CYBERSECURITY: THE CASE FOR ACCURATE INTELLIGENCE AND DATA- Dr. Zakir Durumeric

Parliamentary Intelligence-Security Forum
3 followers
Mar 2, 2026
25th Parliamentary Intelligence-Security Forum
London🏴🇬🇧- August 31- September 1, 2023
-Dr. Zakir Durumeric
Assistant Professor of Computer Science at Stanford University and Co-Founder of Censys
AI Generated·May contain errors
All right, I appreciate I am the very last talk
between the beginning and the end.
Thank you to the forum organizers
for having me here today and for a wonderful two days.
My name is Akira Jurumurch.
I'm a professor of computer science at Stanford
and also chief scientist at Census,
which is a company that broadly does very large scale
mapping of internet infrastructure and exposure.
And so in short, I spend my days often looking
at global trends today.
What does exposure look like?
What does vulnerability look like?
What does attacker behavior today look like?
And given that perspective, perhaps unsurprisingly,
I want to talk a little bit about
the challenges we're facing today.
There's no doubt challenges that are gonna surround
new machine learning algorithms,
large language models, quantum computing,
but these aren't the types of problems
that are really our challenges today.
Many ways these are challenges of the future
while we're still today, I think,
trying to fix many of the problems from yesterday.
There's certainly gonna be nuance
in these new machine learning models,
in these new technologies,
but I think what we need to talk about as well
is the more fundamental foundational problems
that Brian touched on and I want to touch on as well.
The reality is that companies today, utilities,
universities, hospitals, corporations, governments,
face many of the same nation state adversaries
or nation state funded adversaries,
but the reality is that breaches today
are not caused by unknown attacks, zero day attacks.
As you alluded to, these are oftentimes vulnerabilities
that have sat there for hundreds of days.
The reality is there's no need
to use expensive zero day attacks
or to develop zero day attacks
when it's simply not necessary.
It's simply possible to look for a vulnerability
that's been there for six months
or been there for a year or has been there for years.
In many ways, that's the state of the art
in many of the attacks today
is we are exploiting vulnerabilities
that have existed for months or years on these systems.
And our adversaries know this, right?
That there's just simply no need
to develop these more complex attacks
to think about a lot of these issues
around issues like quantum security,
when indeed we simply haven't patched data.
And I think, honestly, this has really been epitomized.
If you look at the recent series of data breaches
that have simply been caused by storage buckets
on the internet, right?
We've seen massive data breaches,
governments, defense contractors, hospitals, universities,
again, who've lost massive amounts of customer data
about patient data, not due to any security vulnerability,
but because they simply put it out on the internet
and had no idea that it was there.
Those are the attacks that we're facing today.
Multiple recent studies have come out
in the last year or two.
What they have found is that on averages,
companies patch between 10 to 25%
of the vulnerabilities that they have on the internet.
So less than a quarter of vulnerabilities
on the internet will actually be patched.
And part of this problem is absolutely a resource issue,
right, we have an absolute talent shortage in this space.
We don't have enough people
to patch many of these vulnerabilities.
But much more than that is that we live in a very,
I would say, data-rich but signal-poor environment.
We simply today don't know how to prioritize
what we do need to fix with the limited talent that we have.
And there is always the hope
that commercial threat intelligence would fix this,
that commercial intelligence would help us understand
which are the vulnerabilities that we needed to fix first,
which of the IP addresses or hosts
were the ones that were malicious.
And I would argue that in many ways,
the market has failed in this space.
Most CISOs will tell you
that they don't look to vulnerability for feeds anymore,
they don't look to threat intelligence feeds,
and researchers see the same thing.
When they compare these threat intelligence feeds
to see what are the most important issues,
these fees don't overlap.
You can buy dozens of different feeds
and each one of them will tell you something different,
and many of them will have very little to do
with the attacks that are actually taking place today.
And as a result, we simply do not know
what to prioritize in many cases as an industry.
The state of the art oftentimes is looking at Twitter.
It's understanding which CVEs or vulnerabilities
are being most actively discussed on Twitter.
And that's how we prioritize in this industry.
We look at the unique identifier and we say,
which one is being mentioned most?
Which one has started to be mentioned more this week?
And that's how we prioritize
what's gonna pass on the internet.
And there's been great guidance recently.
If you look at resources like
the CISA Known Exploited Vulnerabilities list,
this has been a fantastic resource
that folks have been looking at,
and I think it's something for us to push further.
But I think we also have to recognize
that oftentimes what ends up on these lists
is also months behind, weeks to months behind.
Exploits end up on that list
after most of these systems have already been exploited.
In fact, that's sort of the definition of the list.
It's known exploited vulnerabilities, right?
So we add something to a list
and tell people to prioritize
based on what has already been exploited on the internet.
And again, that's sort of the state of the art today.
And so this brings me to my first point,
which actually matches Brian's first point,
despite the fact that we did not coordinate on this.
We need to do a much better job
helping people prioritize what they need to patch,
how they use the limited resources that they have.
And the only way we're gonna get
significantly better doing this
is by having much better collaboration.
And this has to be collaboration
between private industry and governments,
between intelligence.
Each one of these parties has a unique puzzle piece here.
But right now we are not communicating.
And it's something that we are gonna continue
to be retroactively learning what's going to be exploited
based on what has already been exploited
until we figure out how we're able to collaborate
better together between these two groups.
Governments can also play a crucial role though,
not in only developing policy,
but also helping to disseminate tactical data.
In many cases, we know how organizations
are vulnerable on the internet today.
Governments can help these organizations
understand what they need to patch.
They can also help them to understand
what is actively exploitable.
And in many cases, this is because
there's a long tail of critical infrastructure companies
that simply don't have security teams at all.
If we look at hospitals, if we look at utilities,
a lot of these are small organizations
that if we're lucky, have a CISO.
If we're lucky in some cases,
maybe have a dedicated IT team.
And this is a place where government can help.
In a lot of these cases,
private industry knows about these vulnerabilities,
knows about what is exploitable on the internet,
and it's a matter of routing.
It's a matter of getting the right data
to the right people to disseminate this data
and to help people prioritize
which of these vulnerabilities
are the ones that they actually need to look at,
which one is actually important.
To give you a sense, most of these organizations
are facing hundreds of different vulnerabilities,
perhaps on a weekly basis.
It's something that is so far untractable
in terms of what these organizations can fix in practice.
We have to tell them these are the two or the three
that actually matter.
And that, again, is just sort of the state of the world,
the state of the industry,
and I think we all wish it were better.
I think we all wish that we were worried about issues
like quantum computing, but it's simply not.
In other cases, it's not even a security issue.
If we look when, on the commercial side,
when we talk to organizations
about why vulnerabilities haven't been patched,
what we found in many cases is that
even for probably the most advanced companies,
for global 1,000 companies,
companies will not be aware of 25 to 30%
of the assets they own,
of the assets they have on the internet.
In some cases, a lot of these companies
may not even be aware on the security team
of the subsidiaries that they have
and that they're responsible for.
And again, this is one of those conversations
I probably wouldn't have believed
until I had it over and over again,
where security teams didn't even know
the companies that they were responsible for
because organizations have begun to move so quickly
because of M&A activity.
But again, that's where we sit.
And so the world has become much more complex
through M&A activity, through cloud,
through work from home.
In many ways, we haven't caught up to this.
And we haven't caught up to many
of the fundamental problems we're gonna have to solve
in order to actually even get to security.
Do we even know what we own?
Do we have a list of the assets
that we need to consider patching,
of the software that we need to look
for vulnerabilities for?
And again, that's simply where we're at today as an industry.
And so this brings me to my second point.
We need to build a more fundamental understanding
of what works and what doesn't work,
where these problems emerge in practice.
Oftentimes, they're organizational issues.
They're more foundational issues to how organizations run
rather than specific security threats
or specific technologies.
And again, we're only going to make sense of this
through collaboration,
by bringing government sector together,
bringing academic researchers together,
bringing these organizations together,
private industry who have much of this data.
Even today within the private sector,
much of this data is fractured.
Certain organizations know how, what is vulnerable.
Certain organizations know what is exploited.
Certain organizations know which companies
have actually paid out ransomware demands.
None of that data is connected.
And there's a role government can play here
by bringing together the severely fractured ecosystem
to start to make more fundamental progress,
to understand end to end,
which vulnerabilities actually lead to exposure,
which lead to ransomware.
We don't know today.
Third, we need to address the fundamental talent shortage
that exists.
And again, this is something that's only going to happen
through collaboration.
But today, we have a talent shortage on the ground,
folks trying to actually prioritize these vulnerabilities.
We also though are missing key talent
in how we discuss these issues, how we communicate.
How do we communicate and educate
so that we can have successful conversations
between boards and CISOs?
How we can talk about this more as a risk mitigation
issue than necessarily a security problem.
But we need to come together to understand
who we need to educate
and how we're going to help address this talent shortage
because this isn't going to go away.
I think we like to look at AI as a silver bullet,
but silver AI is not going to solve
these organizational challenges for us.
They're not going to solve the fact
that we don't have the data in the right places
to start to analyze it.
Until we solve some of these issues,
AI is going to do very little for us.
It's only once we have this data,
once we understand these problems more foundationally,
that we're going to be able to harness AI
to be able to solve some of these issues.
It's absolutely a tool.
It's absolutely a tool we should be considering how we use,
but we as an industry in many ways aren't there yet today
until we better understand the problems that are happening
and why people are being breached.
That's what I have.
Thank you.
