00:00:00.000 --> 00:00:00.320
[MUSIC]

00:00:00.320 --> 00:00:05.200
MIKE WALKER: Well, you have to understand the rules 
if you want to be able to do what you want. So  

00:00:05.200 --> 00:00:09.120
I have, I have a definite fascination with 
the rules, [LAUGHTER] but the thing you have  

00:00:09.120 --> 00:00:16.640
to understand about the rules is that rules are 
a social construct. And I think that a lot of …  

00:00:16.640 --> 00:00:25.600
in the computer security field, when, when people 
hear the word “security” and they think about the  

00:00:25.600 --> 00:00:33.280
rules that the security system is imposing 
on them, every time a security system asks  

00:00:33.280 --> 00:00:39.280
something of users, then you're asking for their 
trust. And that trust is a precious commodity.  

00:00:39.280 --> 00:00:46.320
And so when you squander user trust, then the 
security regime falls down. So I think it's really  

00:00:46.320 --> 00:00:55.280
important to understand the social context of 
rules instead of just having instant adjudication.

00:00:55.280 --> 00:00:56.037
[THEME MUSIC]

00:00:56.037 --> 00:00:59.440
STANDARD INTRODUCTION: This is 
the Microsoft Research Podcast,  

00:00:59.440 --> 00:01:03.040
where Microsoft researchers—driving 
advancement through fundamental science  

00:01:03.040 --> 00:01:10.702
and technology research—explore the who, 
how, and what’s next in computing and AI.    

00:01:10.702 --> 00:01:14.080
[MUSIC ENDS]
WEISHUNG LIU: Hi, and welcome. My name is Weishung Liu, and

00:01:14.080 --> 00:01:18.240
I’m a principal technical program manager here 
at Microsoft Research. And today, I’m joined  

00:01:19.040 --> 00:01:23.120
by Mike Walker, who is a research manager 
here in MSR, or Microsoft Research,  

00:01:23.120 --> 00:01:28.880
where he leads projects at the intersection of 
AI, cybersecurity, and large distributed systems.

00:01:28.880 --> 00:01:33.840
Before Microsoft, Mike led DARPA’s Cyber 
Grand Challenge, the pioneering program  

00:01:33.840 --> 00:01:39.360
that demonstrated autonomous AI systems capable 
of finding and patching software vulnerabilities.  

00:01:39.360 --> 00:01:44.480
And CGC actually just celebrated its 
10th anniversary this year in 2026.

00:01:44.480 --> 00:01:50.400
Mike's work spans groundbreaking efforts 
that include cloud native memory analysis,  

00:01:50.400 --> 00:01:54.720
automated malware analysis, and 
next-generation cyber defense technologies.

00:01:54.720 --> 00:01:55.960
Mike, thanks for joining me.

00:01:55.960 --> 00:01:57.275
MIKE WALKER: Good to be here, Wei.

00:01:57.275 --> 00:02:02.480
LIU: All right, so, Mike, we have known each 
other for a couple of years now. And I've heard,  

00:02:02.480 --> 00:02:06.800
like, bits and pieces of your story over 
time as we've gotten to talk. But before  

00:02:06.800 --> 00:02:10.880
I really knew who you were, sort of, 
as a person, I, kind of, knew you as,  

00:02:10.880 --> 00:02:14.960
like, “Mike Walker, the hacker 
guy that's working over there,”  

00:02:14.960 --> 00:02:18.400
which I'm sure you love. And I know you 
love that, when I call you “The Hacker.”

00:02:19.360 --> 00:02:22.320
But you've worked on so many things that 
I've never really thought about before.  

00:02:22.880 --> 00:02:27.840
Like, new words have entered my vocabulary as 
a result of knowing you and working with you,  

00:02:27.840 --> 00:02:34.240
like threat intelligence, nation states, and 
I think there are a couple other ones that  

00:02:34.240 --> 00:02:37.920
I think would have never been in 
my lexicon had I not known you.

00:02:37.920 --> 00:02:42.480
So I'm hoping we can spend this time to 
also have the audience get to know you,  

00:02:42.480 --> 00:02:48.000
as well. So we're going to start at the 
beginning then. And the beginning is,  

00:02:48.000 --> 00:02:50.280
I believe, Mike, you grew up in Ohio.

00:02:50.280 --> 00:02:51.360
WALKER: Columbus, Ohio.

00:02:51.360 --> 00:02:56.320
LIU: Columbus, Ohio. What kind of 
a kid were you in Columbus, Ohio?

00:02:56.320 --> 00:03:00.800
WALKER: I think I was a kid who spent 
a lot of time in the woods. [LAUGHTER]  

00:03:00.800 --> 00:03:04.160
I was a skateboard kid. I was a 
basketball kid. But I think mostly  

00:03:04.160 --> 00:03:08.560
I was a kid who read books and lived 
in a world of their own imagination.

00:03:08.560 --> 00:03:11.600
LIU: Were you on a farm in Columbus,  

00:03:11.600 --> 00:03:16.720
Ohio, or were you in, like, or is 
this more suburban, rural, urban?

00:03:16.720 --> 00:03:22.640
WALKER: It was halves and halves. I spent 
my … most of the year in Columbus, Ohio,  

00:03:22.640 --> 00:03:26.480
and then every summer, I would go 
out to the town of Tappahannock on  

00:03:26.480 --> 00:03:30.760
the Rappahannock River in Virginia. 
And I lived completely on the farm.

00:03:30.760 --> 00:03:35.360
LIU: So what was the lifestyle in 
what you were doing different between,  

00:03:35.360 --> 00:03:38.000
like, Columbus, Ohio, and the farm?

00:03:38.000 --> 00:03:44.720
WALKER: So, like, some things that I was doing 
in Columbus, Ohio, is that my father, who was  

00:03:44.720 --> 00:03:50.560
sort of half scientist, half venture capitalist, 
would bring me home industrial equipment and say,  

00:03:50.560 --> 00:03:55.120
“See what you can make out of this,” and drop it 
in my room. [LAUGHTER] Really dangerous stuff. And  

00:03:55.920 --> 00:04:00.400
he would always give me safety warnings, right, 
like, “This is a helium neon laser." You know,  

00:04:00.400 --> 00:04:06.400
“This is … this tube is made out of glass. 
Don't break it.” And then, “This is a really,  

00:04:06.400 --> 00:04:11.680
really high-voltage power supply driving it, and 
it can arc and kill you,” right. And I knew what  

00:04:11.680 --> 00:04:15.600
that meant. [LAUGHTER] And then he would leave me 
with it. And what can you make out of this, right.

00:04:15.600 --> 00:04:20.400
So I would do stuff like, like try to 
replicate the Michelson-Morley experiment,  

00:04:20.400 --> 00:04:23.600
you know, because I had the, had the 
table and the mirrors and I could do that.  

00:04:25.120 --> 00:04:31.920
And then also I spent a lot of time with 
the Apple II, which was this just incredible  

00:04:31.920 --> 00:04:38.720
machine from some other universe that I had 
never gone to. I later found out that was,  

00:04:38.720 --> 00:04:43.120
you know, hackerspace and, and 
information technology culture,  

00:04:43.120 --> 00:04:47.840
but it came from nowhere and there was nobody I 
could talk to about it. And so I spent a lot of  

00:04:47.840 --> 00:04:54.720
time trying to hook the Apple II up to stuff in my 
life and, and try to control it with that thing.

00:04:54.720 --> 00:04:56.960
LIU: OK, so you brought up some hacker stuff,  

00:04:56.960 --> 00:05:03.520
so I have to ask. There was farm stuff. There 
was, you know, “dad brought home interesting  

00:05:04.240 --> 00:05:09.360
tools and equipment.” How did all of 
that get you interested in cybersecurity?

00:05:09.360 --> 00:05:15.200
WALKER: I think I was always in 
cybersecurity. I just didn't know.

00:05:15.200 --> 00:05:15.920
LIU: Explain.

00:05:15.920 --> 00:05:20.400
WALKER: You know, there's this 
idea that you live two lives:  

00:05:20.400 --> 00:05:25.680
the one before you find out your purpose and 
the one after you find it. This was, sort of,  

00:05:25.680 --> 00:05:31.120
life before purpose, where I didn't know why I 
was so interested in how these things worked.

00:05:31.120 --> 00:05:38.960
So, you know, I ran a bulletin board system. 
Before the internet, this was how computers  

00:05:38.960 --> 00:05:46.000
communicated with each other. Dial-up modems, 
not even backbone linked. Each, each BBS—bulletin  

00:05:46.000 --> 00:05:51.520
board system—[LAUGHTER] was, was a universe unto 
itself, you know, architected by its creator.  

00:05:51.520 --> 00:05:59.280
And so there were all these little universes 
you could dial into. And then at some point,  

00:05:59.280 --> 00:06:04.800
all these things I'd been doing my whole 
life, sort of, got connected to a career.

00:06:04.800 --> 00:06:10.240
When you say, you know, somebody could, right, 
somebody could listen to every single badge  

00:06:10.240 --> 00:06:16.880
swipe event on the network, on campus, and 
then replicate them and badge themselves into  

00:06:16.880 --> 00:06:21.200
any building they wanted because this is all 
in a broadcast LAN [local area network] and  

00:06:21.200 --> 00:06:30.400
somebody should change that, there was a time 
when if you said things like that, the people  

00:06:30.400 --> 00:06:34.080
you told it to would just be like, “Well, 
don't do that,” [LAUGHTER] right? And then …

00:06:34.080 --> 00:06:35.520
LIU: Did you listen to them?

00:06:35.520 --> 00:06:38.080
WALKER: No. [LAUGHTER] I was young.

00:06:38.080 --> 00:06:45.600
Um, and then at some point, the world started 
getting serious about imposing rules onto  

00:06:45.600 --> 00:06:51.200
networked computers, and they started hiring 
people who are interested in that. And that's  

00:06:51.200 --> 00:06:59.200
… suddenly it was a career, right. I was like, 
“Oh, I was always here, but now it has a name.”

00:06:59.200 --> 00:07:04.880
LIU: So I have to ask then, like, you 
brought up the notion of rules. I feel  

00:07:04.880 --> 00:07:09.680
like cybersecurity and rules has a 
relationship there. How much of a  

00:07:09.680 --> 00:07:14.400
rule follower were you growing up versus how 
much do you think you are a rule follower now,  

00:07:14.400 --> 00:07:20.080
and was there a difference or have you 
always been some version of “rule following”?

00:07:20.080 --> 00:07:25.200
WALKER: Well, you have to understand the rules 
if you want to be able to do what you want.

00:07:25.200 --> 00:07:26.000
LIU: Yes.

00:07:26.000 --> 00:07:30.560
WALKER: So I have a definite fascination with 
the rules, [LAUGHTER] but the thing you have  

00:07:30.560 --> 00:07:42.080
to understand about the rules is that rules are a 
social construct. And I think that a lot of … in  

00:07:42.080 --> 00:07:50.400
the computer security field, when, when people 
hear the word “security” and they think about  

00:07:50.400 --> 00:07:58.640
the rules that the security system is imposing on 
them, every time a security system asks something  

00:07:58.640 --> 00:08:04.960
of users, then you're asking for their trust. And 
that trust is a precious commodity. And so when  

00:08:04.960 --> 00:08:11.440
you squander user trust, then the security 
regime falls down. So I think it's really  

00:08:11.440 --> 00:08:18.480
important to understand the social context of 
rules instead of just having instant adjudication.

00:08:18.480 --> 00:08:22.400
LIU: OK. I'm going to have to ask you then about  

00:08:22.400 --> 00:08:26.200
a game that we played recently 
called No Vehicles in the Park.

00:08:26.200 --> 00:08:33.440
WALKER: No Vehicles in the Park is … it's a 
game about content moderation is one way to  

00:08:33.440 --> 00:08:40.880
look at it. I think it's a fascinating thought 
experiment. But it just takes the thesis that  

00:08:40.880 --> 00:08:50.000
rules are simple and challenges it with putting 
the player in the position of judge. And so the  

00:08:50.000 --> 00:08:54.240
name of the exercise is the name of the 
rule, which is no vehicles in the park.

00:08:54.240 --> 00:08:58.800
And then it asks you some very 
simple commonsense questions about,  

00:08:58.800 --> 00:09:04.400
is this a violation of the rule no vehicles 
in the park? Like if the International Space  

00:09:04.400 --> 00:09:08.840
Station overflies the park, is that a vehicle 
in the park, and is it against the rules?

00:09:08.840 --> 00:09:14.800
LIU: Mike, do you think the ISS flying 
over the park is a violation of the rules?

00:09:14.800 --> 00:09:18.520
WALKER: No. [LAUGHTER]  

00:09:19.920 --> 00:09:29.480
But it's important that we have a resilient way to 
adjudicate things that, that bridges social norms.

00:09:29.480 --> 00:09:33.760
LIU: OK. How much did you 
agree with the population,  

00:09:33.760 --> 00:09:35.200
by the way, of No Vehicles in the Park?

00:09:35.200 --> 00:09:38.240
WALKER: I've rarely ever broken 50%. Yeah.

00:09:38.240 --> 00:09:39.440
LIU: I got like 90 …

00:09:39.440 --> 00:09:40.240
WALKER: I play often.

00:09:40.240 --> 00:09:46.080
LIU: I think I got 96% on my first 
try, so I think I … does that mean  

00:09:46.080 --> 00:09:51.160
I'm kind of like a lemming that, like, just 
understands the social construct of rules?

00:09:51.160 --> 00:09:57.351
WALKER: I think it means you, 
you channel the zeitgeist.

00:09:57.351 --> 00:09:59.440
LIU: Mm, OK, I've been assimilated.

00:09:59.440 --> 00:10:00.555
WALKER: Right …

00:10:00.555 --> 00:10:01.520
LIU: That's what it feels like.

00:10:01.520 --> 00:10:08.400
All right. So let's talk about 
something that I love, which is  

00:10:08.400 --> 00:10:17.040
you had a short segment in 60 Minutes where you 
had a conversation with Lesley Stahl, and she  

00:10:17.040 --> 00:10:23.680
described you as “one of the world's top 
hackers,” at least at the time in which she  

00:10:23.680 --> 00:10:29.840
interviewed you. And so now I have to know, 
how did you find hacker cultures? What was  

00:10:29.840 --> 00:10:34.160
the extent of your involvement and, you 
know, how did you really get into it?

00:10:34.160 --> 00:10:40.720
WALKER: It, sort of, happened all at once, 
and I can tell the beginning of the story,  

00:10:40.720 --> 00:10:44.080
and then I can tell you the 
day that the story stops.

00:10:44.960 --> 00:10:51.360
So the beginning of the story is that I 
was working at a, at a startup building  

00:10:51.360 --> 00:11:02.480
telecommunications technologies on Sept. 11, 
2001, and I was living in Maryland, outside of  

00:11:02.480 --> 00:11:11.360
Washington, D.C., but my wife was in Washington, 
D.C., on that day, and, um, and then all the  

00:11:11.360 --> 00:11:22.800
phones stopped working. My workplace was overflown 
by an F-16, and, and after that, in the nation's  

00:11:22.800 --> 00:11:29.280
capital, there was this sudden talent search for 
people who understood computer security or people  

00:11:29.280 --> 00:11:35.760
who'd always been there but didn't know what 
it was called. And that talent search found me.

00:11:35.760 --> 00:11:39.440
LIU: What, uh, what were the qualifications, like,  

00:11:39.440 --> 00:11:44.320
that you had that, you know, matched with their 
list of talent that they were looking for?

00:11:44.320 --> 00:11:45.960
WALKER: I wish I knew.

00:11:45.960 --> 00:11:49.200
LIU: World's top hacker?

00:11:49.200 --> 00:11:55.106
WALKER: I think, I think that that 
label was placed on me much later.

00:11:55.106 --> 00:11:56.480
LIU: [LAUGHS] All right, fine.

00:11:56.480 --> 00:12:06.240
WALKER: I think that I became part of a series of 
cultures after that day, and one of them was the,  

00:12:06.240 --> 00:12:12.000
the group of people who play in DEF CON Capture 
the Flag, which is a hacking contest in Las Vegas.  

00:12:13.520 --> 00:12:20.480
And when you start getting involved in trying to 
get, get better at something in a way where you  

00:12:20.480 --> 00:12:26.240
can compete at a national or international level, 
especially in sort of a coffeehouse environment  

00:12:26.240 --> 00:12:32.000
where everyone else is trying to get better, 
too, it sort of propels you upward very quickly.

00:12:32.000 --> 00:12:36.720
LIU: OK, let's, let's go back later 
to talk about Capture the Flag,  

00:12:36.720 --> 00:12:41.760
because I think it's related to the, sort 
of, next arc of your life, which was your  

00:12:41.760 --> 00:12:47.520
time as a project manager—program manager—at 
the Defense Advanced Research Projects Agency,  

00:12:47.520 --> 00:12:51.680
also known as DARPA, which is also why Lesley 
had been interviewing you at the time. It was  

00:12:51.680 --> 00:12:56.720
associated with your affiliation at DARPA, 
and you were there for about four years.

00:12:57.520 --> 00:13:03.520
How did you, how did you go from, you 
know, let's say call, call to action from  

00:13:03.520 --> 00:13:11.880
the nation's capital looking for people with your 
specific set of skills to being at DARPA as a PM.

00:13:11.880 --> 00:13:15.360
WALKER: So we should, we should 
be clear, we've just, sort of,  

00:13:15.360 --> 00:13:21.120
skipped over 11 years of Mike 
trying to get better at hacking.

00:13:21.120 --> 00:13:23.360
LIU: [LAUGHS] You were world's top—look,  

00:13:23.360 --> 00:13:27.400
I just believe world's top hacker, 
one of the world's top hackers.

00:13:27.400 --> 00:13:35.360
WALKER: Well, I think, uh, I think my, my 
friends, colleagues, and teammates from the time,  

00:13:35.360 --> 00:13:42.800
if they were here to, to hear that accusation 
being made, would give me endless grief about it.

00:13:44.240 --> 00:13:49.040
Certainly, there was a, there was, sort 
of, a golden age of Capture the Flag,  

00:13:49.040 --> 00:13:52.080
and I got to be part of it 
and see some of the people.  

00:13:53.200 --> 00:13:59.920
I got to see the game change over time. 
But it was a competition, so ... and it  

00:13:59.920 --> 00:14:05.040
was very reflective of a skill set that was of 
national interest. I think at the same time,  

00:14:05.040 --> 00:14:12.480
the think tank RAND had published an opinion 
piece called Hackers Wanted, right, which,  

00:14:12.480 --> 00:14:19.760
sort of, described the immense need for more 
people who could think that way at a high level.

00:14:19.760 --> 00:14:26.880
And DARPA, specifically Norm 
Whitaker and Dan Kaufman, who  

00:14:26.880 --> 00:14:28.775
were leading I2O [Information Innovation Office] …

00:14:28.775 --> 00:14:28.792
LIU: DARPA Dan!

00:14:28.792 --> 00:14:36.160
WALKER: DARPA Dan … looked at the game of hackers, 
right, and if you could think of that as soccer,  

00:14:36.160 --> 00:14:42.000
their immediate question was, what about robot 
soccer? You know, could we make automated systems  

00:14:42.000 --> 00:14:49.120
that could play this game? And so they encountered 
myself and some colleagues giving a talk about,  

00:14:49.120 --> 00:14:56.240
sort of, the, the interleaving of Capture the 
Flag as, you know, hacking contests and talent  

00:14:56.240 --> 00:15:01.040
development. And they said, you know, could you 
do talent development for computers the same way,  

00:15:01.040 --> 00:15:05.600
right? Could machines play Capture the Flag? 
And that, that was the beginning of the  

00:15:05.600 --> 00:15:09.080
conversation about building the world's 
first all-machine hacking tournament.

00:15:09.080 --> 00:15:15.040
LIU: OK, so now, now maybe let's go back and 
explain what DEF CON Capture the Flag was,  

00:15:15.040 --> 00:15:21.040
or CTF, because I think a lot of us who 
are again not in the space, you say,  

00:15:21.040 --> 00:15:25.680
“Capture the Flag,” and I'm thinking, like, summer 
camp. And, you know, each … there are two teams,  

00:15:25.680 --> 00:15:29.440
and there's kids who hide flags, and then 
you try to run around and find the flag,  

00:15:29.440 --> 00:15:34.160
you find the defenders and whatnot. So 
what was the DEF CON version of this?

00:15:34.160 --> 00:15:38.880
WALKER: It's changed over time. That's the 
thing about this game that I should lead  

00:15:38.880 --> 00:15:44.320
with is that Capture the Flag is different with 
every generation, and at least in my opinion,  

00:15:44.320 --> 00:15:49.200
I think it's gotten better with every generation, 
right. Each hacking group that’s picked up  

00:15:49.760 --> 00:15:54.400
the lessons of the past and the game 
has made it better, but it is, sort of,  

00:15:54.400 --> 00:16:00.240
based around that playground idea, right. My team 
has some flags. Your team has some flags. We're  

00:16:00.240 --> 00:16:05.520
going to try to take each other's flags. Whoever 
takes the most wins, right. It is that simple.

00:16:05.520 --> 00:16:12.320
Now there are more than two teams, 
somewhere between eight and 16, and  

00:16:12.320 --> 00:16:16.720
each team is trying to take flags from every 
other team. They're trying to protect their  

00:16:16.720 --> 00:16:22.376
own flags. But, of course, the 
flags are made of data, right, …

00:16:22.376 --> 00:16:22.388
LIU: Mm-hm.

00:16:22.388 --> 00:16:28.400
WALKER: … or control over a networked system. 
And the way that you protect your own data and  

00:16:28.400 --> 00:16:33.920
control of your own network system is network 
defense, right, and computer defense. And the  

00:16:33.920 --> 00:16:39.440
way that you take flags from other, other teams 
is you hack into their servers and you take their  

00:16:39.440 --> 00:16:46.320
flags and you do it without getting caught. And 
then what makes this game, sort of, unique is that  

00:16:46.320 --> 00:16:52.880
all the software that you're protecting is written 
just for that game. It will never be used again.  

00:16:52.880 --> 00:16:57.760
It's boutique. It was written by the organizers, 
kept in secret, revealed to the teams the minute  

00:16:57.760 --> 00:17:02.240
they sit down. They've got new software they've 
never seen before. They don't know what it does,  

00:17:02.240 --> 00:17:06.480
and they're in a race to understand it, and 
they're racing everybody else. Because if  

00:17:06.480 --> 00:17:10.480
you can understand it, then you can protect it, 
and if you can understand it, you can hack it.

00:17:10.480 --> 00:17:16.160
LIU: OK. So let's get into this thing 
called DARPA's Cyber Grand Challenge,  

00:17:16.160 --> 00:17:23.280
or CGC, which just celebrated its 10th 
anniversary. And so from what I understand,  

00:17:23.280 --> 00:17:31.112
CTF—humans playing soccer. CGC … robots playing 
soccer? Is that probably a good way to, to ...

00:17:31.112 --> 00:17:31.920
WALKER: That's exactly it.

00:17:31.920 --> 00:17:40.560
LIU: OK, so now explain to me a little bit 
further, like, what was CGC, what was your role,  

00:17:40.560 --> 00:17:44.160
and then more importantly, did it feel 
historic at the time or just, kind of,  

00:17:44.160 --> 00:17:48.400
impossible and you somehow figured out 
with a large group of folks at DARPA,  

00:17:49.040 --> 00:17:51.120
“This is how we're going to 
solve this impossible thing”?

00:17:51.120 --> 00:17:54.880
WALKER: Well, there's, there's two 
ways to answer that. There's the,  

00:17:54.880 --> 00:17:59.600
the day before it worked and the day after 
it worked, right. [LAUGHTER] There were  

00:17:59.600 --> 00:18:04.080
actually three and a half years before 
it worked. So if you'd ask me, I think,  

00:18:04.080 --> 00:18:09.280
in any of that three and a half years, 
I would have said let's wait and see.

00:18:09.280 --> 00:18:12.240
LIU: So, so what was it, though, like—by the way,  

00:18:12.240 --> 00:18:19.192
I think you can watch the first CGC online. 
Like, there's a whole recording of the event.

00:18:19.192 --> 00:18:29.600
WALKER: There is a recording of the event, which 
we had sports casted live with announcers, and,  

00:18:30.720 --> 00:18:36.400
and so you can watch it as it happened, 
and it was, it was a capture the flag,  

00:18:36.400 --> 00:18:44.640
right—protect data, steal data—that was run 
entirely by autonomous systems behind an air gap,  

00:18:44.640 --> 00:18:50.520
so if they failed, there was no person even 
to turn them off and turn them back on again.

00:18:50.520 --> 00:18:58.560
LIU: So how did … so if I think about how you 
just described Capture the Flag with DEF CON,  

00:18:58.560 --> 00:19:01.920
but there were, like, people on the 
floor coding, I'm assuming, though,  

00:19:01.920 --> 00:19:07.853
you're saying CGC did not have that. So were 
there teams that had, like, pre-created something?

00:19:07.853 --> 00:19:14.080
WALKER: Yes. They had spent many years of their 
life actually building these autonomous systems,  

00:19:14.080 --> 00:19:22.080
and you can't really preflight an autonomous 
system whose purpose is to fight. You can  

00:19:22.080 --> 00:19:26.400
preflight an autonomous system whose 
purpose is to maybe go for a drive,  

00:19:26.400 --> 00:19:32.080
right. You can take it for a drive. But if you're 
going to fight other systems, you have to, like,  

00:19:32.080 --> 00:19:39.360
build another one, right. And we did give them 
some exposure to a, sort of, scripted opponent.

00:19:41.680 --> 00:19:47.920
The DARPA infrastructure team built a scripted 
opponent and attempted to give teams some  

00:19:47.920 --> 00:19:56.320
pre-exposure. But that first confrontation 
between adversarial hacking machines, there  

00:19:56.320 --> 00:20:02.560
wasn't really a great way to prepare for it. And 
so the teams were in this really tough position  

00:20:02.560 --> 00:20:12.760
of watching the systems play—win or lose—in a 
way where they couldn't do anything but watch.

00:20:12.760 --> 00:20:19.440
LIU: So, you know, I want you to think 
back on that time, as well, because,  

00:20:19.440 --> 00:20:24.480
like, now when we're talking about AI, it seems 
a little bit more commonplace, right? We're like,  

00:20:24.480 --> 00:20:30.240
oh, you know, we have GPT, we have Claude, 
we have a lot of other models, as well. So,  

00:20:30.240 --> 00:20:37.600
like, how historic was that? Because I think 
it can't be understated how vastly different,  

00:20:37.600 --> 00:20:43.440
sort of, autonomous or artificial intelligence 
capabilities were like 10 years ago versus now.

00:20:43.440 --> 00:20:51.360
WALKER: It was, it was a different universe 
of capability, and Cyber Grand Challenge had  

00:20:52.320 --> 00:20:58.480
a role in the development of the technology 
stacks that we're seeing today, but it was  

00:20:58.480 --> 00:21:07.520
an expert systems contest. There was, there was 
no—at that time, we would have called it machine  

00:21:07.520 --> 00:21:12.480
learning—but there's no, no real involvement, 
critical involvement, in machine learning.

00:21:12.480 --> 00:21:19.280
But there was this fusion of tooling where you 
had, sort of, an autonomous set of tools that  

00:21:19.280 --> 00:21:23.840
could perform a lot of these tasks as far as 
conventional compute could push the state of  

00:21:23.840 --> 00:21:31.040
the practice at the time. I think if we look 
back from the perspective of today, which is,  

00:21:31.040 --> 00:21:38.000
which is still a perspective that's difficult to 
keep up with and difficult to appreciate how, how  

00:21:38.000 --> 00:21:45.680
radically different the intelligence displayed by 
today's models is from what we had in 2013, 2016,  

00:21:47.280 --> 00:21:52.400
we're still seeing the technology 
stacks that we would have called  

00:21:52.400 --> 00:21:57.840
“expert systems” in 2016 being used 
as what we call a harness today.

00:21:59.040 --> 00:22:05.360
So if … just to talk about another DARPA 
project, AIxCC [AI Cyber Challenge], which  

00:22:05.360 --> 00:22:13.600
happened a few years ago, some of the people 
from Team Atlanta, which won that competition,  

00:22:13.600 --> 00:22:19.920
are building a harness called MDASH, which is, I 
think we could call it a cyber reasoning system  

00:22:19.920 --> 00:22:24.960
that is powered by models and used to 
do these sorts of tasks autonomously.

00:22:24.960 --> 00:22:30.560
LIU: You were interviewed by The 
New York Times in 2016 about CGC,  

00:22:30.560 --> 00:22:37.200
and you said that making networked technology 
safer was a “challenge for civilization.” And,  

00:22:37.200 --> 00:22:42.960
you know, what do you think about that? Has 
civilization made progress on, you know,  

00:22:42.960 --> 00:22:47.920
resolving that challenge? And what are some 
tangible takeaways that you think, you know, from  

00:22:47.920 --> 00:22:55.960
10 years ago till now, think through sort of how, 
how we're making things safer for civilization?

00:22:55.960 --> 00:23:04.000
WALKER: I'd say that's as true now as it 
was then, maybe more so. And I think our  

00:23:04.560 --> 00:23:11.120
dependence on networked systems for every piece 
of civilization has only increased. I think it's  

00:23:11.120 --> 00:23:19.520
difficult to buy a car or to participate in 
a utility grid that isn't networked, right.

00:23:22.720 --> 00:23:28.640
How we're doing? I think that we're seeing a 
great disruption of the status quo right now.  

00:23:29.920 --> 00:23:35.840
I think that we reached a pretty 
steady-state equilibrium, where our,  

00:23:36.480 --> 00:23:40.320
our security in our networked 
civilization was, sort of,  

00:23:40.320 --> 00:23:47.920
dictated by offensive scarcity. There was a 
known cost for high-end computer network attack,  

00:23:47.920 --> 00:23:56.080
and that cost equation has been completely 
upended by the capabilities of models.

00:23:56.080 --> 00:24:04.160
LIU: So in thinking about, sort of, capabilities 
of models then, some terms that, that I've heard  

00:24:04.160 --> 00:24:10.640
used are things when we're testing models, for 
example, as we call it, red teaming. But then  

00:24:10.640 --> 00:24:15.520
I also hear that there's something called blue 
teaming. And then, of course, if you've taken  

00:24:15.520 --> 00:24:20.400
primary colors or understood kindergarten 
primary colors, if you put red and blue together,  

00:24:20.400 --> 00:24:26.080
you get purple teaming. Can you explain to the 
audience a little bit about those differences?

00:24:26.080 --> 00:24:32.560
WALKER: Yeah. Red team and blue team is sort 
of an artificial separation between being  

00:24:32.560 --> 00:24:42.880
an attacker and being a defender, and it's 
also, sort of, an oversimplification and an  

00:24:42.880 --> 00:24:53.120
abnegation of capability, right. Attackers attack; 
defenders defend. But if you, if you take that  

00:24:53.120 --> 00:25:00.560
simplification away and you look at what's going 
on under the hood, you'll see that the difference  

00:25:00.560 --> 00:25:07.760
between offense and defense most of the time 
is, sort of, who's in control or who's driving.

00:25:08.400 --> 00:25:14.240
If you were to give a name to a program that 
encrypted every file in your computer and then  

00:25:14.240 --> 00:25:21.440
unlocked it when you needed it, that is either a 
cryptography utility or it's ransomware depending  

00:25:21.440 --> 00:25:30.480
on who holds the key. If you were to give a name 
to a program that, that allows you to log in to  

00:25:30.480 --> 00:25:36.640
your computer remotely and control it, maybe your 
home automation setup, that is either malware or  

00:25:36.640 --> 00:25:41.440
remote administration utility [LAUGHTER] depending 
on who holds the key, right? Who holds the key.

00:25:41.440 --> 00:25:47.440
So there's, there's this, sort of, blurring 
of offense and defense where computer security  

00:25:47.440 --> 00:25:55.840
capabilities at their most atomic basic level 
aren't offensive or defensive. They're dual use.  

00:25:55.840 --> 00:26:05.120
And so the idea of red teaming and blue teaming 
is sort of this idea that we can impose single use  

00:26:06.000 --> 00:26:13.600
onto dual-use capabilities, and purple team is 
the acknowledgment that that's not true, right.

00:26:13.600 --> 00:26:19.920
Attackers, as soon as they break into a network, 
they have an intrusion that they need to defend.  

00:26:19.920 --> 00:26:24.640
You'll hear all these terms like moving laterally, 
which just means breaking into more computers,  

00:26:25.600 --> 00:26:29.840
or exfiltration, which means taking data. 
But the whole time they're doing that,  

00:26:29.840 --> 00:26:35.040
they're trying not to get caught, right. 
So they are defending their attack, right.

00:26:35.040 --> 00:26:35.252
LIU: Mm.

00:26:35.252 --> 00:26:44.640
WALKER: And at the same time, when defenders are 
doing things like log analysis or really any—or  

00:26:44.640 --> 00:26:55.360
traffic analysis—their tools themselves can be 
broken into, right. They have a duty to defend  

00:26:55.360 --> 00:27:06.000
their own tooling, and they also need to, in order 
to prevent attackers, they need to be able to find  

00:27:06.000 --> 00:27:11.120
vulnerabilities, find weaknesses, and then grade 
weaknesses in their own software. And the way  

00:27:11.120 --> 00:27:14.880
that you find vulnerabilities is vulnerability 
research, and you grade weaknesses by writing  

00:27:14.880 --> 00:27:20.640
exploits or at least proof of vulnerability, and 
that's offense, right. So purple team is just this  

00:27:20.640 --> 00:27:27.440
idea that attack and defense are the same skill 
set. They're the same tooling. They're blended.  

00:27:27.440 --> 00:27:33.720
And it's really a contest to control the logic of 
memory of networked computers run all the time.

00:27:33.720 --> 00:27:37.120
LIU: All right. So, so I'm going to 
ask you a little bit about the team  

00:27:37.120 --> 00:27:42.560
that you manage then. It's called 
New Security Ventures. And would  

00:27:42.560 --> 00:27:46.320
you classify the type of work that you 
and your team do as being red, blue,  

00:27:46.320 --> 00:27:50.600
or purple? Or are we going to introduce 
a different coloring from the rainbow?

00:27:50.600 --> 00:27:55.360
WALKER: I think that we do highly 
offense-aware defensive tool  

00:27:55.360 --> 00:28:01.800
development. So we build defensive tools, but 
we build them knowing that they are a target.

00:28:01.800 --> 00:28:07.200
LIU: What's one example of a project 
that your team has been working on?

00:28:07.200 --> 00:28:13.840
WALKER: We have a project we call Project Freta. 
It's named after the birthplace of Marie Curie,  

00:28:15.680 --> 00:28:21.040
the pioneer of battlefield imaging. 
[LAUGHTER] And the idea is,  

00:28:21.040 --> 00:28:26.320
is that … well, first I actually need 
to introduce tampering. We think of  

00:28:26.320 --> 00:28:30.126
ourselves as the solution to tampering, 
but, but first, what's tampering, right?

00:28:30.126 --> 00:28:32.000
LIU: What’s tampering?
WALKER: What problem are we solving?

00:28:32.000 --> 00:28:38.320
So let's say that my job is to break into a store 
undiscovered, right? It's mission impossible,  

00:28:38.320 --> 00:28:42.480
right? I'm going to break in, take 
the Doritos, not get caught, right.

00:28:42.480 --> 00:28:44.680
LIU: Wait. Hold on. What flavor of Doritos?

00:28:44.680 --> 00:28:46.240
WALKER: I don't know.

00:28:46.240 --> 00:28:47.280
LIU: What?!

00:28:47.280 --> 00:28:49.880
WALKER: I don't actually eat them. I'm sorry.

00:28:49.880 --> 00:28:50.880
LIU: OK, fine.

00:28:50.880 --> 00:28:54.500
WALKER: But that's, you know, it's 
mission impossible. I have to take them.

00:28:54.500 --> 00:28:55.440
LIU: Fine. You got to go get Doritos.
WALKER: I have to not show up on the video.

00:28:55.440 --> 00:28:55.600
LIU: Yeah.

00:28:55.600 --> 00:29:00.560
WALKER: I can compute this path of not being 
observed by the video camera. It's really hard,  

00:29:00.560 --> 00:29:06.240
right. Maybe I do some, some kung fu, some 
gymnastics. I slide into the laser wire,  

00:29:06.240 --> 00:29:11.680
I grab it, and I get out. But there is 
another way to break in and not get caught,  

00:29:11.680 --> 00:29:18.080
which is to walk in, right, wave at the 
camera, cut the feed on the camera, take it  

00:29:18.080 --> 00:29:23.120
down, and replace it with a fake video camera that 
just plays good video, like there's nobody in the,  

00:29:23.120 --> 00:29:29.200
in the store, and then go in and change the 
tape and leave. That's tampering, right?

00:29:29.200 --> 00:29:29.512
LIU: OK.

00:29:29.512 --> 00:29:35.680
WALKER: And it sounds actually less complicated, 
right, than trying to simply not be observed. And  

00:29:36.960 --> 00:29:42.400
if you look in the computer security world at 
tampering, tampering is often the easier way  

00:29:42.400 --> 00:29:46.320
to accomplish the objective of an attacker, 
which is to break in and not get caught and  

00:29:46.320 --> 00:29:51.520
defend their intrusion, right. They're doing 
defense while they're doing offense. And so  

00:29:51.520 --> 00:29:57.680
Project Freta is an attempt to solve that at 
cloud scale, to make it so that when defenders  

00:29:57.680 --> 00:30:02.480
acquire defensive signals, right, telemetry 
signals about what's happening in computers,  

00:30:02.480 --> 00:30:06.800
they don't do it with software that can 
be removed or tampered with. They do it  

00:30:06.800 --> 00:30:14.160
by acquiring signals from cloud hardware. 
And so we're a whole-system memory analysis  

00:30:14.160 --> 00:30:19.440
environment built from the ground up for looking 
at memory that's acquired from cloud hardware.

00:30:19.440 --> 00:30:25.440
LIU: So, OK, so you just revealed, for 
example, how Project Freta works and how  

00:30:25.440 --> 00:30:31.760
it protects or is trying to defend against 
something like tamper. But then, you know,  

00:30:31.760 --> 00:30:37.200
I'm kind of curious, though. You just revealed 
how it works. So, you know, you're kind of telling  

00:30:37.200 --> 00:30:44.320
the attackers what you're doing and how you're 
defending. How does that work if you're like,  

00:30:44.320 --> 00:30:49.400
“This is how I'm defending, like, this is 
how I'm defending my castle. Here you go”?

00:30:49.400 --> 00:30:55.360
WALKER: Well, you can reveal a technology 
or you can reveal its deployment. And I've  

00:30:55.360 --> 00:30:59.200
revealed a technology, and I haven't 
said anything about its deployment.

00:30:59.200 --> 00:31:05.360
LIU: OK, so there's a, there's a difference, 
and there's layers that you've described so  

00:31:05.360 --> 00:31:10.000
that, you know, even though you've revealed 
something, you haven't revealed the full extent.

00:31:10.000 --> 00:31:14.480
WALKER: Well, you're, you're bringing up a 
really interesting part of the game theory of,  

00:31:14.480 --> 00:31:20.880
of adversarial computer network attack-computer 
network defense, which is that attackers condition  

00:31:20.880 --> 00:31:28.960
their behavior based on their perception of what 
the defense is doing. And so that gets into these,  

00:31:28.960 --> 00:31:36.160
these strange effects of, is the best way to 
protect a network to say that you have something  

00:31:36.160 --> 00:31:41.440
or to actually have it or to have a mismatch 
between what you say you have and what you have.

00:31:41.440 --> 00:31:43.360
LIU: So you're like bluffing in poker.

00:31:43.360 --> 00:31:48.635
WALKER: There's a reason this game 
is played in Vegas. [LAUGHTER]

00:31:48.635 --> 00:31:55.120
LIU: OK, OK. So we've had, we've had some, like, 
more serious, like, topics for a while. And I want  

00:31:55.120 --> 00:32:00.960
to get into something that maybe a lot of folks 
who maybe even know you don't know about you,  

00:32:00.960 --> 00:32:05.640
which is you love—I think I'm going to 
use … I think I can say love—cycling.

00:32:05.640 --> 00:32:07.000
WALKER: That's true.

00:32:07.000 --> 00:32:08.880
LIU: How did you get into cycling?

00:32:09.440 --> 00:32:14.880
WALKER: I got into cycling in, sort of, the way 
that I've gotten into a lot of things in my life,  

00:32:14.880 --> 00:32:19.280
which is I got into it through 
what I thought at the time was,  

00:32:19.840 --> 00:32:23.760
was a disaster. I blew out 
my knee playing basketball,  

00:32:23.760 --> 00:32:30.000
which was one of my first great loves, right. And 
I went through surgical repair. And afterwards,  

00:32:30.000 --> 00:32:34.560
the surgeon told me you can't do the 
stuff with this knee that you used to do.

00:32:34.560 --> 00:32:38.240
I said, “Well, what can I do?” You know, I 
hadn't even thought to ask this question before  

00:32:38.240 --> 00:32:43.520
surgery. You know, I'm sitting there trying 
to, like, rehabilitate the knee. And he said,  

00:32:43.520 --> 00:32:47.920
“Well, you could try swimming 
or you could try riding a bike.”

00:32:48.480 --> 00:32:53.680
I didn't really know anything about either 
one. You know, I used to ride a bike to get  

00:32:53.680 --> 00:33:00.240
around. And then a friend of mine in college 
introduced me to mountain biking, which is  

00:33:00.240 --> 00:33:04.560
sort of this idea that you could take a bicycle, 
which is a very stable system—you know, you go  

00:33:04.560 --> 00:33:10.160
out and you ride it down the sidewalk; it works 
great—and ride it over, you know, just horrible,  

00:33:10.880 --> 00:33:17.640
you know, really steep, rough terrain that 
no one should ever try to ride a bike on.

00:33:17.640 --> 00:33:25.920
LIU: OK. And you did this recently at a 
100-mile endurance race called High Cascades.

00:33:25.920 --> 00:33:27.520
WALKER: High Cascades 100.

00:33:27.520 --> 00:33:31.680
LIU: And what was it like to prep for that? 
Because I feel like you've been, you've been  

00:33:31.680 --> 00:33:36.640
working on some very important things on the 
professional side. And then you also had this big  

00:33:36.640 --> 00:33:42.840
100-mile endurance race that you were prepping 
for on the personal side. How did that go?

00:33:42.840 --> 00:33:46.960
WALKER: There are very few things in life 
where you have a fixed date and time where  

00:33:46.960 --> 00:33:53.360
you have to show up and deal with uncertainty, 
and ultra-endurance racing is one of those. Um,  

00:33:53.360 --> 00:33:56.560
there's, there's the parts that you can 
prepare for and the parts you can't,  

00:33:56.560 --> 00:34:02.240
right. You can't prepare for the smoke forecast or 
the weather. You can just have a range of options.

00:34:02.240 --> 00:34:04.320
But then, you know, as far as the conditioning,  

00:34:04.320 --> 00:34:08.560
I do spend a lot of time on the indoor 
trainer. I just wake up really early. I have,  

00:34:08.560 --> 00:34:16.000
I have a whole structured plan. I think 
I started around January to race in July.

00:34:16.000 --> 00:34:20.232
LIU: Do you follow them like 
you follow rules? [LAUGHS]

00:34:20.232 --> 00:34:24.880
WALKER: Well, I'm much more strict about 
following my training plan. Actually,  

00:34:24.880 --> 00:34:29.600
I think, I think I've been advised—this 
is real—not to be trainer robot.

00:34:29.600 --> 00:34:30.560
LIU: Wow.

00:34:30.560 --> 00:34:33.040
WALKER: So I had to back 
off on being trainer robot.

00:34:33.040 --> 00:34:39.240
LIU: OK, so you had to ease up on the rules that 
you were following, that you wrote for yourself.

00:34:39.240 --> 00:34:45.680
WALKER: Yeah. There's this mode … all the 
cycling fans on the … watching this are going to,  

00:34:45.680 --> 00:34:49.360
are going to be like, [WHISPERS] “Oh my gosh, 
he’s talking about erg mode.” There's this mode  

00:34:49.360 --> 00:34:54.320
called erg mode where the, the trainer decides 
how hard you should pedal. And it took me … and  

00:34:54.320 --> 00:34:58.720
so I would just set it to that and do what it 
told me, right. And so it was this big thing to,  

00:34:59.520 --> 00:35:04.960
to actually go into free ride and 
try to, try to ride the terrain.

00:35:04.960 --> 00:35:05.432
LIU: OK.

00:35:05.432 --> 00:35:08.480
WALKER: Yeah. Back, back off and try to,  

00:35:08.480 --> 00:35:14.960
try to listen to your body a little bit 
rather than just do whatever it takes.

00:35:14.960 --> 00:35:19.840
LIU: [LAUGHS] I feel like there's a relationship 
between what we are talking about now and  

00:35:19.840 --> 00:35:28.000
something maybe related to the cybersecurity side 
of Mike. But I have to … I really want to get a  

00:35:28.000 --> 00:35:34.619
photo of this on this podcast, this video podcast. 
So I am going to ask you about Orange Crush.

00:35:34.619 --> 00:35:34.635
WALKER: Oh, OK.

00:35:34.635 --> 00:35:37.600
LIU: Can you tell the 
audience, what is Orange Crush?

00:35:37.600 --> 00:35:40.240
WALKER: Orange Crush is the project bike that  

00:35:40.240 --> 00:35:44.320
I built over the year when I 
was out with a blown shoulder.

00:35:44.320 --> 00:35:46.040
LIU: I think that's when I met you.

00:35:46.040 --> 00:35:51.920
WALKER: Yeah. The problem with riding 
the bikes down the steep hills is,  

00:35:51.920 --> 00:35:54.120
like, sometimes, it doesn't go completely perfect.

00:35:54.120 --> 00:35:58.480
LIU: Didn't we have … we also had a colleague 
who also likes to mountain bike who, like,  

00:35:58.480 --> 00:36:03.020
broke his back at around the same 
time you had a shoulder injury.

00:36:03.020 --> 00:36:03.960
WALKER: Yeah.
LIU: Also mountain biking.

00:36:03.960 --> 00:36:06.480
WALKER: Yeah. Kind of depends on 
how you break your back versus how  

00:36:06.480 --> 00:36:09.040
you break your shoulder on which 
one of those works out better.

00:36:10.480 --> 00:36:16.480
I got a … I messed my shoulder up pretty good, 
so it took some time, but it meant that I got to,  

00:36:16.480 --> 00:36:21.760
kind of, conceive of something and 
put it together, mostly left-handed,  

00:36:21.760 --> 00:36:29.280
which is its own challenge. So it's, it's 
a gravel bike, but it's … and, of course,  

00:36:29.280 --> 00:36:36.080
my wife named it. She gets the naming rights 
because she's better at it, but it is very orange.

00:36:36.080 --> 00:36:40.560
LIU: Um, we're, we're coming up close 
on time, so I'm going to start wrapping  

00:36:40.560 --> 00:36:46.080
up with a few questions, but I, I love this 
question and I can't help but ask you, Mike,  

00:36:46.080 --> 00:36:51.280
what is the difference between suffering on a bike 
and suffering through a hard technical project?

00:36:51.280 --> 00:36:55.120
WALKER: Well, I don't think of riding a bike 
as suffering at all. That's like the good part.

00:36:55.120 --> 00:36:59.200
LIU: Even, even, OK. Because in your 
100-mile endurance race on High Cascades,  

00:36:59.200 --> 00:37:02.766
I think you said it was, like, nine hours of …

00:37:02.766 --> 00:37:05.760
WALKER: Yeah.
LIU: … nonstop riding, like, in terrain.

00:37:05.760 --> 00:37:07.080
WALKER: Racing. Yeah.

00:37:07.080 --> 00:37:11.200
LIU: That seems like there might 
have been a little bit of suffering.

00:37:11.200 --> 00:37:15.200
WALKER: No, that's absolutely the best 
part. [LAUGHTER] No, no, the suffering is,  

00:37:15.200 --> 00:37:19.360
the suffering is the checklist and the 
logistics the day before. Suffering is  

00:37:19.360 --> 00:37:23.520
going out for a pre- ride and getting a flat 
and trying to find a bike shop that can get a  

00:37:24.240 --> 00:37:31.200
tire patch in the day before High Cascades. 
[LAUGHTER] No, the logistics and the,  

00:37:31.200 --> 00:37:38.000
and the worrying, right, that's always the hard 
part. Once, once, once you're at … you know,  

00:37:38.000 --> 00:37:43.040
it just gets worse and worse up until the start. 
And then you're racing, and it's wonderful.

00:37:43.040 --> 00:37:46.880
LIU: And I want the audience 
to know—because High Cascades,  

00:37:46.880 --> 00:37:48.792
you can actually look it up on the internet …

00:37:48.792 --> 00:37:49.431
WALKER: That's true.

00:37:49.431 --> 00:37:56.400
LIU: … Mike finished 69th out of 329 
participants. I’d say that's pretty,  

00:37:56.400 --> 00:37:59.400
that's pretty darn good. I don't 
even know how to ride a bike, Mike.

00:37:59.400 --> 00:38:04.720
WALKER: Um, I, I race these days 
to learn about myself, you know.

00:38:04.720 --> 00:38:08.560
LIU: Do you work on hard technical 
projects to also learn about yourself?

00:38:08.560 --> 00:38:18.800
WALKER: I work on hard technical projects half 
because just getting to solve hacking problems is,  

00:38:18.800 --> 00:38:23.520
it’s, sort of, what I've always done. But 
then also, I get to work with great people,  

00:38:23.520 --> 00:38:29.520
and I get to, I get to see impact. 
I get to see the impact of our work.

00:38:29.520 --> 00:38:36.051
LIU: All right. So I have a lightning round 
of questions that I'm going to ask you.

00:38:36.051 --> 00:38:36.075
WALKER: OK. All right.

00:38:36.075 --> 00:38:40.312
LIU: So just come up with—this is not a 
Rorschach test. Just come up with, like, ...

00:38:40.312 --> 00:38:41.360
WALKER: It might be.

00:38:41.360 --> 00:38:48.080
LIU: It requires pictures if I do that. But we're 
going to do a lightning round. So the first one:  

00:38:48.080 --> 00:38:55.120
advice that blew your mind 
or changed your perspective.

00:38:55.120 --> 00:39:02.800
WALKER: Hm, um, I remember when 
I was starting at the agency,  

00:39:05.600 --> 00:39:11.440
I was, I’d, I’d worked in an engineering 
shop for a long time, and the culture of,  

00:39:11.440 --> 00:39:17.600
of many engineering shops is to be critical of 
leadership, right. They would see what leadership  

00:39:17.600 --> 00:39:24.480
did and then take the information they had and 
be critical of the decision. Not all the time,  

00:39:24.480 --> 00:39:32.320
but sometimes. And I had a mentor who took me 
aside and said, “Have you ever considered that  

00:39:32.320 --> 00:39:37.040
the decisions that you don't agree with are 
the result of information you don't have?”

00:39:37.040 --> 00:39:37.940
LIU: Ooh.

00:39:37.940 --> 00:39:44.240
WALKER: And then they took me into a situation 
where I, sort of, was revealed … the,  

00:39:44.240 --> 00:39:49.040
the information I didn't have about the 
decision I disagreed with was revealed to me,  

00:39:49.040 --> 00:39:53.040
and suddenly I thought it was a great 
decision. I was like, “That's incredible.”  

00:39:53.040 --> 00:39:58.560
And then I left and I couldn't talk about 
it, and I saw other people having the same  

00:39:58.560 --> 00:40:04.160
criticism of the decision, and that 
started the process of reformatting  

00:40:04.160 --> 00:40:09.040
how I thought about leadership. Every time 
I hear a decision now, I think to myself,  

00:40:09.040 --> 00:40:13.520
“What do they know that I don't?” And 
that curiosity process is always running.

00:40:13.520 --> 00:40:18.560
LIU: OK. Lesson you learned the hard 
way—which is … I feel like this is a  

00:40:18.560 --> 00:40:23.160
very easy question for you—or 
the misstep that humbled you.

00:40:23.160 --> 00:40:27.675
WALKER: There's so much to 
choose from here. [LAUGHTER]

00:40:27.675 --> 00:40:29.040
LIU: First one that comes to mind.

00:40:29.040 --> 00:40:31.280
WALKER: I probably made one on the way here.

00:40:33.760 --> 00:40:40.320
I'll, I'll pick one that's, that's relevant to 
the discussion today, which is that I remember  

00:40:40.320 --> 00:40:47.920
distinctly playing in qualifiers in a hacking 
competition and getting, getting the answer,  

00:40:47.920 --> 00:40:55.760
right—getting the key, right—and running up to 
the person who was handing in keys for our team,  

00:40:55.760 --> 00:40:59.280
being like, “I solved”— you know, I don't 
know what it was; every challenge has a name,  

00:40:59.280 --> 00:41:08.480
right—"like, 202,” and they were like, “Oh, we 
solved that an hour and a half ago.” [LAUGHTER]

00:41:08.480 --> 00:41:14.640
And, and I, I remember, you know, “Oh, OK. 
Well, I better find something new to work on.”  

00:41:14.640 --> 00:41:20.320
And I just looked around the room and, you know, 
we're all curled over our hacking workstations,  

00:41:20.320 --> 00:41:26.160
two monitors and a keyboard no one would recognize 
as a keyboard —you know, we all look the same—and,  

00:41:28.800 --> 00:41:34.720
and thinking to myself, I can 
quantify the IQ in this room,  

00:41:34.720 --> 00:41:40.301
but I can't look around and quantify 
how well we're working together.

00:41:40.301 --> 00:41:40.312
LIU: Mmm.

00:41:40.312 --> 00:41:43.680
WALKER: Right? I don't know how aligned 
we are in this goal of getting the most  

00:41:43.680 --> 00:41:47.520
points and winning here. And I can't 
look around and figure out how many  

00:41:47.520 --> 00:41:52.880
people here are on the right challenge or 
on a challenge that's already been solved,  

00:41:52.880 --> 00:41:56.880
have the tools they need, would 
be paired with the right person.

00:41:56.880 --> 00:42:03.440
And myself and a couple of my colleagues 
started thinking about, like, what is the  

00:42:03.440 --> 00:42:11.280
right way to create social intelligence in, 
in technical teams and hacking teams? And,  

00:42:12.400 --> 00:42:18.560
and that has stuck with me. I mean, that's been 
a constant thread, is, is just the abandonment  

00:42:18.560 --> 00:42:25.600
of the hero journey and net IQ as a way that, 
that teams accomplish things, because you will  

00:42:25.600 --> 00:42:30.880
find people working on challenges to get the 
key, right—as a metaphor—where the problem’s  

00:42:30.880 --> 00:42:38.080
already been solved in any organization that isn't 
working on, on its EQ. You’ll find it everywhere.

00:42:38.080 --> 00:42:43.360
LIU: OK, talent or—I was going to ask you 
a bunch more, but we're lightning rounding  

00:42:43.360 --> 00:42:57.748
right now. Talent or accomplishment you're 
most proud of if given the chance to brag.

00:42:57.748 --> 00:43:00.147
WALKER: Mm. Bragging’s not in my 
nature, I don’t think. But ...

00:43:00.147 --> 00:43:01.339
LIU: I know. Which is the ...

00:43:01.339 --> 00:43:03.327
WALKER: But …
LIU: Go on.

00:43:03.327 --> 00:43:03.760
WALKER: Let’s see.
LIU: Flex a little bit.

00:43:03.760 --> 00:43:09.040
WALKER: I, uh, I taught myself to 
juggle while my knee was blown out.

00:43:09.040 --> 00:43:13.000
LIU: [LAUGHTER] That's the thing 
you're most proud of is juggling.

00:43:13.000 --> 00:43:16.320
WALKER: Well, so if you haven't, if 
you haven't learned how to juggle,  

00:43:16.960 --> 00:43:23.600
it's fine, right. But if you have, you know that 
99% of it is dropping them and they go everywhere,  

00:43:23.600 --> 00:43:28.400
which when you're immobilized with a busted 
knee, it's actually harder than it sounds.

00:43:28.400 --> 00:43:32.800
So I engineered a way to teach myself 
to juggle where I just got myself into a  

00:43:32.800 --> 00:43:37.440
small enough space that I couldn't lose them 
and I could just sit there with my blown-out  

00:43:37.440 --> 00:43:41.440
post-surgery knee and teach myself to 
juggle. I did it in a couple of weeks.

00:43:41.440 --> 00:43:42.760
LIU: Can you still juggle?

00:43:42.760 --> 00:43:43.360
WALKER: Oh, yeah.

00:43:43.360 --> 00:43:46.312
LIU: OK, so if I handed you stuff later …

00:43:46.312 --> 00:43:50.386
WALKER: I mean, let's not 
break the room, but sure.

00:43:50.386 --> 00:43:54.320
LIU: [LAUGHS] OK. Last one: career 
path you would have chosen—this is  

00:43:54.320 --> 00:43:58.240
going to be hard—if you weren't, like, 
in cybersecurity. And it feels like it's,  

00:43:58.240 --> 00:44:02.160
like, in your DNA, so I'm not sure, 
but I would love to hear an attempt.

00:44:02.160 --> 00:44:06.000
WALKER: Oh, it's science fiction writer 100%.

00:44:06.000 --> 00:44:06.680
LIU: Really?  
 
 

00:44:06.680 --> 00:44:11.680
WALKER: You know, I thought I was going to be 
a musician, but when you're a … when you go out  

00:44:11.680 --> 00:44:16.480
and you start getting into a music scene, you, 
kind of, find out if you were a natural or not.  

00:44:16.480 --> 00:44:22.720
So I think science fiction writer is still the 
thing where I wish I'd kept at it a little more.

00:44:22.720 --> 00:44:27.120
LIU: All right, I'm going to, I'm going to 
break my own rule, which is not … which is  

00:44:27.120 --> 00:44:31.360
a social construct, and ask you one more, 
which is, can you recommend to the audience  

00:44:32.080 --> 00:44:37.120
a really good—your favorite or one of 
the best sci-fi books you've ever read?

00:44:37.120 --> 00:44:46.720
WALKER: Oh, sure. Easy. It's The Forever 
War. Haldeman’s Forever War. And it's,  

00:44:47.520 --> 00:44:55.120
I think, it's about coming to hand in the 
key except the challenge’s already solved,  

00:44:55.120 --> 00:44:59.840
except sort of writ large in a, in 
a universe governed by relativity.

00:44:59.840 --> 00:45:00.880
LIU: OK.

00:45:00.880 --> 00:45:02.120
WALKER: No spoilers.

00:45:02.120 --> 00:45:04.240
LIU: No spoilers. All right. Fine.

00:45:04.240 --> 00:45:07.760
Thank you so much, Mike—this 
was a great time—for joining me.

00:45:07.760 --> 00:45:10.880
To our audience, thank you for 
listening. If you want to learn a  

00:45:10.880 --> 00:45:14.720
little bit more about any of my colleagues 
in Microsoft Research and their work,  

00:45:14.720 --> 00:45:20.720
check out the Microsoft Research website 
at aka.ms/research [MUSIC] or listen and  

00:45:20.720 --> 00:45:27.520
watch other episodes of the podcast on 
YouTube and major podcast platforms.

00:45:27.520 --> 00:45:28.339
[MUSIC ENDS]