Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

The invisible risk that could undermine your AI strategy

What’s needed for cloud security in 2026.

[Photo: Getty Images]

Profile photo of Douglas Merritt

CEO · Aviatrix

2024 delivered a 33% surge in reported cybercrime losses (equaling $16.6 billion), a 44% jump in weekly attacks on corporate networks, and a doubling of third-party breaches (30% of all incidents). We’re waiting for the final numbers, but 2025 is tracking worse.

While adversaries have grown smarter, more organized, better funded, and AI-enabled, that isn’t the primary cause of the continuing onslaught. The primary issue is the overwhelming complexity of most corporate cloud environments.

We raced to the cloud and are now racing again to layer AI into our systems. Meanwhile, cloud workload security—the security of the applications, services, and AI agents actually running your business—has lagged.

You don’t need to speak Kubernetes or serverless to see the implication: If your cloud architecture is chaotic and disjointed, your AI strategy is fragile. In 2026, that fragility will show up as lost trust, regulatory exposure, and interruptions to your biggest bets.

Here’s how I see the year ahead, and what leaders should do now.

1. Four generations of tech, one expanded attack surface

Most companies are running four eras of technology at once: lift-and-shift systems you can’t rewrite, modern containerized applications, serverless services that appear and disappear on demand, and AI agents that need to talk to all of it.

The risk isn’t in any one layer—it’s in the connections. An AI agent helping a customer might query a legacy database, call a modern billing API, and trigger a serverless refund, all in seconds. If that agent’s credentials are stolen or its instructions manipulated, it moves laterally across systems faster than any human attacker, exploiting connections that no one fully mapped.

Prediction: 2026 will set another record for breaches tied to lateral movement across generations of technology. The warning signs are already here. In 2024, edge devices and VPNs—the very systems meant to protect networks—accounted for 60% of enterprise zero-day exploitation.

2. Zero trust will separate slogans from real strategy

Zero trust is simple: Never automatically trust a user, application, or system just because it’s “inside.” Always verify. (Important tip: In cloud, there is no “inside.” Nearly every workload is exposed to the “outside.”)

The challenge is consistency. If older systems are governed one way, newer apps another, and AI agents sit on top calling into both worlds, you don’t have zero trust, you have fragmented trust.

Here’s a CEO-friendly test: “Only this application and this AI agent can access that customer database.” Can you define that rule once, in plain English, and know it’s enforced everywhere?

The organizations that pull ahead in 2026 will build a single control plane above the complexity, defining policies in the business’s language and enforcing them consistently, whether the workload is a virtual machine, a container, a serverless function, or an AI agent. That’s not housekeeping. It’s resilience.

3. Network security returns to the board agenda

Identity remains essential. But a dangerous myth has spread: If we get identity right, we’re safe.

In the cloud, there are two control points you can always rely on: identity (who or what is acting) and network (what is talking to what). Most breaches follow a familiar arc: compromise credentials, move quietly inside the environment, then exfiltrate data or disrupt operations. We’ve invested heavily in stopping the first step and underinvested in blocking the second.

That’s why network security will move back to center stage in 2026, anchored by one question: For our most critical applications, do we know what traffic is leaving our environment, from which workloads, and why?

4. The most important decision is architectural, not technological

Board conversations still sound like shopping lists: Do we have the right tools? Are we covered by the right acronyms?

Tools matter, but they only see slices. Attackers see the whole system. The shift in 2026 is that leaders will stop asking “Are we covered?” and start asking “What is our cloud security architecture?”

Two practical checks:

  • Can we define security intent once and enforce it everywhere—across clouds, regions, and all four generations of technology?
  • Can we prove in production that policies like “No regulated-data system talks directly to the internet” are actually true?

If the answer is yes, you can move faster with AI because you trust the foundation. If it’s no, every new AI initiative quietly multiplies risk.

WHAT CEOS SHOULD DO NOW

Change the questions you ask. Start with three:

  • Do we have a real-time inventory of every cloud workload and AI agent?
  • Do we know every identity—human and non-human—touching critical systems?
  • Can we see and control every packet flow into, out of, and between workloads?

Then set one architectural objective: Pick one mission-critical application. Map how it really works today, every system, workload, and AI component it touches. Require that its policies are defined once and enforced consistently, regardless of where components run.

If we get this right, AI in 2026 becomes an innovation accelerant. If we get it wrong, the next high-profile failures won’t be because AI was too powerful. They’ll be because we asked it to run on top of a foundation that we never took the time to secure.

Doug Merritt is CEO of Aviatrix.