This episode is presented together with:
Unveiling TITAN AI: A New Era of Threat-Informed Third-Party Risk Management
One minute, AY is telling you about Russian botnets built from compromised routers being used to spy on Ukrainian infrastructure and the next he’s complimenting your throw pillows in the podcast studio.
I think that contrast tells you something important about why I keep inviting him back.
It hasn’t been long since our first conversation. But the world moved. Agentic AI continues to explode onto the scene.
OpenClaw went from a cool open-source project to a security crisis in the span of weeks.
The speed at which threats are materializing has shifted from days to microseconds.
And AY, as usual, was not only watching it unfold but his team was publishing the research that quantified exactly how bad it was.
This is the second installment of conversations with AY on Audience 1st Podcast.
The first episode was about origins, philosophy, and the foundational ideas that built Security Scorecard.
This one is about what happens when those ideas collide with a world that is changing faster than anyone predicted.
OpenClaw is an open-source personal AI agent.
It connects to your email, your calendar, your WhatsApp, your browser.
It can send messages on your behalf, manage your schedule, execute code in your terminal.
People loved it. It went viral.
Thousands of people rushed to install it because the promise was irresistible - wake up to a briefing on your code commits, your weather, your meetings, all handled by an AI that never sleeps. The problem is that almost nobody thought about security before they deployed it.
Security Scorecard’s research team scanned the entire internet and found tens of thousands of OpenClaw instances left completely exposed. No authentication or encryption.
Many of them were named in ways that let you identify the person and the company they worked for.
Hackers could see exactly which instance belonged to a bank, which one belonged to a supermarket, which one belonged to your neighbor.
And because these agents have privileged access to everything - email, messages, calendars, browsing history - breaking into one was like getting handed the keys to someone’s entire digital life.
OpenClaw is not an anomaly.
It’s the first wave.
Every agent framework that prioritizes convenience over security by default is going to produce the same result.
People will deploy it because it’s useful, they’ll skip the security configuration because it’s not the default, and attackers will find the exposed instances before the organization’s security team even knows they exist.
AY’s team built a tool - declaw.io - to help people discover their own exposure.
The fact that a tool like that needs to exist tells you everything about where we are right now.
In the real world, we’ve been trained and conditioned.
You walk into a dark alley at night, you look behind your shoulder.
But in the digital world, those defenses disappear. You click and then you think, instead of thinking and then clicking.
It explains why someone will plug a security camera into their home network and leave the password as admin/admin.
It explains why employees upload confidential documents to ChatGPT without a second thought.
The physical world has friction built into it - distance, darkness, locked doors - that forces you to pause and assess before you act.
The digital world has been engineered for the opposite.
Every interface is optimized to reduce friction, to get you to click faster, to remove the pause that would otherwise protect you.
If the information about OpenClaw’s vulnerabilities was already publicly available, if the research was out there, and organizations still didn’t act, then is the problem really visibility? Or is it something else entirely?
His response was nuanced in a way I appreciated.
The problem is that organizations are drowning in data.
If you have visibility into ten thousand issues and no mechanism to tell you which two actually matter right now, then your visibility is functionally useless.
Enter: contextualized visibility.
Not just seeing what’s exposed, but understanding what it means in the context of your specific environment, your specific threats, your specific business.
A hospital needs to know about the threats being used against other hospitals.
A hedge fund needs to focus on data exfiltration patterns in financial services.
A SaaS company has a completely different threat profile than a manufacturing firm.
Yet most security tooling treats them all the same. Here are your vulnerabilities, ranked by severity, good luck figuring out which ones actually matter to you.
Every vendor in the market is selling visibility. Visibility has become table stakes.
The differentiator is context.
Imagine that every single second, you hire ten new interns. They have access to your systems, your data, your customer information.
The next second, those interns are gone, and five new ones show up.
Some of them might make mistakes. Some of them might expose sensitive systems. And you have no idea who they are, what they did, or whether any of them left the door open.
That is what the agent landscape looks like right now. These are not persistent systems you can audit at your leisure.
They are ephemeral, privileged, autonomous processes that spin up, execute tasks, and vanish, often before a security team even knows they existed.
Traditional identity and access management frameworks were not built for this.
They were built for humans who log in, do their job, and log out on a predictable schedule.
Agents do not operate on a predictable schedule. They operate on demand, in parallel, at machine speed.
For anyone building in the identity security space, this is the architectural challenge of the next five years.
How do you tie an agent to an identity? How do you establish accountability when the entity performing the action existed for three seconds and then disappeared?
AY raised this as a problem he’d want to work on if Security Scorecard didn’t exist.
In our first conversation, AY said something that became a throughline for the entire episode: what you can’t measure, you can’t improve.
So this time I pushed him on the inverse. What happens when you can measure it, when the data is right there, the dashboard is lit up, the risk is quantified, and the organization still doesn’t act?
He said that many companies say cybersecurity is important the same way they put culture values on a wall.
But if the behaviors don’t match the stated values, then you’re not actually practicing it.
He pointed to large banks running ancient versions of software because upgrading is expensive and might break trading systems. The risk is known. The measurement exists.
The action does not, because the cost of fixing it feels bigger than the cost of ignoring it, until the day it isn’t.
We talk about maturity as a technology problem - do you have continuous monitoring, do you have threat intelligence, do you have automated response?
But the gap between measurement and improvement is a behavioral gap.
It is an organizational willpower gap.
It is a leadership accountability gap.
And no amount of tooling will close it if the people at the top of the organization are not willing to treat security investment as a real priority and not just a compliance checkbox.
Organizations are stretched thin. Budgets are flat. The CISO is being asked to do more with less while the C-suite pours money into AI initiatives.
The behavioral failure is real, but it is often the rational response of overwhelmed people operating inside broken incentive structures.
Fixing it requires changing the incentives - tying executive compensation to security outcomes, making risk posture a board-level KPI with real consequences, and giving security teams the authority and resources to actually act on what they can measure.
AY walked me through SecurityScorecard’s maturity framework for third-party risk management, and what struck me was how clearly it reveals the distance between where most organizations think they are and where they actually are.
The framework has three stages.
The first is static, point-in-time assessment. You catalog your suppliers. You run a vulnerability scan or a penetration test. You complete an annual questionnaire. You check the box. This is where the vast majority of organizations live, and many of them believe this constitutes a functioning security program. It does not. It constitutes a snapshot - a single photograph of a landscape that changes every second of every day. AY compared it to putting the results on a top shelf drawer and letting them collect dust until next year. That is not risk management. That is documentation.
The second stage is continuous assessment. This is where you move from once-a-year to always-on. You are being alerted in real time about what is exposed in your environment and your suppliers’ environments. Missing patches. Malware indicators. Configuration drift. The shift from stage one to stage two is significant, but it is not the finish line and this is where I think a lot of organizations get stuck. They achieve continuous visibility and assume the job is done. It is not.
The third stage is threat-informed detection and response. This is where you stop treating every vulnerability as equally urgent and start contextualizing threats based on who you actually are. Organizations are spending enormous amounts of money protecting against threats that are not relevant to them while ignoring the ones that are.
SecurityScorecard works with over thirty-two hundred organizations across forty-six countries. Nine of the top ten banks. Major insurance companies. And even among that cohort,
AY was candid that every organization is at a different stage of the journey and most are not as far along as they believe. R
egulated industries tend to be further ahead, not because they are inherently more sophisticated, but because compliance mandates like PCI-DSS and HIPAA force them to invest.
The organizations without that regulatory forcing function are, in many cases, still living at stage one and calling it a program.
I asked AY whether “continuous” has become a buzzword. He didn’t dodge it. He said that almost every term in cybersecurity is a buzzword at this point and that saying it and doing it are two completely different things.
Fair enough. But then he put a number on the gap.
Most third-party risk management programs operate as a once-a-year activity. One assessment and checkbox. One day. Which means there are three hundred and sixty-four days where nobody is verifying anything.
Three hundred and sixty-four days where a supplier could be compromised, a configuration could drift, a new vulnerability could be exploited and nobody would know because the assessment already happened and the results are sitting in a drawer collecting dust.
The adversaries are not operating once a year. They are operating every second of every day. And so when a vendor puts “continuous” on their website, the real question is: does your definition of continuous match the adversary’s definition of continuous?
The only meaningful definition of continuous in the current threat environment is always on. Twenty-four hours a day, three hundred and sixty-five days a year, without gaps.
AY said you have to move from robustness to resilience. Stop trying to make it impossible for an attacker to get in. Start assuming they’re already inside.
A few months before the Russian invasion of Ukraine, the Ukrainians proactively invited hackers to break into their own infrastructure.
They deliberately discovered their own vulnerabilities before the adversary could exploit them. It meant that when the actual attack came, they had already seen many of the paths an attacker would take and had hardened accordingly.
If you are not actively and continuously trying to break yourself, using the same tools and techniques the attackers are using, then your security posture is based on assumptions, not evidence.
You assume your configurations are correct. You assume your patches are deployed. You assume your third parties are secure. Assumptions are what get breached.
Boards are pushing AI adoption for Wall Street optics while CISOs are left picking up the pieces. I asked AY whether the disconnect has gotten worse.
More and more company budget is flowing into AI initiatives. At the same time, cybersecurity budgets are staying flat or actually declining.
The board is mandating AI adoption from the top, driven by competitive pressure and investor expectations, while the security team is being asked to govern and protect these new deployments with the same resources they had before the AI push started. In some cases, with fewer resources.
AY drew an important comparison. Application security for traditional software development is a well-established discipline.
People know the OWASP Top 10. There are mature static and dynamic code analysis tools. Decades of collective wisdom have been codified into frameworks and best practices.
AI governance has none of that. The tooling is nascent. The frameworks are early. The startups working on it are still finding product-market fit.
And yet organizations are deploying AI into production environments at a pace that far outstrips the maturity of the security ecosystem around it.
This gap between AI adoption speed and AI security maturity is not closing. It is widening. Every quarter that passes without an organization investing in AI governance creates compounding risk.
The companies building tools that help CISOs quantify and communicate that risk to their boards, not in technical jargon, but in the language of business exposure, are solving a problem that is going to get much more expensive before it gets any cheaper.
AY described an internal culture model at Security Scorecard that I think has broader implications for how any security organization should think about innovation.
He calls it Pirates versus Navy. The Navy is your traditional team, handling scalability, maintaining systems, doing things the established way.
The Pirates are a small, isolated R&D group with explicit permission to experiment, break things, and operate outside the usual constraints.
What I find interesting about this is that it’s not a new idea in the abstract - people have been talking about skunkworks teams for decades - but AY is applying it specifically to how security organizations need to evolve in the age of AI.
He also backs it with structural incentives.
Security Scorecard allocates bonus pools to departments based on AI tool adoption.
They mandate reporting on the percentage of code written by AI.
They test candidates on AI competence during the hiring process.
That last part is what separates this from the generic “we encourage innovation” talking point that every company puts in their values deck.
Innovation culture is about making creativity economically rational - rewarding the people who experiment, protecting them when experiments fail, and structurally separating them from the operational teams whose incentives are optimized for stability, not discovery.
Most organizations say they want both.
Very few actually build the structural conditions for both to coexist.
If the first conversation with AY was about foundations - about where conviction comes from, why language matters, and how to build something that changes an industry - this second conversation was about velocity.
About what happens when the pace of change outstrips the pace of defense.
About what it means to operate in a world where the attacker moves in microseconds and the defender is still trying to get a budget approved.
The throughline of everything AY said in this episode can be reduced to a single design principle: the secure path and the convenient path have to become the same path.
Humans, without exception, will choose the convenient thing because the digital world has been optimized for speed and convenience at every layer and security is almost always a friction point that sits outside the natural workflow.
This conversation felt different from the first one.
The first time, AY was telling me about the world he wanted to build.
This time, he was telling me about the world that is arriving whether we are ready for it or not.
The optimism is still there. But it comes with a sharper edge now.
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.