Latest / AI for Founders with Ryan Estes / Agentic AI, Deepfakes and the End of Passwords

Transcript

Agentic AI, Deepfakes and the End of Passwords · AI for Founders with Ryan Estes

0:00
Welcome back to AI for Founders. I'm Ryan Estes.
0:02
There's a war raging online, not between hackers and
0:05
corporations, but between humans and machines pretending to be
0:09
human. Passwords are failing, deepfakes
0:11
are rising, and identity itself is up for grabs.
0:14
Today on AI for Founders, I'm talking with two experts on the
0:18
front lines. Jason, CEO of Beyond Identity, a
0:22
cybersecurity company backed by over $200 million to make
0:26
passwords obsolete. And Josh from Brand Hospitality
0:29
Group, who's deploying this kind of tech in real world high
0:33
turnover environments where security meets simplicity.
0:37
We're diving into how AI impersonation is forcing a total
0:41
rethink of authentication, not just proving who you are, but
0:45
proving who's really acting when code agents and AI systems start
0:49
making decisions for all of us. If you're a founder building an
0:52
AI, SAS or cybersecurity, or just tired of MFA nightmares,
0:57
this one's for you. You'll learn how to future proof
1:00
your identity systems, give your AI agents their own digital
1:04
passport, and maybe even glimpse what happens when you're
1:07
credentials live under your skin.
1:10
Now don't just listen. Upgrade your intelligence feed.
1:13
Subscribe to the AI for foundersnewsletter@aiforfounders.co.
1:18
This is weekly drops of founder frameworks, big AI insights, and
1:22
behind the scenes tools we don't share anywhere else.
1:25
It's where the smartest founders come to sharpen their edge.
1:28
Sign up, forward it to your smartest friend, and pretend it
1:31
was your idea all along. And if this episode makes you
1:34
think differently about identity, leave us a review.
1:37
Give us five stars if you'd let your AI clone manage your inbox.
1:41
This episode is brought to you by Kitcaster, the podcast
1:44
booking agency trusted by over 800 founders and counting.
1:48
Kitcaster helps you get booked on the top podcast that reach
1:51
your ideal audience without the cold emails, the back and forth
1:54
scheduling, or the guesswork. You show up, Share your story,
1:58
and build real influence in your industry.
2:01
Whether you're fundraising, hiring, or scaling your brand,
2:04
podcast interviews are one of the fastest ways to drive away
2:07
awareness, build trust and close deals.
2:10
I've seen it first hand. Founders we work with have
2:12
landed partnership, raised millions, and created massive
2:15
growth just by showing up on the right shows.
2:18
Ready to get booked? Head over to kitcaster.com and
2:21
apply today. That's kick caster by Moburst.
2:25
Hey everybody, you know what's crazy?
2:27
I've been sitting on a goldmine this entire time and I didn't
2:30
even know it. My Gmail inbox, literally
2:33
thousands of warm contacts who already know and trust me.
2:36
But here's the problem every founder faces.
2:38
How do you turn those relationships into revenue
2:40
without feeling like a sleazy salesperson?
2:43
That's exactly why I've been using Warm Star dot AI and
2:47
honestly it's been a game changer for my business.
2:49
This AI platform connects your e-mail, analyzes all your
2:53
existing contacts, and helps you identify who in your network
2:56
might actually be interested in what you're doing.
2:59
Then it writes personalized outreach messages that don't
3:01
sound robotic or pushy. The best part?
3:04
You review every single message before it goes out.
3:07
No spray and pray nonsense. This is relationship based
3:10
selling done right. My listeners know I only
3:13
recommend tools I actually use, and Warm Start has literally
3:16
helped me reconnect with people I'd lost touch with and turn
3:19
those conversations into real business opportunities.
3:23
If you're a founder struggling with outreach, check out
3:25
warmstar dot AI and use code AIFF to get $50 off your first
3:31
month. Trust me, your existing network
3:33
is worth way more than you think.
3:51
All right. Welcome back, everybody.
3:53
We're about to have a great conversation about personal
3:56
identity, AI and a lot of other different topics.
4:01
I'm really excited. We've got Jason today and Josh.
4:03
Jason is with Beyond Identity, Josh is with Grant Hospitality
4:09
Group and what we're going to be talking about is Beyond Identity
4:12
to get going. Now, this is a cybersecurity
4:14
company founded in 2019 with a mission to eliminate passwords,
4:18
thank God, and make authentication truly fishing
4:21
proof. Now their technology binds user
4:24
identity directly to a device using crypto cryptographic keys,
4:28
removing the weakest link in modern security, the password.
4:32
Plus, I mean, we've all got to be just absolutely beat to death
4:37
with two factor authentication and all the passwords.
4:40
We've got to remember it's a mess.
4:42
So they're backed over by over 200 million in funding and
4:45
trusted by global enterprises. Beyond identity is redefining
4:48
with zero trust really means in a world where identity is the
4:52
new perimeter. Welcome to the show guys.
4:54
Thanks. For Anonymous.
4:57
You bet. Now let's get into the AI
5:00
conversation 1st. And the one thing I was
5:02
particularly interested in talking to you guys about is
5:06
kind of, we were talking about agentic support and everybody
5:10
kind of feels like they need to have some kind of agentic plan.
5:13
But now you're starting to see agentic tools basically
5:17
increasingly impersonate humans intentionally.
5:20
So either that's clone voices and synthetic identities and we
5:24
there's deep fake opportunities, but also just the ability to
5:27
like do captchas and and perform tasks on our behalf.
5:34
So I'm just really interested in Mike Beyond Identity's approach
5:37
to basically extend to proving who is acting and how that will
5:41
become more relevant as these agentic solutions that we're all
5:45
building become more and more sophisticated.
5:50
So that's a. That's a very simple short
5:55
question with a very long and deep response.
5:58
I'll try, I'll try and I'll try and give you the menu version
6:01
and we can choose our own adventure.
6:02
You bet. So the first thing is, I would
6:05
say there's the AI enabled adversary.
6:09
And what is the AI enabled adversary to do?
6:12
And so we think about the AI enabled adversary.
6:15
We're thinking of the ability to really kind of mimic victims,
6:18
mimic targets, right? And voice and speech and writing
6:21
and appearance. And, and so we have a line of
6:26
use cases that we call reality check.
6:29
And so the idea of that is how do you defend against the AI
6:32
enabled adversary? And if we pull that thread just
6:35
a little bit on mimicry, you'll start to think, well, wait a
6:40
minute. I have seen a lot of products
6:42
recently about detecting AI like, hey, this is a fake photo
6:45
or this is a fake video. And our response is that's the
6:49
exact wrong solution for two reasons #1 is when you dig into
6:55
the technology, there is a bit of an arms race between from
6:58
detection and detection avoidance or synthesis, right?
7:02
So one person's detector is a great, is the great training
7:06
agent for the next person who's building the next generator,
7:10
right? The next synthesizer.
7:11
So, and you can see this, right, when detections come out and
7:15
their rate will be, their detection rate will be high and
7:17
that'll fall pretty quickly. The second argument is a little
7:20
bit more about like use if we really believe AI is around the
7:23
corner, and we should because it is.
7:26
By the way, did you know I don't speak English?
7:29
I, I this is real time AII don't actually look like this.
7:33
I'm actually, I actually have red hair and a tail, but I'm
7:38
using AI, right? So.
7:41
I didn't want to tell you guys, but I'm actually a porcupine, so
7:44
right look at how great it looks.
7:47
So what does it even mean to detect the presence of AI when
7:50
there are all these valid use cases that we know are coming
7:53
that are going to be helpful in things like customer service,
7:55
customer support, human interaction.
7:59
So better question is where is this content coming from?
8:03
Who authorized the production of the content?
8:05
What's the strength of trust of the identity behind that?
8:08
And we believe that's answerable by device bound hardware backed
8:12
identity. We can get into a lot more of
8:14
what that means in a little bit. The other thread is I'm not
8:18
necessarily defending myself against an AI adversary.
8:21
I'm using AI and what kind of vulnerabilities does this now
8:25
expose me to that I may not be used to?
8:27
And what we're actually finding is AI is not necessarily
8:30
introducing new vulnerabilities. It's taking existing
8:33
vulnerabilities and expanding their surface area by an order
8:36
of multiple magnitudes. And and so we think there's
8:41
similar solutions to like solving that problem around
8:45
device bound hardware backed identity.
8:46
What you need to do is you need to give your agents identity.
8:49
You need to actually string that back to your users.
8:52
Like your user authorizes an agent for a period of time with
8:55
permissions. Those both should actually be
8:58
certified. Those both should be
8:59
cryptographically related, those both should be tracked.
9:02
Even if an agent lives like a Firefly kind of comes and it
9:04
goes away. So we, I'd say we break it into
9:07
those two buckets. Tools the adversary is going to
9:09
use and then tools the developer is going to use.
9:12
And you know, our mission is how do we enable the business?
9:15
How do we make it possible to go fast and build with AI without,
9:19
you know, expanding the universe of, of vulnerabilities that we
9:23
know good and well of like credential expansion, like data
9:26
leakage, like injection? Yeah, you, you know, and this is
9:30
maybe it's good to review kind of the the core of your
9:33
technology, you know, being cryptographic hardware connected
9:37
to identity. So talk a little bit about how
9:40
this is better and different than the passwords we're so
9:43
used. To sure what matters the most in
9:49
terms of security to the typical person listening to this
9:51
audience, it's probably their money, it's their bank, it's how
9:56
they make payments. When you pay with a modern
9:59
credit card, when you pay with Apple Pay, when you pay with a
10:04
mobile phone, you're not using a password action.
10:08
What you're using is you're using single device multi factor
10:12
authentication that's based on a hardware backed device bound
10:16
credential. The merchant over Bluetooth
10:19
sends a receipt or a bill to your phone or your credit card.
10:22
If you're using tap and pay your credit card or your phone then
10:26
looks at that bill and prompts you, hey Ryan, do you want to
10:30
sign this? And if you say yes, it's going
10:34
to say smile or it's going to ask for a PIN.
10:36
That's your second factor. It's going to ascribe a hash of
10:39
that factor to the bill, and then it's going to use a key in
10:43
that enclave on that phone or in that credit card to essentially
10:46
sign over the bill and that second factor.
10:50
And because that key is hardware bound, that becomes your first
10:53
factor, your possession. That is then transmitted
10:55
wirelessly back to the merchant, and the merchant can use that to
10:59
clear a monetary transaction later on in that day.
11:03
That key cannot be stolen digitally.
11:05
It's physically locked in that enclave.
11:08
It's never in the general purpose memory of your phone.
11:11
This is what it means to be device bound and hardware
11:14
backed. The finance industry and the
11:16
mobile payments industry worked all of this out over the last 10
11:19
years. And because the CPU
11:21
manufacturers don't want to build specialized chips, we all
11:24
now get this capability for free.
11:26
Every laptop you buy has this. Every drone you buy has this.
11:30
Every mobile device you buy has this capability.
11:33
They're generally only being used at least until we showed up
11:37
for mobile payments or things like a secure boot, Microsoft
11:43
Secure boot. We take advantage of that
11:45
hardware to create a unique key that represents you on that
11:49
device. And we can use a signature from
11:52
that key to prove possession. Just like the mobile payments
11:55
example. We can use a biometric or a
11:56
local PAN as a second factor. And you know, just like when you
12:01
get on an airplane, you have to prove you're the right person
12:03
and you're safe enough to be on the airplane, right?
12:05
No guns, no knives, no bombs. We will also ascribe or append
12:11
to that bill the security posture of your device in that
12:13
moment. Are you safe enough for the data
12:16
that you're actually asking for? So in that moment, what I've
12:19
just described, it's a little bit technical, but their
12:23
credential never moves, right? And 80% of all security
12:27
incidents today are based on credential movement or the
12:30
exploitation on credentials that can move, whether they're your
12:33
password, whether you're TOTP tokens, QR codes, etcetera.
12:37
We eliminate that movement, we drastically reduce that surface
12:40
area and and that's kind of how, at least at a high level, we
12:44
break that initial attack chain. Well, when you say are you safe
12:49
enough? Do do you have a knife in your
12:51
pocket? Like is that like a score?
12:53
Is that a, a, a range? Is that how do you kind of
12:57
determine that? So we have a policy engine that
13:00
really lets customers kind of draw their own risk profile,
13:03
right. So for instance, we have a
13:05
customer who they're a retail organization, they use this in
13:10
their purchasing app. They're mostly concerned with
13:13
Ato. And so they've written policy
13:15
rules that basically say, give me some proof this device isn't
13:19
jailbroken. And that's all I want you to do.
13:21
We have some other customers that let's call them national
13:24
security customers and they want to understand that their
13:28
specific EDR of choice, Crowdstrike or Sentinel One is
13:31
running. They want to know that it's
13:33
running with the right rules. They want to know every process
13:36
that ever asked for authentication.
13:37
They want to know what loader loaded the process into memory
13:40
from what executable was the executable sign?
13:43
Was it signed by an OEM that's actually on their secure S bomb.
13:47
So you know, you can really dial the knob up or down.
13:50
It's very customer dependent and like it's kind of one of our
13:53
philosophies is we want to put the controls in front of the
13:56
customer and it's kind of on them to to decide what risk is
14:00
acceptable and what facet of their business.
14:05
This episode is brought to you by MO Burst, the AI first growth
14:08
engine trusted by iconic startups like Calm, Robin Hood,
14:12
Reddit, Uber and Google. Moburst is a digital, full
14:16
service, mobile first marketing agency that helps companies to
14:19
scale and become category leaders.
14:22
Whether you're finding product market fit or ready to
14:25
accelerate growth, Moburst delivers creative, data-driven
14:29
solutions that help with scale and efficiency and impact.
14:33
At Moburst, we're not just guessing what works.
14:35
We're predicting it. As an AI first agency, Moburst
14:39
uses cutting edge tools to analyze user behavior, adapt
14:43
creative in real time, and cover hidden growth opportunities.
14:47
The result? Marketing that's faster,
14:49
smarter, and built to convert from early stage startups to
14:52
global brands. We've helped hundreds of
14:54
companies overcome complex challenges, unlock new
14:58
audiences, and optimize every metric that matters.
15:01
At Moberst, we believe that creative thinking, advanced
15:04
technology, and data are the core drivers of success at
15:08
scale. If you're serious about scaling
15:10
your business with precision and purpose, it's time to talk to
15:13
Moberst. Visit moberst.com and let's
15:16
build something category defining together.
15:20
Now, you and I know startups are built in boardrooms.
15:23
They're built in the wild during 3:00 AM.
15:25
Bug fixes, botched launches, and that one meeting where
15:28
everything almost fall apart. That's what Code Story is all
15:31
about. It's a podcast where tech
15:33
founders breakdown the make or break moments behind their
15:36
products, What went wrong, what went right, and what nearly
15:39
wrecked everything. Hosted by NOAA Labhart, Code
15:42
Story goes past the highlight reel and into the honest stuff,
15:46
late pivots, brutal bugs, and somehow coming out alive with a
15:49
product people actually love. If you're in the build stage and
15:53
the scale stage or the Please Let this feature work stage, I
15:56
can relate. This show will speak to your
15:58
soul. Listen to codestory@codestory.co
16:01
or wherever you get your podcast.
16:03
Just search Code Story. It'll be right at the top.
16:05
Because great startups aren't born.
16:07
They're forged one hard moment at a time.
16:09
Code story.co is the the risk profile.
16:13
The the more or the less risk tolerant you are, the more
16:18
obnoxious it is to actually get the the passwords to go through.
16:21
Is that kind of like to supply dramatically?
16:24
Yes. No.
16:25
So there's definitely, there's definitely a way to probe the
16:28
user, right. So for instance, you can ask for
16:32
continuous user verification in the policy engine and that would
16:36
pop, pop back to the user every time, every time it happened.
16:40
We do see customers use that for like highly sensitive
16:42
operations. So imagine in imagine an SRE or
16:45
DevOps engineer who works in a critical data center, you don't
16:49
really want them to have long lived session to session timers.
16:52
You really do want them. It's OK for them to pay for
16:55
friction. And in that scenario, we also
16:57
have them do what's called double device authentication.
17:00
So they'll they'll pop a Yubikey in as well.
17:02
And so they're proving off a signature of the laptop and a
17:04
signature of the Yubikey and doing the posture.
17:08
But all of the posture checks that I've mentioned, you don't
17:12
actually have to interact with the end user.
17:14
They can actually occur silently.
17:16
So that's that's kind of one of the keys is most businesses
17:21
don't feel the need to actually stick their fingers in their
17:25
workers eyes for a security reason.
17:28
If they can verify the controls they expect are running on the
17:31
device and the device is behaving in the way they expect.
17:36
Cool. Josh, how are you guys?
17:38
How are you use utilizing? It sounds like there's kind of a
17:42
some like variability in here with the tools to kind of serve
17:46
a lot of different customers needs.
17:47
How are kind of using beyond identity?
17:50
Yeah, so we're. Gonna start with it actually
17:52
because of a requirement for cybersecurity insurance.
17:55
You know, every year I felt that crazy long questionnaire and
17:59
when we got it back one year, they said, hey, you gotta make
18:02
sure you're doing MFA on all these access points.
18:05
And the big, of course, hassle factor with MFA is that, you
18:11
know, pulling out your phone every 15 minutes to press
18:14
something or enter a code or click OK or whatever that might
18:17
be. And we, as a hotel operator, we
18:20
work with several major brands. And so like Marriott, Hilton,
18:23
IHG, and they all have their own MFA that they use to access
18:28
their systems. So we're already exposed.
18:30
So our typical users, at least our corporate office might have
18:33
3 or 4 authenticators that they're using just for their
18:36
brands based on whatever they need for that identity.
18:39
And then if they're on the finance side, every single bank
18:42
has their own little aspect of it.
18:44
I think 1 user I counted up, they had 15 different Mfas that
18:47
they had to keep track of, which is just crazy.
18:50
And so I'm just like, well, anytime we add something to like
18:56
our operation, especially for like a line level folks, we need
19:00
to try and make it as easy as possible.
19:01
Otherwise it's going to create more of a challenge for them
19:06
than we want. And so we were really looking
19:08
for a solution that could be as frictionless as possible.
19:12
And we'd actually gone down the path if another vendor was we're
19:16
ready to sign. And then that kind of fell apart
19:17
at the end due to some changes they had on their end.
19:20
And we, we came across the end of day.
19:23
I'm not sure how we missed them in the first RFP go around, but
19:26
we're glad we did. And they, they spun something up
19:28
really quick. And we'd said, hey, this would
19:32
be a great fit for our users because we have non-technical
19:35
frontline workers that need to log into things.
19:39
And we're coming from an industry that's historically
19:42
behind. We actually, like, I can tell
19:44
you this now because I think everybody's over it, but I,
19:47
until like 2 years ago, there was generic credentials out on
19:50
pretty much any brand you could probably go to, if you had
19:54
worked in that hotel, you'd probably go behind the front
19:57
desk and login. But they've gotten rid of that
19:58
because they realize like, hey, it's important to, you know,
20:02
everyone should have their own login, but it's just still a
20:04
password based login. Type of aspect.
20:08
So we were looking for that solution to do that.
20:12
So how we've used it is it is a little bit of a lift, but we use
20:15
it across for all of our users. You know, once we get them
20:18
enrolled, once we get over that hump, then it's a seamless
20:22
experience for them. You know, they pop up, they're
20:24
logging in on their workstation, they put it in, they're logged
20:27
in, they have what they need from the brand side.
20:29
Now on the brand, sorry, brand and brand sound very similar.
20:34
But on the brand side, whoever that franchise partner is, you
20:38
know, they still have to use their username and password and,
20:40
and go through the MFA with that.
20:42
But on the brand side, once we've gotten them enrolled with
20:45
that passkey on that device, they're golden.
20:48
Like literally until we replace the device or they replace their
20:53
device, their phones, sometimes people forget to, you know,
20:56
it's, it's bound to the, to the actual device itself.
20:59
So you can't just rely on a backup to restore it.
21:02
You got to actually put that passkey on a new phone.
21:06
And that's why like what we've really found the most useful of
21:09
it is these line level folks who are not technical can have this
21:14
incredible level of security that's usually reserved for much
21:18
more structured positions. It sounds like the kind of
21:25
change management then just kind of lands on Apple because
21:28
everybody's used to like double clicking the side and they can
21:31
like scan things. So they're using their phone,
21:34
you know, in a way that's native and and.
21:38
And for us, you know, a lot of it, we used shared workstations.
21:41
So, you know, at a reception desk of a hotel, you know, once
21:44
they've got that, if they go through the effort of putting
21:47
that pass key on that device, they don't even have to use
21:49
their phone, you know, as long as they've logged in that, that,
21:52
that pin code that they create is their second factor with a
21:55
device that we know that they should be using.
21:57
And so they're just in. But yes, they can use their.
22:01
We do allow the I think it's called roaming authentication
22:06
where they can scan AQR code on on their mobile phone where
22:09
they've got the pass key embedded there and login as
22:11
well. Well.
22:15
OK, I'm totally getting it. Thanks for that.
22:17
Now what's interesting is that like when Sam Altman launched
22:22
agent mode for ChatGPT, like the first use case he had was like
22:27
book my hotel for my trip to Las Vegas and it just went in there
22:33
and filled out everything and did all the passwords and boom,
22:36
boom, boom. And it booked him a room.
22:39
Now maybe you guys did like I did is I, I went to do that and
22:42
it couldn't pass captchas. It couldn't get it couldn't get
22:45
past security. Is there a world in, in
22:48
particularly if we're talking about like kind of device bound
22:52
agentic solutions that, that, that beyond identity fits into
22:57
this modality where I built my agent, my personal secretary
23:00
agent to book my accommodations and my flight.
23:03
And it can actually use my, you know, device bound authenticator
23:09
to to get it to go all the way through.
23:13
So I would say, so there's a couple thing, couple things you
23:16
want to break that down in #1 is we are actually working on what
23:21
we call agentic identity. And we actually have some, some
23:26
PO VS of it occurring right now. And the idea is the agent is
23:30
going to have its own identity. And when a user like in your
23:33
example, goes to authorize an agent, what's actually happening
23:38
under the hood is you imagine a blockchain for a minute.
23:41
You have a user who authenticates and proves that
23:44
they're the right person on the right device.
23:46
And in a, in a business scenario, maybe also you're
23:49
concerned with their posture. Maybe in a consumer scenario,
23:51
you're not the agent. You want to know what agent it
23:54
is running on what device with what posture.
23:58
And those two vectors become linked, cryptographically
24:01
linked. Because you're really wanting to
24:03
say this user on this device with this posture authorizes
24:06
this agent on this device with this posture with these
24:10
permissions for a period of time to go off and to go do something
24:14
of interest. And so we're actually, we have a
24:18
couple PO VS of this happening right now.
24:19
And the problem that this is trying to solve is how do I
24:21
actually have evidence? How do I have audit of exactly
24:25
what happened after the fact? Cause agents are like fireflies.
24:28
They come and they go, they disappear like that.
24:30
They, and there's a lot of valid reasons for that.
24:33
We always want to be able to reconstruct who authorized it on
24:36
what behalf to actually do what the the the other part of the
24:40
conversation is reaching out to a site and doing something on
24:42
their behalf that either requires something called
24:46
password vaulting. If the site doesn't really
24:49
participate, right, you don't really want to be putting your
24:53
passwords into agents. And I get it, all sorts of
24:56
reasons why this, but like a high level analogy, if you think
25:00
back to like computer architecture, you know, there's
25:03
this classic lesson that you learn that you don't mix control
25:06
and data. If you think back to compilers,
25:08
there's this classic lesson of no language can analyze itself.
25:12
That's why you have a type language or a time language.
25:15
If you look at networking, there's always a control layer
25:17
versus a data layer. It's a separation of
25:19
responsibilities. When we build agents, those
25:23
separation of control and data kind of get mudged.
25:25
When you think about a template that a lot of agents are working
25:27
for, that template kind of mixes, hey, here's some data and
25:32
here's what I want you to do with it.
25:34
Or it gets end up, it ends up getting mixed in the context
25:37
window of the of the model itself.
25:40
And so like this is kind of fundamentally where where the
25:43
problem arises, where it just becomes really, really hard to
25:47
know. It might bring in data that can
25:48
actually inject control. Do I have data that can leak out
25:54
of the context through some other some other query?
25:57
If the context is long lived and I can't actually track user
26:01
authorizations into the agent, then then short of guardrails,
26:06
it's very, very hard to control that sort of leakage.
26:09
So, yeah, I need to be careful. I could go forever on this
26:12
topic. Number one, it's, it's a
26:14
fascinating set of technical problems.
26:16
But if I were to back up a little bit, you need to give
26:19
your agents identity. If you, if you can, you need to
26:23
actually latch that into your identity and you need to employ
26:26
policy to try and separate those two things out and really
26:29
understand data flow for the purposes of data leakage and
26:33
injection. We're we're very close to that.
26:37
In fact, I would say there's, at least in the enterprise world,
26:40
there's working. Examples of that already, at
26:42
least with us. On the consumer side, I think it
26:47
gets a little bit more, certainly possible, but
26:50
mysterious when you consider the security aspects.
26:54
Very important time to kind of go over that right now,
26:56
especially where we're at right now, because I think a lot of
27:00
organizations, even big organizations have said, Oh my
27:03
gosh, we've got N8N subscriptions for everybody and
27:06
Zapier subscriptions for everybody.
27:08
Figure out how we can make agents to make everything
27:11
better. We don't want to get left behind
27:13
when truly kind of the security measures and like the data
27:16
policies are just kind of out the window.
27:18
And so now they're kind of backing up and be like, what are
27:21
people doing now? So, so they're, they're maybe
27:24
now they're like, well, we spent 152 hours building these agents.
27:28
Are we getting anything out of them?
27:29
So people are like re examining like this is this is kind of the
27:32
backside of that agentic, you know, pendulum that swung 8
27:36
weeks ago, which is fascinating and largely something that you
27:39
said actually, I wanted to dive into a little bit more, which is
27:42
they're like fireflies. They pop in and they pop out and
27:46
you have to give them identity. Now, I think some of this just
27:49
the word that was adapted for them agents.
27:51
And then they're like, oh, see, you have an employee agent and
27:55
see it has a face and it has a voice.
27:57
And so you kind of personify this computer software that
28:01
like, oh, it's like a robot in the computer.
28:04
But truly it is it anything? Is it a thing or is it just a
28:09
Firefly that pops in, it's there for a minute and then it's gone,
28:11
and then it's like, OK, well, who's responsibility or
28:14
responsible for the actions of that agent?
28:20
Ultimately, it's the company that built it.
28:22
I think that answer is very clear.
28:24
But but yeah, So what is an agent?
28:29
I would say, hey, this is going to this is a topic that a bunch
28:32
of people will debate. I'll tell you what I think it
28:34
is. I think of an agent almost as
28:37
like a think of it as an executable, right?
28:40
It's clearly running on a different, a different type of
28:42
operating system, right? We don't have and it's running
28:46
on different type of hardware. So like when you, when you think
28:49
about agents, agents are using LO Ms. to manage a context
28:54
window to try and manage the, the manipulation of data with
28:57
the ejection of control. If an LOM wasn't trained on your
29:01
data, you've got to bring data to it to do anything at
29:04
interest, right? And so the way you bring data to
29:06
the LLM is obviously you can tell it in the prompt.
29:10
You can inject it through legacy services, you can inject it
29:14
through RAG searches, you can inject it through MCP services.
29:19
You quickly get to limitations when you're building these sorts
29:21
of things. But like everything I just
29:23
described, you could see metaphors or parallels in
29:25
computing, right? So like, think of computer
29:27
architecture. I have registers, I have
29:29
reservation stations, I have a LUS and comparators and these
29:33
sorts of things. And when you think about, you
29:37
know, how you build agents in the substrate that executes
29:40
these agents, it's similar but different.
29:42
And so I do wonder if there are parallels to be drawn from it.
29:46
But when we say you want to give an agent identity, what we
29:51
really mean is an instance of a class of an agent has to be
29:58
uniquely identifiable even when that instance is dead and long
30:02
removed. You need to be able to attribute
30:05
what models was it using, what legacy services was it using,
30:09
what MCP services was it using what what, what rag services was
30:13
it using? On what device was it running
30:15
and what was the security posture of the device.
30:17
Those all come together in what we call the agents identity
30:20
vector. And of course we want that to be
30:23
sealed under this thing called an attestation.
30:25
So it's tamper evident as in if someone were to try and change
30:29
the record in the future, it would be obvious that they tried
30:31
to change the record. And, and, and, and the other
30:36
thing is an agent is always acting on behalf of someone.
30:39
In your consumer example, it was acting on behalf of Altman.
30:43
But in a lot of the scenarios that I think are probably more
30:46
near, at least for my customers, it's acting on behalf of really
30:49
that customer, not necessarily any individual person.
30:53
It may be acting on behalf of that customer's customer for a
30:56
moment in time. And so you have to be able to
30:59
track that relationship so that you can understand permissions
31:04
and potential permissions violations and really the
31:07
permissions of the data as it flows through that kind of messy
31:10
ecosystem, which is, you know, the agent ecosystem.
31:17
Weird times. Do you see any of this company
31:22
when you started this company? Like OK, we got to create
31:24
identity for agents. We have to define what an agent
31:27
is and we got to make sure there's some accountability and
31:31
and and record of what it did before it pops out of existence.
31:34
Pretty crazy. Some of it, yeah, Some of it,
31:37
no, we started this company. We, we knew we wanted to to, to
31:41
really solve some hardcore security problems in a way that
31:44
was highly usable, which is kind of the the hallmark of what
31:47
security products are not, right?
31:49
Security products usually slow organizations down.
31:52
And so we anticipated automated workloads, we anticipated
31:55
running on unattended machines. We anticipated kind of cross
32:01
domain signings, like kind of like that dual vector I
32:04
described to the user and the agent.
32:06
We absolutely did not anticipate an agent in the, in the current,
32:11
you know, sense of agentic AI. No, that, that was that, that
32:17
that's, that's been, no, that's just been the fun of the last
32:20
few years, right. Totally.
32:22
You guys have seen, you guys have seen absolute explosive
32:26
growth. Was that something you foresaw
32:28
in the, in the beginning, you know, 2020 or you're like, oh,
32:31
triple, triple, double, double or what do you attribute the,
32:35
the kind of massive growth you guys have experienced?
32:39
Well, no one ever starts a startup thinking they're going
32:41
to fail, right? And you know, especially when
32:46
you start a startup with somebody like like TJ or Jim
32:50
Clark or whatnot, these guys are famous for, what was it?
32:53
I think the first company talk, Jim showed up and gave his idea
32:58
of pumping us up was to tell Netscape stories.
33:01
He's like my first year we did 50 million in revenue.
33:04
In my second year, we did 250 million in revenue.
33:06
In my third year, we did half a billion in revenue.
33:09
And we IPU it at month 18. So there's your targets.
33:15
So I mean, clear, clearly Jim captured a moment in time that
33:19
was special. But like I, I do think we are at
33:22
an interesting moment in time right now with what's happening
33:26
with, you know, the, the big world of capital A and capital
33:30
I. And then of course the small
33:31
world of just helping organizations move into that
33:37
space with their eyes open around security problems, right?
33:42
Not expanding that surface area, but also not slowing down
33:45
development. Yeah, yeah, cool.
33:49
And Congrats for all that success.
33:51
We're looking at the future. You know what's going to be more
33:55
important? A pass key or my Social Security
33:58
number. A pass key clearly, look, the
34:04
government's always going to want to track you and, and, and
34:07
so you're always going to have identifying numbers, whether
34:10
it's that an employee ID or whatnot.
34:13
But in terms of the importance of a kind of like device bound
34:20
so so passkey is an interesting proxy.
34:22
A lot of people will that that name will stick in people's
34:24
brains. The technical jargon, if you
34:27
stick your face under the water is device bound ad test station,
34:31
and that's a way to kind of prove that it's a person's kind
34:36
of logical asset on a very specific targeted device.
34:39
Most of us, I'd say most of the world right now just thinks of
34:42
pass keys as helping a person log into Google or log into
34:45
Amazon. But as we move into this world
34:50
of agentics, as we move into this world of mimicry, as we see
34:55
the AI slop, right, understanding the Providence of
34:58
data is going to be incredibly important.
35:01
So when I say Providence, I mean things like the equivalent of
35:04
like watermark. Now, watermarks are really easy
35:07
to remove, right? So that we're not talking about
35:09
solving the problems of, of, of pulling Providence off, we're
35:14
talking about the Providence of putting things on.
35:16
So how do I know this data came from this person on this device?
35:20
Well, let's, let's drill in more.
35:22
How do I know this data came from the actual camera sensor on
35:26
this device? How do I know it wasn't actually
35:29
molested in some way by an insider at the certificate
35:33
authority before they issued my certificate?
35:36
So this concept of data Providence, I think is going to
35:39
become much more important given everything that's happening in
35:42
the world of AI and device bound attestation or what a pass key
35:45
does, I think is going to move from just being about humans and
35:49
and automated accounts. It's almost be about anything
35:52
that produces data. So you can almost always track
35:56
things back to their point of origin.
35:58
And, and I think some of this is validated when you actually look
36:01
at the big silicon manufacturers in what capabilities they're
36:05
starting to bake into their their devices.
36:09
It's almost impossible to buy consumer grade electronics today
36:13
without a hardware enclave, which is how you do this device
36:16
bound attestation. And, and, and you know, we're
36:20
already seeing people call out this concept of molested data in
36:27
terms of so Acme. Acme is a protocol for
36:30
certificate enrollment. And very clearly in the opening
36:34
lines of Acme, it's like, look, we can't trust TLS for our
36:37
security properties because the way tools are built today, TLS
36:42
is not end to end. It's from your computer to your
36:44
company's proxy. It's from that proxy to
36:46
Cloudflare CDN. It's from Cloudflare CDN to
36:49
Amazon's application layer load balancer.
36:53
It's from that load balancer to maybe you have a service mesh
36:57
controller in your Kubernetes cluster.
36:59
Then it gets unraveled again across some sort of message
37:02
broker and maybe it ends up at some micro service or some some
37:05
monolithic service that's actually going to reply to it.
37:08
All those hot points I just mentioned have your secrets in
37:11
data, right? Most of those hot points are
37:14
probably not managed by your engineers.
37:16
They're probably managed by Cloudflare, Akamai, Amazon, F5
37:19
networks, right? We've all read the news.
37:23
We all know all those companies have had security incidents of
37:26
of 1 shape or another, F5 being the most recent.
37:30
So yeah, Providence of data and not relying on on or realizing
37:36
that you kind of have to build your security at a layer above
37:40
transport or TLS. I think that's, I think that's
37:43
coming. I think that's coming fast.
37:46
Pass keys will enable it. But essentially you're, you
37:48
know, the camera on your phone's going to have a pass key, the
37:51
microphone on your phone's going to have a pass key.
37:53
Anything that generates data is going to have some way of
37:56
attesting to the Providence of that data.
37:59
Just a matter of time if we're thinking about data creation,
38:02
you know, as the technology gets closer to us, Airpods, you know,
38:06
wearable glasses, AR neural link, it's like getting inside
38:11
of us, right? So like the, the cameras will be
38:14
our eyes, the, the microphone will be our mouth, you know,
38:18
speakers. So I guess the speakers would be
38:20
your mouth, the microphone would be your ear.
38:23
I don't know. But I guess the point is like 20
38:27
years from now is Beyond Identity, a hardware company
38:30
that does implants on infants just like sets authentication
38:34
immediately. So one of our one of our
38:36
founders has already taken that step and he actually has
38:40
enclaves in his body. No kidding, no true statement.
38:44
He he can open doors and do other things with a wave of his
38:49
hand. He's he's going full Jedi.
38:54
Look, you're if, if the security space is full of interesting
38:57
people, the start up space is full of interesting people.
39:00
When you combine those two, you get, you get boundary pushers.
39:03
And yeah, look, I, I have no idea what the world's going to
39:07
look like in 20 years. That's, that's beyond, it's hard
39:11
to think about what the world's going to look like in five
39:12
years. But we can certainly see
39:14
patterns, right? We can see human brain
39:17
interfaces already, right? We can see.
39:20
So, so let's think about Neuralink for a minute.
39:23
I'm just spitballing. I don't know.
39:24
These people have never talked to the company.
39:26
But. I can, I, I could absolutely
39:28
believe that data provenance from from some sort of the human
39:32
implant is very, very important from a securities perspective,
39:37
whether it's a neural interface or, or think about any other
39:42
sort of diagnostic capability. How do I know this came from
39:46
this person? How do I know it wasn't
39:48
manipulated or doctored in some way?
39:51
We know a huge part of the criminal element, right, as it
39:55
was demonstrated over the last three years, doesn't care about
39:58
the general health and safety of the world, right?
39:59
Especially if it's not their locale, right?
40:02
We've seen those emergency rooms get shut down.
40:04
We saw these things get shut down.
40:06
If ransomware could start encrypting medical data, could
40:08
start essentially testing the veracity of where did this thing
40:14
come from? Pay up or I'm not actually going
40:16
to reveal the origin like more. There's a lot more crazy
40:22
scenarios in front of us than in behind us, which is again why I
40:25
think provenance. Of data is probably going to
40:29
become a first class thought in any data producing platform.
40:34
That's encouraging. Yeah, the if you're fans of what
40:40
is it? The The Expanse, Snow Crash, all
40:45
of these dystopian science fiction novels, it turns out
40:49
they're, they're probably pretty accurate, at least on the
40:52
problems of the future that we will end up facing.
40:57
But you know, the good news is there's solutions to these
41:00
problems, right? Like you can't actually
41:02
attribute the Providence of information.
41:05
I mean, actually, I guess the biggest good news that at least
41:08
Beyond Identity has to say is if a credential doesn't move, it
41:12
can't be stolen. And if you use a hardware backed
41:16
device bound credential, it literally cannot move.
41:19
It's never in memory. Totally.
41:22
Now that now that we're sufficiently terrified of our
41:26
robot overlords coming for us, what who would be in the in the
41:31
best position? Who are the best customers right
41:33
now for Beyond Identity? You know, is it, is it like an
41:36
early stage startup it or is it dependent on kind of the, the,
41:40
the Providence of the data and like the, the security of the
41:45
data? Who are the best customers and
41:46
who should definitely be considering this right now?
41:50
From a technical perspective, the best customers for us are
41:53
people who already have an identity system.
41:57
If you don't have an identity system of record, you probably
41:59
shouldn't start with us. The typical deployment of our
42:02
product we plug into your identity stack and we help
42:05
defend your organization. We don't displace your identity
42:07
stack. We have put a lot of tooling in
42:12
place to make that as easy as possible.
42:16
Some new things that Josh, unfortunately you didn't get to
42:19
experience as a long time customer, but but yeah, auto
42:23
enrollment through an MDM is actually being devote now.
42:29
But from a business perspective, it's really anyone who wants to
42:34
eliminate 80% of their sock incidents, right.
42:37
If you follow the average by whether whether it's crowd
42:41
strikes threat report, Mandiant threat report or Verizon DBR
42:45
threat report, 80% of the incidents that are hitting your
42:47
organization are identity related and we stop them.
42:51
Amazing. Cool.
42:52
And Josh, I mean you guys seem pretty forward acting, at least
42:56
you could identify some of these problems early on and and take
42:59
steps to remediate against them. Like what are the next 1224
43:03
months? What's what's big in security
43:06
and IT for hospitality? Well, I think for us, we're just
43:10
trying like, I mean, I'm seeing first just for Brad, it's like
43:12
what our goals are is just really to get that adoption
43:15
level up because it is a little bit of a lift, you know, to get
43:17
it's a it's a mind shift of like, hey, you know, you got a
43:20
new employee. We have a lot of line level
43:23
employees that kind of there's turnover that happens in our
43:25
industry. And you know, some people come
43:27
in and they're they're not used to using it like a device bound
43:30
passkey like this. It's a little bit of a mind
43:32
shift for them. Same thing with like leaders to
43:34
get them like on. Board like when we're opening a
43:35
new. Hotel like hey, we don't use a
43:37
password. This is how it works.
43:38
You know, it's kind of like we just recently opened a hotel in,
43:43
in the, in the Denver area. And I remember the GM or the
43:45
AGM, one of them was like, wow, this is really great.
43:47
Like it's like auto magic, you know, like once they're
43:50
enrolled, you know, it just worked.
43:51
But then when they got over to the brand sighted stuff like,
43:54
Hey, how do I get this to work? You know, on this utility, I'm
43:57
like, oh, I'm sorry that that can't work because it's not a
43:59
brand application. That's like our partners
44:01
application from the brand. And so unfortunately, like
44:04
there's no, there's no way that you can use our pass key to
44:07
authenticate into their system. So I would, if I had a wish list
44:12
of what I could control outside of Brandt, I would love for on
44:16
some of these organizations, especially in the franchise
44:18
model where we're working with a brand of this providing, you
44:21
know, a lot of resources and intranet their applications and
44:25
all of that that they're developing and working on and
44:27
they're making secure. But The thing is they're using
44:31
their own identity structure. It'd be really great if there
44:35
was in the future a way that we could create some kind of trust
44:39
model where like, hey, Brands Hospitality Group uses Beyond
44:42
Identity, this is our IDP, so we can trust automatically.
44:45
You know, there's things that are happening on the brand level
44:47
so that we could integrate that in.
44:49
Like that would be a phenomenal win for us because things are
44:53
changing so fast and it is hard to keep up when you've got one
44:58
way of doing something in one way.
45:00
And we're trying to be a little bit more forward thinking with
45:03
the device from Passkey. But then we have a brand partner
45:05
that's still using, you know, duo push or I mean, not to pick
45:08
on duo, but you know, a push or, you know, you got to get, you
45:11
know, put this app on and pick this character or whatever it
45:15
is. It would be great if there was
45:18
some kind of convergence that could happen so that we could
45:20
use trust, you know, in, in between our different identity
45:24
structures to allow that to happen.
45:29
So I think I diverged, I diverged a little bit from your
45:31
main question as far as like what our security things are in
45:33
the next 1224 months. But really to sum it up is just
45:37
we're trying to really hone in on that beyond identity passkey
45:42
and making sure that all of our applications that we use do
45:45
support like single sign on through Sam Lehro, IDC or
45:49
something like that. So that we know that they're
45:52
when they are logging in, they're authenticating against
45:54
that passkey and not with a legacy password.
45:57
And I think that's the biggest challenge that we face is as
45:59
we're working with vendors from a large, you know, vendors have
46:02
run for 20 years and our legacy platforms are they're trying to
46:05
modernize. And then there's new vendors
46:07
that are coming up and like, Hey, we've got this great
46:08
solution to solve this new problem, you know, that you're
46:11
dealing with an hospitality. And then we ask like, hey, do
46:13
you support like, you know, how does, how does your identity
46:16
structure work? And they don't support single
46:18
sign on or have any mechanisms for us to integrate that into
46:21
our passkey. Well, then that's kind of it's
46:23
getting to a point where now we're like, hey, get that fixed
46:25
first before you come talk to us because we're not interested in
46:29
adding another password to our folks after we've finally done
46:32
the work of eradicating them across all of these applications
46:34
that we used. Totally.
46:38
You're like, dude, we're we're injecting your credentials
46:42
subcutaneously. Yeah.
46:45
Well, hopefully we're in the future.
46:47
Yeah, hopefully we don't start doing that during open
46:49
enrollment. That would be a little bit
46:52
deterrent probably in our in our world.
46:57
You never know, people catch on. Just got to get a couple
47:00
influencers doing it, then everybody's doing it.
47:03
I tell you. I mean, just if we're talking
47:05
about like, I got a, a Nissan Aria, you know, which is an
47:09
electric car and they're so cool.
47:11
They're like spaceships and just if I have my keys on me and I
47:13
walk by and it like blinks at me and says hi.
47:16
So a lot of this like device location stuff, it's like if I
47:20
could put that into my hand, that would be great, you know,
47:22
and then I can just wave my hand and like, you know, remember Kit
47:26
from Knight Rider? Now I could just talk to my car
47:29
and it could come find me and getting.
47:31
Closer and closer. There's like a paradigm shift
47:35
that has to happen with people. Like everyone's so used to the
47:38
idea of like, hey, started the company, this is my username and
47:41
this is my password, you know, and the idea that now you get a
47:43
pass key and it's, it's bound to something.
47:46
It's a mental 5th. Like I was just going through
47:48
North Dakota. That's that's where my driver's
47:50
license is issued. They just got the mobile
47:53
driver's license deployed like I think October 1st.
47:56
So I was super excited about that.
47:57
Signed up right away and I was excited to try it at the TSA
48:00
because I know that they, you know, have those fancy readers
48:03
now. So I'm at the Fargo airport and
48:05
they've got it and I go online and they see they've got their
48:08
signed digital idea. I'm all excited to give it a
48:10
try. And I'm like, the guy's like
48:12
driver's license and I'm like, oh, I've got it on my phone.
48:14
He's like, no driver's license. I'm like, but you're signed.
48:16
I'm like, but I didn't want to argue with the TSA because
48:18
that's like worse than carrying a bomb.
48:20
And so I just gave him my driver's license and it was
48:24
really in the morning and there's a flaker in front of me.
48:26
So they went in and one of them got random.
48:28
And of course that defaults to you if you're the next in the
48:31
TSA pre-check line. So I'm the guys like swabbing my
48:34
shoes. I just said, Hey, can I ask you
48:36
a question? And I just asked him.
48:37
I said, hey, I saw you guys have that sign.
48:38
I have the, you know, we just got the mobile driver's license.
48:40
But did they, it sounds like your system doesn't actually
48:43
support it yet. It was a little bit, you know,
48:44
like what's the story there? And and then the supervisor
48:48
overheard and she's like, oh, she said, actually we do have
48:51
it. I just don't think anyone's ever
48:53
come through with it yet. And so like, they've been
48:56
trained on it, but they didn't really know how it worked.
48:58
They maybe like seen it, but nobody's actually ever tried it
49:01
yet. And she said, do you mind going
49:02
back and, like testing it? I'm like, well, I, you know, I
49:06
just finally got through. I don't really want to, you
49:09
know, do this all over again. It's like, no, no, I'll escort
49:12
you. So she took me back there and
49:14
like, the guy's a super long line.
49:15
So I'm sure they were annoyed too.
49:17
Like why is this guy back again? But like, we did it and it
49:20
literally took like 2 seconds And the guy's like hold, it's
49:22
amazing. You know, the, the agents on
49:25
like the TSA agent who is doing the ID checks.
49:27
He liked it because it made his life so much faster and more
49:31
efficient. But like his mind that he was
49:33
still so trained to be like, hey, no phones.
49:35
I need the ID because that's, that's just what it is.
49:38
And I think that's just a small glimpse of the change is going
49:41
to happen. I think people just have to get
49:43
used to the idea of like, your phone is no longer.
49:49
I mean, I don't have to say this properly, but that secure
49:53
enclave that's now exists in these electronic devices where
49:55
you can store these cryptographic information, solve
49:58
so many problems. But there's such a mindset with
50:02
people because of before we issued unsecure things and bad
50:07
stuff happened. If you went on your bank on your
50:09
phone or you put in your like Social Security number on your
50:12
phone, you know, like in an unsecure way.
50:14
And so there has to be this mental shift of like, hey, this
50:16
is the way the world's trending and this is how we do it.
50:19
So how do we use the proper syntax and like training to kind
50:22
of help people through that transition?
50:27
Because like, it is the total shift of how we've done things.
50:34
Or a habit. So you guys, this has been
50:40
great. I really appreciate the time.
50:42
I know we're slim for time right now.
50:44
I want to give a couple of shout outs to some charitable
50:47
organizations that you guys support to make sure that we're
50:50
giving them some love. I want to give a shout out to
50:52
stonewoodny.org, which is a First Harvest Food pantry.
50:59
So it looks like it's, it's making sure that food pantries
51:02
are, are stocked with local harvests, which looks really
51:07
cool. Also shout out to vetdogs.org,
51:11
which are service dogs for veterans.
51:13
I, I come from a marine family and veterans are near and dear
51:16
to my heart. So definitely, you know, service
51:19
who doesn't want a golden retriever or any kind of a
51:22
Labrador at that point, you know what I mean?
51:24
So like definitely a shout out to them for for taking care of
51:27
our vets and then a safe house project, which is sex
51:32
trafficking detection training, a survivor LED training designed
51:35
to empower you to spot, report and prevent sex trafficking
51:38
where you live or work. So shout out to those
51:40
organizations, gentlemen, I really appreciate the time.
51:43
Thank you so much. Hey founders, let me ask you
51:48
this. Are you building with AI or
51:51
getting left behind by it? If you're serious about scaling
51:54
smarter, faster and bolder, you need to check out AI for
51:59
founders.co. It's your go to resource for
52:02
cutting through the noise and actually using AI to grow your
52:05
startup. Each week, AI for Founders drops
52:09
a sharp, no fluff newsletter, dives deep with expert guests on
52:13
the podcast, and hosts monthly workshops where you can get your
52:17
hands dirty with the real tools and strategies.
52:20
Whether you're precede or Series B and beyond, launching
52:24
something new or just staying sharp, AI for Founders gives you
52:28
the edge you need to win in 2025 and beyond.
52:32
Hit AI for founders.co to subscribe to the newsletter,
52:36
tune into the podcast, and grab your spot in the next workshop,
52:40
and it's all free. Build smarter, move faster,
52:43
think bolder with AI for Founders.
52:46
Now back to the show.