AI Adoption, Trust, and the Fundamentals: Lessons From Optiv Field CISO Nathan Wenzler

Watch on: YouTube, Spotify, Apple

-

Nathan Wenzler has worked in cybersecurity for close to 30 years and now serves as a Field CISO at Optiv, where his role is to advise security leaders across the company's client base. By his own description, sorting out what is real in AI from what is marketing has been most of his job for the past six to twelve months.

On this episode of the Cybersecurity Ecosystem Show, he explained why the organizations succeeding with AI are the ones taking a methodical approach, why AI differs from earlier technology shifts in a way that matters for security, which fundamentals now require attention, how the CISO role has changed, and how the buying market for security tools is behaving under the current uncertainty.

Here are the main lessons from the conversation.

The organizations succeeding with AI have a plan

Nathan's central observation is that the pressure to adopt AI quickly has pushed many organizations past the basic planning they would apply to any other investment.

"The organizations that are starting to have success with it, they're the ones who fall back to having a plan and a strategy and building governance and talking to their organization," he said. "I hate to say taking a slow approach, but it's really more of a methodical approach. Get pen and paper out, write down a plan. What's the business case? What are you trying to solve? How much is it gonna cost? Do a benefits analysis, do a risk analysis."

The alternative has produced visible failures. "We've seen a lot of companies who have jumped in the last year or two. Companies firing staff and then six months later going, oops, maybe we should have hired everybody back. You didn't have a plan."

He sees CISOs increasingly leading this conversation with boards and executive teams, framing AI as an investment decision subject to the same discipline as any capital expenditure. The organizations that do this are rolling out tools more successfully, demonstrating ROI, and feeling more confident about security.

His response to the competitive-pressure argument is direct. "We hear, but our competitors will get out ahead of us. I hate to break it to you, they're not ahead of you. They're having the exact same conversation. They're flailing just like everyone else is."

Slow adopters are not the problem

Taylor asked how Nathan approaches organizations that have been hesitant to adopt AI. His answer is that most of them have good reasons.

The slow adopters tend to be critical infrastructure operators, government agencies, healthcare providers, and manufacturers, organizations where errors have physical consequences. "Anywhere where there's life and death on the line, you get a lot more hesitation, because their risk tolerances are vastly different than a tech company."

That caution becomes the foundation of the plan. A food manufacturer's concern that an AI error could make people sick defines the baseline: implementations must be safe. From there, the organization can identify where AI can be deployed, who can be trusted with sandboxes to test it, which use cases make sense, and, equally important, which use cases are excluded entirely. "Identify those places we're just gonna exclude, because it's the manufacturing line or it's the hydro dam up in the mountains that we just can't have go wrong."

The typical starting point for these organizations is a knowledge base with an LLM front end, drawing only on known internal data sources. When they move to more complex applications, their testing requirements become stringent, which is appropriate given the stakes. "Skeptics are easy. They just move slower because they know the risks at play."

Fast adopters face a trust problem, not a technology problem

Organizations that adopted AI broadly have a different set of issues. Leadership typically reports two dozen models in use across the workforce with no central control, combined with concern over headlines about AI agents going rogue.

Nathan's assessment of those headlines is that the underlying security failures are familiar. Reviewing one widely reported incident, he noted that "from a security person's perspective, there wasn't anything novel there. These were all well-known security flaws. They misconfigured stuff, they didn't sandbox correctly, they didn't apply permissions correctly. These are things we've been talking about for 30 years."

The larger risk, in his view, is one that does not make headlines. "What AI compromises in a lot of ways is trust."

He contrasted AI with earlier shifts. When organizations moved from on-premises data centers to the cloud, the underlying concept did not change. "Cloud is just a data center on somebody else's environment. The scary part there was I just need to learn how to use the tool. But the tool does the same thing I've been doing before." Network segmentation in AWS is still network segmentation.

AI is different because its output can be wrong in ways users cannot detect. "It's a text prediction engine at its most basic level. It's not designed to give you accuracy. It's designed to give you what is statistically most likely to sound like a good answer. But if I don't know what a good answer looks like, boy, it sounds really good. It must be true."

His standard question for every organization: "How do you know?"

The business cost of unverified AI output

Nathan illustrated the scale of the problem with a study he cited on Google's AI-assisted search results, which found an error rate of roughly nine percent. Applied to Google's search volume, he calculated that this amounts to tens of millions of incorrect answers per hour. The average user has no way to identify which answers are wrong.

Inside an organization, the same dynamic produces business decisions based on bad data. Legal counsel using a public model for research, where courts have already sanctioned lawyers for citing cases that do not exist, could expose the company to losses on contracts or acquisitions. A marketing team that publishes AI-generated claims without verification could create false advertising liability.

The question he puts to leadership: "What happens when your own employees start making bad decisions because AI told them to?" And the follow-up: "Do you really want nine percent of the decisions being made by your employees to be based on bad data? Nobody in leadership says yes to that question."

The fix is cultural. "We've got to get back to that mode of verify first, then trust." That means teaching employees differently, putting validation mechanisms in place, and enforcing the practice of checking AI output before acting on it, especially where it affects customers, legal matters, or personnel. He noted that fast-moving organizations are already pulling back from unrestricted use.

He also expects the tools to improve. The major AI vendors are under enough pressure that they will have to add transparency into how their models reach conclusions and demonstrate data integrity. "Six months from now, we might be able to trust these tools a lot more."

AI is an application, so the fundamentals apply

Asked which fundamentals matter most now, Nathan reframed AI in familiar terms. "If you think about what AI, as we use it today, what it is structurally, it's an application. There's a front-end interface of some type, there is some calculations and math and algorithms in the middle, and it's all accessing data."

That means the required controls are the ones security programs should already have: data security, identity and access control, and application security. The problem is that most organizations have not executed well on them, particularly data security.

"Data security, as an example, is hard. There's a lot of business decisions that have to be made. What data is critical? Who owns that data? Who decides who gets access to the data? These can be really heavy processes. They're time consuming. And because of that, frankly, they kind of got moved to the back burner."

Patching and penetration testing are easier to execute and more satisfying, so they got the attention. AI removes the option to defer. "Now we have this application that isn't contained. It's not something you can force to live on-prem necessarily. If you're gonna go hard in the AI world, data security is super important. Identity practices are really important. And if you haven't been doing a good job of that, you have to now."

Security is a risk management function

Nathan's broader argument is that the industry has not fully absorbed a change in its own identity.

Cybersecurity as a practice is only 35 to 40 years old, and its first generation came out of IT. "We were the folks who worked on networks and databases and were your help desk admins. There is still a mindset that security as a practice is a technology function. And I've been saying for a long time, that's done. It's over. Security is a risk management function. We have far more in common now with finance and legal and HR than we actually do with IT and Dev and DevOps."

Technical depth remains necessary, because the data and systems security protects are technical. But the function is risk, and that brings human questions into scope: how to allow access to an application that cannot be fully trusted, how to educate users to operate it responsibly, and how to ensure boards and executives make decisions on sound data.

The most common failure he sees follows from the old mindset. "We focus on the tooling at the cost of the people and process side of it. And you can't separate those two out anymore, especially not in the AI world." Tools for data security and identity are necessary, but without addressing trust and education, "it doesn't matter which one you put into place. It will fall apart."

What a Field CISO actually does

Taylor asked Nathan to define his role, since the title is used inconsistently. His answer started with what it should not mean. "I have been in situations where someone has said, I'm a field CISO, and you go, well, where were you a CISO at? They go, well, no, I work in marketing. Well, that's not okay."

In his definition, a Field CISO is someone who has held the CISO seat, no longer runs an operational security program, and advises other CISOs from direct experience. The value is peer credibility and perspective: "I can talk to a CISO and say, listen, I've talked to 35 of your peers in the last three months. Here's what they're talking about."

The role sits adjacent to sales but is not a sales role. "I don't carry any quota. I don't have an agenda when it comes to buying. If you buy the thing, it doesn't affect me at all." That independence is what allows a Field CISO to tell a client their strategy is sound, or to offer candid advice about a tool based on having used it.

A second function is internal: reviewing marketing and sales messaging before it reaches security buyers. "You might put something out there where I go, no, no, no, you can't say this. Any security person who's been doing this a while is going to eat us alive."

The CISO as organizational translator

Nathan connected the Field CISO's bridging function to what the in-house CISO role has become. His example: HR asks whether a new HR platform is secure enough for employee data. Answering that requires IT's input, but HR and IT do not typically communicate directly and use different vocabularies. The CISO translates between them and builds a solution that addresses the actual business concern.

"Being that natural bridge in organizations, that's a huge part of the modern CISO role today. And the ones who get good at that are the ones who I see being very successful."

He described the shift as "CISO version 2.0." Ten or fifteen years ago, the most technical security person was promoted into the role regardless of communication ability. "The role's changed. It really legitimately has changed a lot, even just in the last five years. It's all soft skills. It's all communications and people management and understanding organizational culture."

The cause, in his analysis, is that technology is now inseparable from business operations. In the early days of email, executives did not need to understand how it worked. Today, a business cannot sell without it, and leaders know it. "They know that the technology is just fully intertwined with their business decisions." Once that happened, security could no longer be kept at a distance, and security leaders had to become fluent in business language.

The regulatory environment is expanding and fragmenting

Nathan's short assessment of the compliance picture: regulation is increasing but distributing. Individual US states are adopting their own requirements, and countries across Southeast Asia and other growing technology markets are building their own frameworks. "Everything we do now is global. It just is." He estimated roughly 60 distinct privacy regulations now exist worldwide, each with its own reporting and notification requirements.

The mitigating factor is that most stay close to established best practice. "If you are following the NIST Cybersecurity Framework or ISO 27001, chances are you're probably pretty close to a lot of what these folks are doing, outside of reporting requirements."

How CISO buying has changed

For the founders and go-to-market professionals in the audience, Nathan described a buying market shaped by company size and by uncertainty around AI.

Smaller organizations are increasingly priced out of the major vendors. "I've only got 500 people or 300 people. I can't afford a quarter of a million dollar software contract." Their response is to build in-house, which AI makes more feasible, or to partner with startups as design partners. In parallel, a category of service providers targeting small and mid-size businesses is growing, offering penetration testing, software assessment, and vulnerability assessment as services.

Larger organizations retain a preference for established vendors. "Nobody ever got fired for buying IBM. I think we're still a little bit there." But AI security is the exception. Companies that normally sign three-year contracts are limiting AI security tools to one-year terms, because they expect the tools to be obsolete or absorbed by major vendors within that window.

His assessment of the vendor landscape was candid. "Everyone says they're solving it. Every customer I talk to knows that's not true. There are a lot of folks that are just smoke and mirrors." The result is a pause: organizations conduct due diligence and may buy something today, but they are hedging. "I don't know if I've ever really seen a buying market quite like this, because there's so many question marks and so much hype and so many unknowns."

The fundamentals are the plan

Nathan's conclusions converge on one point. AI adoption succeeds when it is treated as a business decision with a documented plan, a defined use case, and a risk analysis. AI security succeeds when organizations execute the fundamentals they have deferred, particularly data security and identity, and build a culture of verifying AI output before acting on it. Both depend on security leaders who operate as risk managers and translators.

The full conversation is worth your time. You can reach Nathan on LinkedIn or at [email protected].

Here are the key takeaways from this blog:

• Plan before adopting AI: the organizations succeeding with AI treat it as an investment with a business case, cost, benefits analysis, and risk analysis, and their competitors are not ahead of them.

• Slow adopters have good reasons: life-and-death risk tolerance defines the baseline, and the plan starts by identifying where AI will never be used.

• AI compromises trust, not just technology: unlike cloud, AI produces output that users cannot easily verify, so the organizational standard must be verify first, then trust.

• AI is an application: data security, identity and access control, and application security are the required fundamentals, and deferred data security work is now mandatory.

• Security is risk management: the function has more in common with finance, legal, and HR than with IT, and focusing on tooling at the expense of people and process fails.

• The modern CISO is a translator: the role now depends on communication and organizational fluency, because technology is inseparable from business operations.

• Buyers are hedging on AI security: small companies are building or partnering, large companies are signing one-year terms, and most buyers doubt vendor claims.

The Cybersecurity Ecosystem Show connects practitioners, investors, vendors, regulators, and everyone in between. New episodes drop weekly. Subscribe so you don't miss one.