Having Security Controls Is Not The Same As Being Secure

Francis Dinha is CEO and cofounder of OpenVPN Inc., a leading enterprise network security company.

getty

​A finding in Proofpoint’s 2026 AI and Human Risk Landscape report reveals something alarming: half of the organizations that already had AI security controls in place still experienced a confirmed or suspected AI-related incident. Despite controls being present, the incidents still happened for half of everyone using them! That means that when these organizations implemented what they believed were protections against AI security incidents, they may as well have flipped a coin to determine if these protections would work. ​

That disparity is not a technology failure, as tempting as it may be to mark it as such. Rather, I believe it is a measurement failure, which reflects a structural confusion running considerably deeper than AI security alone.​

The Distance Between Policy And Enforcement​

To be fair, there is an understandable logic behind the procurement model of security. Every tool addresses a documented threat category, every policy maps to a compliance requirement and every framework produces a report that leadership can point to as evidence of diligence. In this way, most security programs are structured to optimize for auditability rather than resilience, a distinction that rarely surfaces in procurement conversations but becomes painfully apparent after an incident. ​

Shadow AI is an example of how the difference between policy and practice can cause serious damage to organizations. In May 2026, Community Bank reported to the SEC that an unauthorized AI-based software application disclosed customer names, social security numbers and dates of birth.

Research from Purple Book Community and ArmorCode found that nearly 90% of enterprises claim full visibility into their AI footprint, while 59% acknowledge that shadow AI exists within their environments. Both cannot be true simultaneously, and the difference reflects what enterprises are consistently missing.​

What we’re seeing here is a problem with enforcement. It is not a new phenomenon, but AI has made it considerably more widespread. ​

Why Fragmentation Compounds The Problem​

Part of what sustains this disparity is the sheer fragmentation of the modern security stack. The same Proofpoint report found that 94% of organizations describe tool management as challenging, a figure that will surprise no one who has observed how enterprise security programs tend to evolve in practice. Tools tend to accumulate over years, and each one addresses a specific threat surface. They are rarely designed to share signals with adjacent systems. The result is an environment where individual components may function exactly as their vendors intended, while the composite picture across the organization remains unclear.​

Research from AlgoSec reinforces this, with 71% of security teams reporting struggling with visibility in their networks in 2025. All too often, organizations deploy separate monitoring tools across cloud environments, on-premises infrastructure and endpoints without considering how they all connect. Each one might enforce policy in isolation, but attackers can often move through the seams between those siloed systems.​

This fragmentation problem is sharpened further by the pace at which AI is being introduced into production. Proofpoint’s data shows that 87% of organizations have moved AI assistants beyond the pilot stage, with three-quarters actively rolling out autonomous agents. Yet more than half describe their security posture as catching up, inconsistent or reactive. When deployment accelerates on one timeline and governance matures on an entirely different one, the distance between the two is precisely where we find dangerous exposure.​

What Meaningful Protection Actually Requires​

Closing this distance is a different problem than building a security program from scratch, because the tools, in most cases, already exist. What is missing is the integration and visibility necessary to know whether they are doing their jobs.​

This begins with a shift in the standard of evidence.

Attestation, the practice of confirming that a control has been deployed, is categorically different from validation, which confirms that a control is functioning as intended under real conditions. Programs that stop at attestation are at risk of producing the scenario Proofpoint documented. Their protections will be in place, but incidents will still occur, and they likely won’t understand why.​

Protection must be continuous rather than periodic.

Point-in-time assessments capture a snapshot of posture at the moment of measurement, but in environments where AI systems generate activity around the clock across dozens of integrated channels, a snapshot is an unreliable measurement. Meaningful oversight requires inspection to be ongoing, with traffic consistently evaluated and access continually being validated in real time rather than against the memory of a quarterly review cycle.​

Visibility must also be unified.

Disconnected tools produce disconnected signals, and disconnected signals produce the kind of blind spots that attackers have learned to navigate all too well. A coherent security architecture requires a layer that aggregates those signals and makes them actionable, operating not as a reporting interface but as a live enforcement layer positioned between what systems are authorized to do and what they are observed doing.​

Posture Is Earned, Not Declared​

The uncomfortable reality is that many organizations cannot answer a basic question: Which of their security controls demonstrably reduce risk today? They can show purchase orders, deployment records, compliance reports and policy documents. What they often cannot show is evidence that those controls are still functioning effectively against current threats. ​

Security posture is not established at the moment a tool is licensed or a policy document is approved. It must be earned continuously, through enforcement that operates in production, at the pace of the environment it is meant to protect and across every channel where enterprise activity actually flows.

Organizations that close the distance between what their protections claim and what their environments reveal will find themselves defended at a much higher level. Those that do not will unfortunately learn the hard way the consequences of letting this incongruence slide.​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?