When enterprises first shifted to the cloud, speed was everything. Developers spun up Amazon Web Services (AWS) instances overnight to launch new apps, test new features, and accelerate digital transformation. I had a front-row seat to this while leading Splunk’s transformation to a cloud-first company, where our cloud offerings, the heart of our growth during my tenure, were tied closely to AWS.
Having teams move rapidly to the cloud was the right move. Enterprises couldn’t afford to wait. But those urgent deployments often bypassed governance and security teams entirely. CIOs and CISOs woke up to discover their organizations running dozens of cloud environments outside their oversight. That was the birth of shadow IT. The move to cloud fueled progress, but it also created extraordinary risk.
Today, we are watching history repeat itself. Only this time, it is happening faster, more invisibly, and with far higher stakes. Welcome to shadow AI.
WHAT IS SHADOW AI?
Shadow AI refers to the unsanctioned use of artificial intelligence tools across the enterprise, often invisible to traditional oversight. Business leaders want productivity gains, and employees adopt AI without waiting for IT approval or security vetting.
It might look like a developer using GitHub Copilot to generate code, a marketer relying on ChatGPT to draft campaigns, or a sales rep uploading sensitive data into an AI assistant for analysis. On the surface, it is all speed. Underneath, it creates new layers of risk: internal data flowing into external models, ephemeral AI agents creating their own access patterns, and systems interacting in ways governance teams never anticipated.
WHY IT’S DIFFERENT THIS TIME
Shadow IT was disruptive, but at least the underlying technologies—cloud workloads and SaaS applications—were largely static and deterministic, behaving in relatively predictable ways. Shadow AI is different.
Generative AI is nondeterministic. Identical prompts generate different results. That is a powerful feature of foundation models, but radically different from how prior systems behaved. AI agents also self-direct. They adapt. They operate over encrypted channels that look like normal traffic, making them nearly invisible to firewalls and identity tools. They can even generate new credentials and carve their own pathways through internal systems.
The same lateral “highways” that make AI efficient also expose organizations to attackers. It is no coincidence that 70% of successful breaches involve lateral movement, the same paths AI uses inside the cloud. Without new approaches, the tools driving your productivity could just as easily open doors to systemic compromise.
YOU CAN GUIDE SHADOW AI
Just as with shadow IT in the clouds, shadow AI is not going away. Employees and teams will adopt AI tools because the competitive pressure is too great. The question is not whether you can stop it—you cannot—but whether you can shape the environment so shadow AI works for you, not against you.
That requires a mindset shift. Traditional defenses such as perimeter firewalls, endpoint protection, and access controls do not extend into the places where shadow AI lives. The real solution lies deeper, at the infrastructure layer.
Enterprises need to embed zero trust for workloads and establish fabric-level visibility across their cloud networks. That means:
- Governing how AI agents move laterally inside your environment
- Monitoring encrypted traffic patterns for anomalies
- Enforcing trust dynamically at the network layer, not just at the edge or identity layer
With these controls, enterprises can allow AI to flourish, boosting productivity and creativity, while still retaining the visibility and remediation needed to manage risk.
THE LEADERSHIP LESSON
Shadow AI is not a glitch to be stamped out. It is a natural expression of how people and systems evolve. When people see an opportunity to move faster, they take it. That instinct fuels innovation. Our universe itself is emergent, expanding at an accelerating rate. That expansion is beautiful, but it also creates entropy. Left unmanaged, it produces disorder. Guided with the right guardrails, it produces resilience and growth.
At Aviatrix, I see this paradox every day. Customers are racing to modernize security in the cloud, embracing speed and experimentation while trying to protect critical systems. The ones who succeed are not those who try to contain change, but those who put the right lightweight guardrails in place so their people can innovate safely.
The same is true with shadow AI. Governance, done properly, is not an obstacle to innovation. It is the framework that allows innovation to accelerate. Leaders must recognize that insecurity is not failure. It is the natural condition of progress. By shaping environments where fragility is visible and actively managed, we can allow the creativity of AI, and of our people, to expand safely.
Doug Merrit is chairman, CEO, and president of Aviatrix.