.500 --> 00:45:37.159 with these primitives. When they write it back, they would 937 00:45:37.179 --> 00:45:39.239 write it back as a, whatever the latest number was, 938 00:45:39.260 --> 00:45:41.460 which we version two, and it will be the, whatever 939 00:45:41.480 --> 00:45:43.179 the new, the new cipher suites is. So you can 940 00:45:43.199 --> 00:45:45.369 always read the old data and you would write back 941 00:45:45.630 --> 00:45:48.489 using the new version. And the way office does it 942 00:45:48.889 --> 00:45:51.550 is it's an XML file. There's an XML file header. 943 00:45:52.139 --> 00:45:55.639 And it contains all the cryptographic primitives, like AES, it's 944 00:45:55.659 --> 00:46:01.769 cipher blockchaining, here's the initialization vector, here's the password, the 945 00:46:01.789 --> 00:46:06.789 key derivation count, here's the key derivation algorithm, and lots 946 00:46:06.829 --> 00:46:09.829 of other metadata. So you can actually build the cipher 947 00:46:10.409 --> 00:46:14.929 collection completely dynamically, which is really, really nice. 948 00:46:16.619 --> 00:46:20.079 What network hardware will have to change in the, yes. 949 00:46:20.559 --> 00:46:22.900 I mean, we won't, we'll be able to go to 950 00:46:22.960 --> 00:46:25.179 Best Buy or whatever it is you have in the 951

00:46:25.280 --> 00:46:29.150 UK and buy an off the shelf, you know, router 952 00:46:29.230 --> 00:46:33.230 that is crypto happy. Oh, I mean, in theory, I 953 00:46:33.250 --> 00:46:34.710 think a lot of quantum happy rather. 954 00:46:34.730 --> 00:46:37.150 I mean, unless they're doing some kind of inspection, if 955 00:46:37.170 --> 00:46:39.050 they're just like passing data on, there should be no 956 00:46:39.130 --> 00:46:40.909 need for it, right? They're not decrypting stuff, but if 957 00:46:40.929 --> 00:46:44.019 they are decrypting stuff, then yeah, they're going to have 958 00:46:44.039 --> 00:46:46.239 to have access to these algorithms. If you're doing, you know, 959 00:46:46.460 --> 00:46:48.320 TLS termination, then yeah, for sure. 960 00:46:48.889 --> 00:46:50.769 The other that's not something you'll have in your home 961 00:46:50.969 --> 00:46:53.329 i did that stuff in racks of computers for companies 962 00:46:53.389 --> 00:46:54.070 but ah but. 963 00:46:54.010 --> 00:46:56.090 There's a fly in the ointment there's a huge fly 964 00:46:56.110 --> 00:46:57.989 in the ointment though and that is your laptop and 965 00:46:58.010 --> 00:47:00.679 your computer with tpms and trusted boots and that sort 966 00:47:00.699 --> 00:47:03.900 of stuff right right now those keys are probably rsa 967 00:47:03.920 --> 00:47:06.340 and well they are rsa or elliptic curve. 968 00:47:06.019 --> 00:47:08.139 They are and we're just going through the secure boot 969 00:47:08.179 --> 00:47:10.119 crisis thanks very much right. 970 00:47:09.900 --> 00:47:12.699 So that will all get that will all get busted 971 00:47:13.099 --> 00:47:15.980 and um you know so the hardware industry out there 972 00:47:16.429 --> 00:47:17.909 We're being mean here, Michael. 973 00:47:17.929 --> 00:47:20.139 Richard's squinting, but I think you've got to look at 974 00:47:20.179 --> 00:47:22.320 it like this. Hey, you've got to buy a new laptop. 975 00:47:22.960 --> 00:47:23.900 That's not a bad thing. 976 00:47:24.400 --> 00:47:26.800 But it's also like we're fixing keys all the time. 977 00:47:27.260 --> 00:47:30.800 These are BIOS updates. This will come in firmware. 978 00:47:31.000 --> 00:47:32.619 You think? There'll be new drivers. 979 00:47:34.179 --> 00:47:37.650 And they might be slower than a hardware-dedicated version of it, 980 00:47:37.760 --> 00:47:41.139 but I just don't feel– a lot of the stuff 981 00:47:41.159 --> 00:47:43.579 you're describing here, Michael, to me sounds like configuration settings 982 00:47:43.639 --> 00:47:46.610 in Azure. you've already done the work in the browser. Like, 983 00:47:46.619 --> 00:47:49.170 as long as I haven't done anything stupid in code, 984 00:47:49.630 --> 00:47:50.889 I don't think I have to do anything as a 985 00:47:50.929 --> 00:47:51.489 web dev. 986 00:47:51.849 --> 00:47:53.110 That's correct. 100% correct. 987 00:47:53.349 --> 00:47:54.809 I just got to make sure that my admins have 988 00:47:54.849 --> 00:47:55.530 set stuff right. 989 00:47:55.590 --> 00:47:57.449 Well, if you're using IIS, you're going to have to 990 00:47:57.510 --> 00:47:59.630 make sure you're in the latest version that supports it. 991 00:47:59.690 --> 00:48:00.730 But in Azure. 992 00:48:01.150 --> 00:48:04.050 Well, http.sys... So I actually wrote a blog post on 993 00:48:04.070 --> 00:48:08.929 this because about six-ish weeks ago, we released a version 994 00:48:08.949 --> 00:48:12.170 of S Channel, which is the Windows TLS stack that 995 00:48:12.190 --> 00:48:19.449 supports TLS 1.3 hybrid. And I wrote a dumb ASP.NET application, 996 00:48:19.469 --> 00:48:21.050 you know, dumber than a bucket of rocks just to 997 00:48:21.070 --> 00:48:25.340 keep it really, really simple and did no configuration whatsoever 998 00:48:25.380 --> 00:48:27.760 in the code. And that's a really important point. Like, 999 00:48:27.800 --> 00:48:31.099 you know, thou shalt not do any TLS configuration in code, 1000 00:48:31.159 --> 00:48:33.280 like leave it to the OS or to some configuration 1001 00:48:33.320 --> 00:48:36.659 somewhere else, definitely not in code. And yeah, I just 1002 00:48:36.699 --> 00:48:39.280 turned on that. I wanted X35519 MLChem768 in priority zero. 1003 00:48:43.219 --> 00:48:47.010 in the Cypher suite and group order. And I ran 1004 00:48:47.050 --> 00:48:49.949 this application and I connected it, connected using a browser 1005 00:48:50.010 --> 00:48:51.630 and I was living in the future. 1006 00:48:52.570 --> 00:48:52.829 Nice. 1007 00:48:53.090 --> 00:48:55.650 Michael, how long do we have before we need to 1008 00:48:56.329 --> 00:48:57.789 configure things correctly? 1009 00:48:58.010 --> 00:49:01.230 How long do we have? I mean, I mean, it 1010 00:49:01.269 --> 00:49:05.449 depends on risk. I was talking to a large retailer. 1011 00:49:06.309 --> 00:49:07.719 You know who they are, but I can't say who 1012 00:49:07.739 --> 00:49:07.969 they are. 1013 00:49:07.989 --> 00:49:08.340 Perfect. 1014 00:49:08.710 --> 00:49:09.610 And they're not concerned. 1015 00:49:09.650 --> 00:49:10.880 The one I bought my computer from? 1016 00:49:10.889 --> 00:49:15.619 They're not concerned about right now anyway with their devices 1017 00:49:15.739 --> 00:49:18.389 that they use for shop floor inventory management, right? Because 1018 00:49:18.429 --> 00:49:21.980 it's just shop floor inventory management. It's not sensitive data. 1019 00:49:22.519 --> 00:49:26.079 So their Android devices that they're using will never support post-quantum. 1020 00:49:26.239 --> 00:49:29.650 They're okay with that. Now, their corporate systems that run 1021 00:49:29.989 --> 00:49:32.969 HR and payroll and all that sort of stuff, yes, 1022 00:49:33.010 --> 00:49:35.710 they do care. So how long do we have? I mean, 1023 00:49:35.730 --> 00:49:38.989 the clock started now for certain types of data. Depends 1024 00:49:39.050 --> 00:49:42.010 on the data. If you've got

really sensitive data or 1025 00:49:42.070 --> 00:49:46.719 data that must be secret for a long time, healthcare information, 1026 00:49:46.820 --> 00:49:51.820 military secrets, intelligence, financial records in some cases, perhaps. I 1027 00:49:51.840 --> 00:49:56.079 don't know. I'm not a regulatory person. You know, they 1028 00:49:56.119 --> 00:50:00.869 have a time that they must maintain their secrecy. And 1029 00:50:00.909 --> 00:50:03.079 that clock's already started. Because if we take a 2029, 1030 00:50:03.079 --> 00:50:06.469 2030 timeframe, that's only four years. It's not even four 1031 00:50:06.510 --> 00:50:07.039 years away. Hmm. 1032 00:50:07.969 --> 00:50:10.400 On the screen behind you, a sentence came up a 1033 00:50:10.420 --> 00:50:14.389 little while ago. There's no such thing as low-risk data. 1034 00:50:14.929 --> 00:50:17.630 But apparently this large retailer seems to think there is. 1035 00:50:17.829 --> 00:50:20.420 I think it said no low-risk secrets. 1036 00:50:21.150 --> 00:50:23.010 Oh, low-risk secrets. That's right. Yeah. 1037 00:50:23.050 --> 00:50:24.610 Yeah. Low-risk secrets. 1038 00:50:25.070 --> 00:50:25.780 Secret is secret. 1039 00:50:25.800 --> 00:50:28.880 So, if it's a secret by default, it's by definition 1040 00:50:28.940 --> 00:50:29.219 a risk. 1041 00:50:29.559 --> 00:50:32.420 Well, a small, air quotes, small secret can lead to 1042 00:50:32.480 --> 00:50:36.440 access to bigger secrets. Like a low credential, for example, 1043 00:50:36.480 --> 00:50:37.860 it can lead to something like a key. 1044 00:50:38.320 --> 00:50:38.610 Sure. 1045 00:50:38.730 --> 00:50:38.929 Yeah. 1046 00:50:38.949 --> 00:50:42.670 The old classic, I got your Wi-Fi password from your 1047 00:50:42.750 --> 00:50:45.010 light bulb because you thought, well, it's just a light bulb. 1048 00:50:45.030 --> 00:50:46.289 What are you going to do to me? Right. 1049 00:50:46.530 --> 00:50:48.610 But the fact that I got chain attack, once I 1050 00:50:48.630 --> 00:50:50.269 got into the wifi, there was a whole lot of 1051 00:50:50.309 --> 00:50:51.219 other things that were there. 1052 00:50:51.280 --> 00:50:53.599 Yeah. And that's why I put all my IOT stuff 1053 00:50:53.619 --> 00:50:56.400 on his own virtual network is our own isolated. 1054 00:50:56.460 --> 00:50:57.619 Oh, you're darn right. You do. 1055 00:50:57.699 --> 00:50:58.739 Absolutely. Yeah. 1056 00:50:58.920 --> 00:50:59.099 Yeah. 1057 00:50:59.199 --> 00:51:02.139 But, uh, at the same time, you know, it's still, 1058 00:51:02.219 --> 00:51:05.219 I'm still sorting through what do I have to code 1059 00:51:05.360 --> 00:51:09.650 as a dev versus what I have to, uh, you know, 1060 00:51:09.710 --> 00:51:12.849 what's going to come to me, uh, just by making 1061 00:51:12.909 --> 00:51:15.869 sure we're using the latest bits. I love your info 1062 00:51:15.929 --> 00:51:18.570 on always encrypted. It's like, hey, now I want to 1063 00:51:18.590 --> 00:51:20.510 go

talk to my DBA. And it's like, we're up 1064 00:51:20.590 --> 00:51:22.690 on this version, right? Because we have all this encrypted 1065 00:51:22.730 --> 00:51:25.550 and REST stuff, and you don't want a crisis. So 1066 00:51:25.570 --> 00:51:30.050 if you're already running always encrypted in SQL, then we 1067 00:51:30.070 --> 00:51:33.469 should be good. When the new algorithms are needed, they'll work. 1068 00:51:33.989 --> 00:51:36.630 The big issue, and this is one thing that a 1069 00:51:36.690 --> 00:51:38.530 lot of products at Microsoft are having to deal with, 1070 00:51:38.710 --> 00:51:41.750 is the symmetric stuff's fine. It's the key wrapping that 1071 00:51:41.769 --> 00:51:44.010 has to change. The beauty of it, though, is it's 1072 00:51:44.030 --> 00:51:46.969 just the key wrapping. So if you have a 256-bit 1073 00:51:47.369 --> 00:51:51.110 AES key, you just decrypt it or unwrap it using RSA, 1074 00:51:51.150 --> 00:51:56.159 for example, and then you rewrap it using AES-KW. which 1075 00:51:56.360 --> 00:51:59.349 managed HSM in Azure, and now Key Vault, actually, Key 1076 00:51:59.369 --> 00:52:03.880 Vault Premium in public preview, supports AESKW key wrapping, which 1077 00:52:03.940 --> 00:52:04.860 is post-quantum resilient. 1078 00:52:05.139 --> 00:52:06.519 Yeah, go ahead. 1079 00:52:06.739 --> 00:52:09.619 So there's two issues that I see developers need to really, 1080 00:52:09.639 --> 00:52:12.800 really think about. The number one is please don't put 1081 00:52:12.900 --> 00:52:15.019 any TLS anything in your code. 1082 00:52:15.119 --> 00:52:16.619 Yeah, it'll hurt you later. 1083 00:52:16.860 --> 00:52:20.599 Don't restrict protocol version. Don't restrict cipher suites. Don't do 1084 00:52:20.739 --> 00:52:25.219 any of that. Just let it be configured completely outside 1085 00:52:25.260 --> 00:52:29.139 of the application. That's number one. Number two is this 1086 00:52:29.179 --> 00:52:35.289 whole crypto agility thing. If you've got crypto code with 1087 00:52:35.329 --> 00:52:37.570 the algorithms hard-coded in the code. 1088 00:52:37.590 --> 00:52:38.230 You're in trouble. 1089 00:52:38.909 --> 00:52:42.659 And you're encrypting squeaky bytes of data... You know, you've 1090 00:52:42.679 --> 00:52:44.280 got to update your code and probably can't read the 1091 00:52:44.360 --> 00:52:45.840 old data. So now you've got, you know, it's just 1092 00:52:45.900 --> 00:52:48.170 a mess. And that's where the whole crypto agility comes in. 1093 00:52:48.199 --> 00:52:52.179 And honestly, if you haven't got crypto agility in place, 1094 00:52:52.199 --> 00:52:57.179 there are patterns you can use to wrap existing non-crypto, 1095 00:52:57.309 --> 00:53:01.210 crypto agile blobs into like some sort of crypto agile 1096 00:53:01.289 --> 00:53:02.949 envelope that contains all the metadata. 1097 00:53:03.150 --> 00:53:06.670 So Azure's really good about letting me know when I 1098 00:53:06.710 --> 00:53:09.989 need

to move off of some version of something and 1099 00:53:10.070 --> 00:53:13.119 onto something else because it's being deprecated. Do you think 1100 00:53:13.159 --> 00:53:15.900 that sometime in the future, we're going to get an 1101 00:53:15.980 --> 00:53:18.800 Azure thing when we log into the portal that says, hey, 1102 00:53:19.519 --> 00:53:23.480 you need to upgrade your whatever it is to be 1103 00:53:23.599 --> 00:53:28.110 quantum happy. You think that's going to happen? Like, should 1104 00:53:28.119 --> 00:53:30.219 I just leave it up to Azure to warn me 1105 00:53:30.260 --> 00:53:33.130 about things like that? Or is there stuff that I 1106 00:53:33.170 --> 00:53:33.869 need to do now? 1107 00:53:34.329 --> 00:53:36.650 I think it depends. If you look at the shared 1108 00:53:36.690 --> 00:53:40.969 responsibility model, that's where it really becomes important. Like if 1109 00:53:41.010 --> 00:53:43.300 you've got a platform managed key to encrypt data, then 1110 00:53:43.320 --> 00:53:45.139 we're going to take care of that, right? Because a 1111 00:53:45.159 --> 00:53:48.079 platform managed key. But if you've got a customer managed key, 1112 00:53:48.099 --> 00:53:52.340 which is basically a key wrapping key, then they'll probably 1113 00:53:52.360 --> 00:53:54.699 look at, I can't predict the future, but my guess, 1114 00:53:54.860 --> 00:53:59.070 just my guess, Is that there will be notifications to say, hey, 1115 00:53:59.300 --> 00:54:03.360 you know, we've now got the ability in Azure, blah, blah, blah, 1116 00:54:03.980 --> 00:54:08.099 to wrap your encryption keys in quantum resilient keys. Would 1117 00:54:08.119 --> 00:54:11.239 you like to do this? Yeah. And click here. Yeah. 1118 00:54:11.360 --> 00:54:14.599 And the consequence is going to be very likely that 1119 00:54:14.659 --> 00:54:17.539 the data streams are going to get larger because the 1120 00:54:18.400 --> 00:54:21.389 quantum resilient keys are bigger. But I don't know what 1121 00:54:21.429 --> 00:54:22.570 other impacts I'm going to see. 1122 00:54:22.670 --> 00:54:24.670 No, you shouldn't see it. No, because the only thing 1123 00:54:24.690 --> 00:54:27.409 you're really changing is the key wrapping. Right. Which is, 1124 00:54:27.929 --> 00:54:29.219 so if you've got a 256-bit AES key, you're going 1125 00:54:29.239 --> 00:54:35.849 to wrap it in AES-KW. And in fact, it'll probably 1126 00:54:35.869 --> 00:54:38.099 be smaller than RSA, to be honest with you. But 1127 00:54:38.119 --> 00:54:39.880 if you wrap it in MLChem, it will be bigger. 1128 00:54:39.900 --> 00:54:40.360 Right. 1129 00:54:40.380 --> 00:54:41.739 Yeah, but it's just the key. 1130 00:54:41.940 --> 00:54:42.559 It's just the key. 1131 00:54:42.699 --> 00:54:43.719 Yeah, the data doesn't change. 1132 00:54:43.780 --> 00:54:46.340 Am I even choosing that or I'm just configuring to 1133 00:54:46.400 --> 00:54:48.269 allow this and it'll optimize itself? 1134 00:54:49.110 --> 00:54:52.250 No, what do you mean? When do I have to 1135 00:54:52.289 --> 00:54:57.199 make a decision about this, about the rewrap? Well, first 1136 00:54:57.239 --> 00:54:59.679 of all, the nice thing is it's very low friction. Right. Very, 1137 00:54:59.739 --> 00:55:02.500 very low friction. Like it's microseconds to do the work. 1138 00:55:02.940 --> 00:55:06.550 You're not decrypting and re-encrypting petabytes of data. 1139 00:55:07.070 --> 00:55:07.619 Right. 1140 00:55:07.949 --> 00:55:11.650 So my guess is personally, as soon as it becomes available, 1141 00:55:11.829 --> 00:55:15.550 I would just opt in and just re-wrap your keys. 1142 00:55:15.829 --> 00:55:18.030 So I'm going to wait for your blog post and 1143 00:55:18.070 --> 00:55:19.690 then I'm just going to switch this stuff on. 1144 00:55:19.809 --> 00:55:21.489 No, we'll push it through Azure Update. 1145 00:55:21.630 --> 00:55:21.750 Okay. 1146 00:55:21.909 --> 00:55:23.869 I guess it'll be through the Azure Updates website. 1147 00:55:24.010 --> 00:55:24.170 Yeah. 1148 00:55:24.190 --> 00:55:24.820 Yeah. Yeah. 1149 00:55:24.860 --> 00:55:27.659 Good. And is this imminent like in the next year? 1150 00:55:27.679 --> 00:55:27.760 Yeah. 1151 00:55:28.239 --> 00:55:28.400 Wow. 1152 00:55:28.440 --> 00:55:31.360 It'll depend on the service. Yeah. So, so, so I 1153 00:55:31.360 --> 00:55:33.639 really want to point something out here is that, you know, 1154 00:55:33.659 --> 00:55:36.000 this post-quantum stuff is very layered, right? So the very 1155 00:55:36.039 --> 00:55:37.619 bottom of the layer, you've got the algorithms. Well, that 1156 00:55:37.639 --> 00:55:39.280 stuff's been in place in windows for. 1157 00:55:39.300 --> 00:55:39.900 For ages. 1158 00:55:40.219 --> 00:55:43.389 Um, we have called sim crypt available in windows, Linux 1159 00:55:43.429 --> 00:55:47.489 and Mac hand optimized C code. Absolutely beautiful to read. 1160 00:55:47.550 --> 00:55:50.670 It's so well written. Honestly, it really is. Then above that, 1161 00:55:50.730 --> 00:55:55.210 you've got, um, say TLS 1.3 with hybrid, right? You 1162 00:55:55.269 --> 00:55:57.710 open those floodgates because now people can use that stack. 1163 00:55:58.530 --> 00:56:00.429 So basically on windows is to use the new S 1164 00:56:00.449 --> 00:56:02.489 channel stack. And in the case of Linux use open 1165 00:56:02.590 --> 00:56:05.690 SSL 3.5, um, which has ML chem built into it. 1166 00:56:05.730 --> 00:56:08.690 So you've got those two options are now available to you. Um, 1167 00:56:08.889 --> 00:56:10.789 so that's, that's the data in transit taken care of. 1168 00:56:11.309 --> 00:56:13.489 And then the last one, which is incredibly important is 1169 00:56:13.510 --> 00:56:17.639 the key wrapping. And we now have that in, in Azure. 1170 00:56:17.679 --> 00:56:22.420 So managed HSM has had a key wrapping since day one. And, um, 1171 00:56:22.969 --> 00:56:25.369 as your Key Vault now has it, ASKW. So

what's 1172 00:56:25.389 --> 00:56:27.369 going to happen is that's going to open these floodgates 1173 00:56:28.090 --> 00:56:34.119 and there'll be a Cambrian explosion of services coming online 1174 00:56:34.179 --> 00:56:37.460 right as they adopt. And look, it won't happen like overnight, 1175 00:56:37.639 --> 00:56:40.079 because everyone's going to do testing, make sure it works, 1176 00:56:40.159 --> 00:56:43.010 make sure it fails correctly and that sort of stuff. 1177 00:56:43.550 --> 00:56:46.869 But Richard, to your point, next year, We'll see a 1178 00:56:46.909 --> 00:56:48.590 lot of services rolling this out. 1179 00:56:48.610 --> 00:56:49.630 Starting to switch over. 1180 00:56:49.869 --> 00:56:52.349 But I also remember when we started switching up keys, 1181 00:56:52.389 --> 00:56:54.739 when attacks got smarter and so forth, there was a 1182 00:56:54.820 --> 00:57:00.199 period where we changed a number of times different flavors of, yes, SSL, 1183 00:57:00.280 --> 00:57:04.639 and then into TLS. We are restarting this in some 1184 00:57:04.659 --> 00:57:07.809 respects with these new cryptography. I've got to think. The 1185 00:57:07.849 --> 00:57:10.710 first move we make may not stick for more than 1186 00:57:10.750 --> 00:57:12.809 a year or two before it's like, hey, now we 1187 00:57:12.829 --> 00:57:15.090 found some problems and you probably want to switch to this. 1188 00:57:15.889 --> 00:57:18.469 I just remember going through this with AAS and a 1189 00:57:18.510 --> 00:57:21.969 few others. Like, we've done this before. In some ways, 1190 00:57:21.989 --> 00:57:25.230 we've gotten really lazy because it's worked so well for 1191 00:57:25.349 --> 00:57:26.010 so long. 1192 00:57:27.150 --> 00:57:30.469 Well, actually, it's worse than that. And that is that, 1193 00:57:31.050 --> 00:57:33.699 so we've actually become really lazy because of RSA. 1194 00:57:33.969 --> 00:57:34.199 Right. 1195 00:57:34.579 --> 00:57:39.300 RSA is interesting. RSA can do all the crypto primitives. Right. 1196 00:57:39.460 --> 00:57:41.460 It can do signing, it can do encryption and all 1197 00:57:41.480 --> 00:57:44.679 that sort of stuff. Elliptic Curve and Diffie-Hellman cannot. They 1198 00:57:44.699 --> 00:57:47.460 can't do everything. And so we've been used to, like 1199 00:57:47.500 --> 00:57:49.980 you said before, Richard, I'm not trying to laugh at you, 1200 00:57:50.000 --> 00:57:51.420 but he said, you know, prime numbers. Well, that's just 1201 00:57:51.460 --> 00:57:53.860 an RSA thing. That's not an elliptic curve thing. Right. 1202 00:57:53.940 --> 00:57:55.699 Because everyone thinks of RSA. 1203 00:57:55.860 --> 00:57:56.079 Yeah. 1204 00:57:56.159 --> 00:57:58.980 And RSA is this Swiss army knife. It does absolutely everything. 1205 00:57:59.159 --> 00:57:59.380 Sure. 1206 00:57:59.480 --> 00:58:02.070 And the problem is it does absolutely everything. 1207 00:58:02.159 --> 00:58:02.909 Absolutely everything. 1208 00:58:02.969 --> 00:58:05.809 Which means we've got to change it absolutely everywhere. Yeah. 1209 00:58:06.949 --> 00:58:10.269 It went everywhere. Well, the other side of this was 1210 00:58:10.289 --> 00:58:14.269 because compute, I remember when it was expensive to do encryption, 1211 00:58:14.309 --> 00:58:18.210 where we literally had separate servers from the website for 1212 00:58:18.269 --> 00:58:21.429 doing just the encrypted pages. It's okay, well, now you're 1213 00:58:21.449 --> 00:58:23.809 going to take your shopping cart and start a payment 1214 00:58:23.849 --> 00:58:27.400 cycle that needs to be asked to sell. Now, it 1215 00:58:27.440 --> 00:58:29.480 was this big thing about moving the cart over to 1216 00:58:29.519 --> 00:58:34.000 the other much more expensive dedicated stack for completing a transaction. 1217 00:58:34.659 --> 00:58:37.980 Because encryption used to be so costly. You know, these 1218 00:58:38.019 --> 00:58:40.099 days our CPUs are so damn fast. They're sitting around 1219 00:58:40.119 --> 00:58:42.679 playing poker and smoking cigarettes waiting for something to do. 1220 00:58:43.380 --> 00:58:46.780 So sure, encrypt everything. Who cares? RSA is fast. 1221 00:58:46.969 --> 00:58:49.889 Well, the funny thing is that first of all, it's 1222 00:58:49.929 --> 00:58:51.769 just the key wrapping and unwrapping part. 1223 00:58:51.869 --> 00:58:52.670 Yeah. 1224 00:58:52.710 --> 00:58:56.050 Let's just call it key establishment because there's all different 1225 00:58:56.070 --> 00:58:56.570 ways of doing it. 1226 00:58:56.570 --> 00:58:57.550 That's really what's going on. 1227 00:58:57.630 --> 00:59:01.230 That's the expensive part because it's all asymmetric stuff. And Windows, 1228 00:59:01.369 --> 00:59:03.159 I think it still exists. There used to be a 1229 00:59:03.260 --> 00:59:09.519 registry key called modexp offload for modular exponentiation offload, which 1230 00:59:09.579 --> 00:59:14.019 is a very expensive RSA operation. And there was a 1231 00:59:14.039 --> 00:59:16.739 way you could actually set a DLL in there, a 1232 00:59:16.760 --> 00:59:19.139 name for DLL. It would actually offload that work to 1233 00:59:19.199 --> 00:59:21.420 a DLL, which was a shim into some hardware to 1234 00:59:21.440 --> 00:59:24.860 actually do the modular exponentiation work. But we don't need 1235 00:59:24.929 --> 00:59:28.190 any of that stuff anymore. You know, when I look 1236 00:59:28.210 --> 00:59:32.190 at the work that... you know, as your front door 1237 00:59:32.210 --> 00:59:33.829 have done, for example, they've done a whole bunch of 1238 00:59:33.889 --> 00:59:39.579 analysis on performance and the performance is like just a 1239 00:59:39.659 --> 00:59:42.519 couple of percent, you know, going from elliptic curve to 1240 00:59:42.539 --> 00:59:46.980 elliptic curve plus ML, ML chem. And that's because of 1241 00:59:47.619 --> 00:59:48.960 optimizations made in the library. 1242 00:59:49.260 -->

00:59:49.449 Sure. 1243 00:59:50.119 --> 00:59:52.019 I got to tell you, Michael, there's this weird dichotomy 1244 00:59:52.059 --> 00:59:55.349 going on where it's like the catastrophizing of quantum destroying 1245 00:59:55.409 --> 00:59:58.969 everything and the, oh, just touch this button, problem goes away. 1246 00:59:59.130 --> 01:00:02.769 Yeah. Yeah. I'm feeling that too. You're right. 1247 01:00:03.710 --> 01:00:05.010 I wish it was that simple. 1248 01:00:05.309 --> 01:00:07.860 Yeah. I keep waiting for what's not the simple part. 1249 01:00:08.230 --> 01:00:10.449 I mean, for me, it's the frontline dev. 1250 01:00:10.510 --> 01:00:10.949 Right. 1251 01:00:11.070 --> 01:00:14.909 I mean, the administrator in me is a little sticky 1252 01:00:14.949 --> 01:00:17.369 in the shorts right now because this is all very scary. Like, 1253 01:00:17.429 --> 01:00:19.469 I want to check everything. I don't want to be 1254 01:00:19.510 --> 01:00:21.139 the guy who didn't do his homework. 1255 01:00:21.179 --> 01:00:21.389 Right. 1256 01:00:21.820 --> 01:00:24.420 But for the dev, unless you've done something dumb, I 1257 01:00:24.599 --> 01:00:27.800 think you're good as long as your administrators are on it. 1258 01:00:27.860 --> 01:00:30.460 As I mentioned before, the two big dumb things are 1259 01:00:30.980 --> 01:00:34.500 baking in crypto algorithms into your code and not being crypto-natural. 1260 01:00:34.699 --> 01:00:38.039 That's dumb thing number one. Dumb thing number two is 1261 01:00:38.099 --> 01:00:40.019 if you hard code your TLS configuration. 1262 01:00:40.820 --> 01:00:40.940 Right. 1263 01:00:40.980 --> 01:00:43.840 You're right. Those are the two big ones. Let's just 1264 01:00:43.880 --> 01:00:48.530 ignore compatibility for a moment. If you change a server 1265 01:00:48.550 --> 01:00:52.030 to use TLS 1.3 hybrid and only hybrid, by the way, 1266 01:00:52.130 --> 01:00:53.989 the reason why it's called hybrid is because you do 1267 01:00:54.369 --> 01:00:57.659 elliptic curve and MLChem together. The keys are derived from both. 1268 01:00:58.289 --> 01:01:00.630 That's why it's called hybrid. But if you do hybrid 1269 01:01:00.650 --> 01:01:02.769 and the client doesn't talk hybrid for whatever, you got 1270 01:01:02.789 --> 01:01:07.199 like a crusty old Java application or C sharp application 1271 01:01:07.219 --> 01:01:10.480 written 15 years ago, it's probably not going to work. 1272 01:01:10.820 --> 01:01:11.039 Yeah. 1273 01:01:11.199 --> 01:01:12.360 And it's really a question of, is it going to 1274 01:01:12.460 --> 01:01:14.559 fail with some grace to tell you what the hell's 1275 01:01:14.599 --> 01:01:15.159 going on? 1276 01:01:15.199 --> 01:01:15.400 Yeah. 1277 01:01:15.739 --> 01:01:16.829 Right. Yeah. 1278 01:01:17.050 --> 01:01:19.949 It says, don't understand this cipher suite and gives you 1279 01:01:19.989 --> 01:01:22.190 a hex value. Yeah. Really useful. 1280 01:01:22.389 --> 01:01:22.630 Yeah. 1281 01:01:22.849 --> 01:01:25.570 Well, that's better than object not found. 1282 01:01:27.289 --> 01:01:29.090 One of the best ones in office back in the 1283 01:01:29.110 --> 01:01:29.829 day was out of memory. 1284 01:01:30.190 --> 01:01:32.820 Yeah. Michael, tell us about your book. 1285 01:01:33.070 --> 01:01:33.909 Oh, my new book, man. 1286 01:01:34.119 --> 01:01:34.340 Yeah. 1287 01:01:34.360 --> 01:01:37.820 So, I wrote this book with Sean Hernan, Lee Holmes, 1288 01:01:37.880 --> 01:01:39.980 and Sherry DeGrippo. So, Sean and I have known each 1289 01:01:40.000 --> 01:01:44.139 other for many, many years. He's a partner engineering manager 1290 01:01:44.239 --> 01:01:46.309 in Azure Security. I've been around for a long time, 1291 01:01:46.349 --> 01:01:50.309 got the utmost respect for Sean, great guy. Lee Holmes, 1292 01:01:50.349 --> 01:01:52.070 he was the security guy in PowerShell. 1293 01:01:52.289 --> 01:01:53.650 Yeah, he's been on the show before too. 1294 01:01:53.769 --> 01:01:55.349 Yeah, my boy Lee, great guy. 1295 01:01:55.429 --> 01:01:56.030 Yeah, he's a good man. 1296 01:01:56.190 --> 01:01:58.670 He and I actually worked together in the earliest days 1297 01:01:58.710 --> 01:02:02.909 of the SDL because when Monad, as it was back 1298 01:02:02.929 --> 01:02:05.150 in the day, had to go through all its SDL checks, 1299 01:02:05.750 --> 01:02:06.510 I was the. 1300 01:02:07.050 --> 01:02:08.989 Precursor to PowerShell? 1301 01:02:09.030 --> 01:02:09.929 To PowerShell, correct. 1302 01:02:10.130 --> 01:02:10.809 Yeah. 1303 01:02:11.030 --> 01:02:13.920 I was like the SDL contact for Monad. And so 1304 01:02:13.980 --> 01:02:15.579 Lee and I got to know each other incredibly well. 1305 01:02:16.500 --> 01:02:20.260 And he's also a partner engineer in Azure. And then 1306 01:02:20.360 --> 01:02:22.920 Sherrod DeGrippo, she's actually left Microsoft now. She goes to 1307 01:02:22.940 --> 01:02:28.469 Palo Alto Labs. She is easily one of the preeminent 1308 01:02:28.510 --> 01:02:32.349 experts in the world on threat actors. She knows absolutely 1309 01:02:32.369 --> 01:02:35.429 everything about threat actors. So the book is Threat-Driven Software Development. 1310 01:02:35.989 --> 01:02:38.389 And basically what it is is looking at software development 1311 01:02:38.409 --> 01:02:41.030 through the eyes of threat actors, like what do threat 1312 01:02:41.070 --> 01:02:41.820 actors actually do? 1313 01:02:41.840 --> 01:02:41.960 Mm-hmm. 1314 01:02:42.550 --> 01:02:45.420 And every chapter starts off... So first of all, Sherrod 1315 01:02:45.440 --> 01:02:47.880 has her own chapter at the beginning. But every chapter 1316 01:02:47.980 --> 01:02:51.480 after that looks at the contents of that chapter through 1317 01:02:51.500 --> 01:02:55.679 the lens of threat intel. So every chapter starts off 1318 01:02:55.719 --> 01:02:59.840 with anywhere between half and two pages of threat intel perspective, 1319 01:02:59.880 -->

01:03:02.500 where Sherrod gives it a whole bunch of color. And 1320 01:03:02.539 --> 01:03:06.969 then Lee, myself, and Sean go through and sort of 1321 01:03:07.030 --> 01:03:10.489 explain that particular topic in detail. And a lot of it's... 1322 01:03:11.579 --> 01:03:14.099 A lot of it's not just, don't think of it 1323 01:03:14.159 --> 01:03:15.559 like just security best practices. 1324 01:03:15.739 --> 01:03:16.159 It's not. 1325 01:03:16.860 --> 01:03:23.079 It's DevOps as well as it's operational security, AppSec, and 1326 01:03:23.119 --> 01:03:26.139 also Threat Intel all sort of munched together into one book. 1327 01:03:26.400 --> 01:03:26.880 Sounds good. 1328 01:03:27.280 --> 01:03:28.340 Yeah, a lot of fun. I did one when I 1329 01:03:28.340 --> 01:03:31.559 was in the red team. Funny thing is, so Mark 1330 01:03:31.599 --> 01:03:34.719 Rasinovich wrote the forward. And my manager at the time, 1331 01:03:35.809 --> 01:03:38.269 Craig Nelson, who's CVP of the red team, he said, 1332 01:03:38.309 --> 01:03:40.110 oh man, I really love this book. I gave him 1333 01:03:40.130 --> 01:03:44.530 draft to look at. Can I write a chapter? I'm like, well, 1334 01:03:44.550 --> 01:03:45.889 why don't you write the afterword? I said, I've never 1335 01:03:45.909 --> 01:03:47.090 had an afterword in a book. Why don't you write 1336 01:03:47.110 --> 01:03:53.000 the afterword? So he did. Nine pages. But the afterword... 1337 01:03:53.239 --> 01:03:56.739 The after chapter. Yeah, the after chapter. Look, I'm not 1338 01:03:56.760 --> 01:04:02.940 trying to besmirch Craig at all. That afterword is absolutely brilliant. 1339 01:04:03.420 --> 01:04:04.840 If you were to sum up the afterword in like 1340 01:04:05.380 --> 01:04:08.519 one sentence or two words, it would be So what? 1341 01:04:09.380 --> 01:04:12.300 And he does a really, really good job of answering. 1342 01:04:12.340 --> 01:04:14.199 So what? Yeah. It's really good. Yeah. 1343 01:04:14.219 --> 01:04:16.750 That's good. It's really cool. Michael, what can we say? 1344 01:04:16.789 --> 01:04:20.829 It's been enlightening having you here and talking about all 1345 01:04:20.869 --> 01:04:23.570 this crazy stuff that we barely understand. Well, I barely 1346 01:04:23.610 --> 01:04:26.230 understand anyway, but I understand a little more thanks to you. 1347 01:04:26.269 --> 01:04:27.230 So thank you very much. 1348 01:04:27.449 --> 01:04:28.889 You're welcome. Thanks for having me on. 1349 01:04:28.929 --> 01:04:30.880 Great show, friend. Thank you so much. And we'll talk 1350 01:04:30.889 --> 01:04:33.119 to you next time on. NET Rocks!. NET Rocks! 1351 01:04:54.269 --> 01:04:57.030 NET Rocks is brought to you by Franklin's Net and 1352 01:04:57.090 --> 01:05:01.809 produced by Plop Studios, a full-service audio, video, and post-production 1353 01:05:01.869 --> 01:05:05.849 facility located physically in New London, Connecticut, and, of course, 1354 01:05:05.909 --> 01:05:14.050 in the cloud, online at pwop.com. Visit our website at dotnetrocks.com 1355 01:05:14.489 --> 01:05:19.190 for RSS feeds, downloads, mobile apps, comments, and access to 1356 01:05:19.210 --> 01:05:22.590 the full archives going back to show number one, recorded 1357 01:05:22.610 --> 01:05:23.829 in September 2002. 1358 01:05:23.829 --> 01:05:27.139 And make sure you check out our sponsors. They keep 1359 01:05:27.219 --> 01:05:30.400 us in business. Now go write some code. See you 1360 01:05:30.420 --> 01:05:30.840 next time.