Why Culture Is the First Line of Defense in the Age of Agentic AI

The arrival of agentic AI rewrites the rules of engagement for cybersecurity. As new tools and workflows create novel attack surfaces, the velocity and sophistication of AI-driven threats now demand a response that transcends technology alone. This new reality calls for a profound shift in our thinking toward a security-conscious culture, one where trust and empowerment form our first line of defense.

Every part of a business must embrace security as its own critical responsibility. This means ensuring our employees are well-equipped and empowered to make sound, secure decisions. It means fostering an environment where people feel comfortable speaking up when they spot something that doesn’t seem right. And, critically, it means ensuring every leader across the business knows how to communicate and collaborate effectively if the worst happens and a breach occurs.

The new battlefield: Agentic AI and our widening vulnerabilities

In my years specializing in computer crime investigations, including my time as a Special Agent with the Air Force Office of Special Investigations, I’ve seen firsthand how the frontlines of the cyber conflict shift. Today, it’s clear that networks worldwide are the primary arena for those who wish to do harm — whether it’s nation-states aiming to steal vital secrets or disrupt our critical infrastructure, or cybercriminals looking to cripple business operations for their financial gain.

Agentic AI magnifies this challenge considerably. When we talk about agentic AI, we’re essentially describing AI that has been given its own “arms and legs” to take independent action — a powerful way to visualize it, as our CEO, Nikesh Arora, often describes. This reality propels us into what I can only describe as an “arms race.” We must continuously ask ourselves one question: Will our defenses be nimble and smart enough to keep pace with those on the offensive, or will attackers gain the upper hand? At the heart of this race is the speed with which attackers can use agentic AI to devise entirely new capabilities and coordinate their efforts with astonishing efficiency. It’s also the speed with which we, as defenders, must detect these actions and respond effectively.

We can no longer think of our defenses like a fortress with a simple, hard outer wall. The attack surface — all the ways attackers can try to get in — is now much more fluid. It encompasses our mobile devices, our cloud computing environments, and what remains of our traditional networks. We need clear visibility and the ability to identify malicious actions at every conceivable point — from one computer to another, as well as between applications and the various layers of our digital infrastructure.

The erosion of trust: AI-powered deception

One of the things that concerns me about advanced AI is how cleverly it can be used for manipulation, adding another layer of complexity to our work. Attackers are already using AI in numerous ways, particularly in crafting social engineering schemes that are more convincing than ever. Language barriers, for instance, which once might have provided subtle clues of an attack, have been virtually eliminated.

This capability now extends alarmingly to voice and video. It’s possible for attackers to take a mere 5–10-second snippet of someone’s voice and then replicate it with frightening accuracy, making it incredibly difficult to detect fraudulent calls to a help desk or other deceptions that rely on voice. The rapid advancement into deepfake video capabilities further blurs the line between what’s real and a manipulated imitation. Figuring out if you’re talking to a colleague or an AI-generated fake will get tougher and, I suspect, become a more common challenge.

This means we cannot solely rely on the ways we’ve traditionally verified identity. If an attacker’s aim is to compromise someone’s identity to access sensitive information, then it’s paramount that all the subsequent steps in our processes are even more secure. Every transaction involving our important data — how it’s accessed, changed, or moved — must have robust verification at every single stage.

Beyond technology: The enduring power of data, process, and people

With the cost of data breaches now averaging nearly $5 million^[1] for organizations, being strong on cybersecurity is, without a doubt, a real business advantage. In my experience, success in this demanding environment hinges on having access to the right information at the precise moment it’s needed to detect an attacker’s activity. Then, almost instantaneously, we must determine: Is this a legitimate action, or is it something malicious?

Organizations that do this well have great people and effective technology. They also ensure that the visibility their technology provides is centralized. This allows their systems to automate much of the initial work of detection, freeing up their skilled employees to focus on investigating the most complex and nuanced situations. Conversely, a jumble of different security tools that don’t talk to each other effectively creates inherent hurdles for our defenders — hurdles that attackers are all too quick to exploit.

One of the most pressing challenges I see organizations grappling with today is “shadow AI.”  I hear frequent questions from CIOs and CISOs: “How can I ensure we’re using AI in our organization safely? How do I even get a handle on what AI applications are being used across different departments? And, what company data might be fed into them?” The larger and more distributed the organization, the more complex this becomes. This makes a clear, centralized AI strategy — complete with approved applications and strong measures to prevent data leakage — more critical than ever. We need the ability to specify which AI applications are approved for use and ensure employees aren’t inadvertently introducing new, unsanctioned applications into our environment.

However, even with these strategies, significant challenges remain. Stopping sensitive company data from accidentally being fed into public AI tools is something we’re continuously working on. Ensuring our internal defenses can match the sophistication of AI-powered attacks is another ongoing effort. And, critically, we must address the challenge of how much we can trust the outputs of AI systems, which still often require human oversight and validation to guard against problems like “hallucinations” or simple inaccuracies.

Culture: The ultimate human firewall

When I look at the kinds of cyber dangers we’re dealing with now, they’re faster, more intricate, and happening on a bigger scale than ever before. We’re seeing nation-states borrow techniques from cybercriminal groups, and attackers exploit vulnerabilities across global supply chains within minutes of them becoming known. This situation highlights a simple truth I’ve come to learn through years on the frontlines: Technology by itself, no matter how advanced, isn’t a magic bullet.

My ultimate advice, therefore, goes beyond just technology. It’s about acquiring the latest tools and having brilliant people concentrated solely on the security team. Fundamentally, it’s about cultivating a pervasive, deeply ingrained security culture within every organization.

What does this culture look like in practice?

• Shared responsibility: From the legal department to operations, finance to HR, every single part of the business must recognize and internalize that security is their responsibility too.
• Empowerment: Our employees must be well-positioned and genuinely empowered to make secure decisions in their daily work. They need to feel it’s both safe and encouraged to raise their hand when they see something that doesn’t look right.
• Communication and preparedness: Our leaders across the business must clearly understand their roles and responsibilities. Crucially, they must know how to communicate effectively with one another and with security teams if a breach occurs. The more we practice and test our responses to various scenarios, the better prepared and more secure our organizations will inevitably be.

In this era, where agentic AI is relentlessly speeding up the pace of cyber challenges, I believe a deeply ingrained security culture — one built on a bedrock of trust, shared responsibility, and continuous vigilance — is our most resilient and adaptable line of defense. It’s about fostering an environment where every individual understands their vital role in protecting the organization. By doing so, we transform our entire workforce into an active, engaged, and ultimately formidable part of our collective security solution.

This article was adapted from Wendi’s appearance on the IBM AI in Action podcast.

---

^[1] Cost of Data Breach Report, IBM, 2024.