Right now, somewhere in your organization, someone is sharing confidential data with AI for analysis. A developer is pasting source code into an AI coding assistant to help them be more productive or fix a problem. A marketing team is building an AI-powered content-generation pipeline, blissfully unaware that their colleagues elsewhere did the same last quarter.
In fact, almost half (46%) of U.S. employees now say they use AI, according to Gallup.
Given the potential increases in productivity, shouldn’t all of this AI usage be considered a good thing? Not always, especially if a business is unaware of how AI is being used and lacks the governance to manage it effectively.
And, unfortunately, this seems to be a pretty common situation: Gartner found that 69% of cybersecurity leaders suspected employees were using unapproved AI tools. Even if only half of those leaders are correct in their suspicions, this is definitely not good.
This is shadow AI, and it keeps coming up in my conversations with security, compliance and technology leaders as an emerging and significant business risk.
Shadow AI mirrors the shadow IT problem—such as unsanctioned use of cloud storage or software-as-a-service (SaaS), or the business use of personal devices—that we all faced over the last decade, but it can be harder to identify and may have more serious consequences.
The AI data leakage problem
When it comes to the risks of shadow AI, leaking confidential data and intellectual property is front of mind for most organizations. This occurs as employees share data with AI tools with the best of intentions, potentially using their own personal accounts, outside of data governance.
Depending on the configuration of the AI tools, these inputs can be incorporated into the corpus of information used for the next training cycle, so confidential data and/or intellectual property could be shared inadvertently with other users of the same tool. In fact, Gartner predicts over 40% of enterprises will have security or compliance incidents linked to unauthorized shadow AI by 2030.
The opportunity cost of duplication
Beyond data leakage, the second risk that comes to the fore is around organizational efficiency. Most employees use AI to accomplish tasks more quickly, but without central coordination, teams can end up with incompatible solutions that follow separate paths. This pattern mirrors earlier enterprise technology cycles around cloud adoption.
This fragmented approach can result in the same costs being incurred and, in some cases, the same work being undertaken, without sharing best practices or pooling capabilities. One department may spend weeks solving a problem that another business unit solved last quarter. And when these siloed projects are eventually discovered and merged, the organization has already borne the cost multiple times.
The Nutanix Enterprise Cloud Index, published in early 2026 and drawing on 1,600 IT and engineering executives globally, found that 79% of respondents had encountered AI technologies being implemented by staff outside IT functions.
5 ways to avoid the risks and reap the rewards
Simply attempting to ban the use of unsanctioned AI tools is understandable but can be counterproductive. The goal of (most) teams or individuals using AI is to improve business productivity. Businesses need to combine policy and education to enable the adoption of new technologies that maximize business benefits, while controlling risk:
1. Create clear AI usage policies that define approved tools, establish data-handling guidelines, and classify AI applications into tiers: sanctioned, limited-use, and prohibited.
2. Provide a streamlined approval pathway for new AI technologies, as complex, long-winded governance processes risk being bypassed.
3. Create an internal AI council to maintain oversight of the above and ensure accountability across the organization.
4. Keep up with new regulations shaping governance, e.g., the NIST AI Risk Management Framework, which identifies and manages AI-related risks and is being adopted across North America.
5. Consider security tools that sit between users and AI tools, both to identify usage and implement policy controls.
Looking at number five more specifically, AI Gateways apply policies in real time to, for example, redact personally identifiable information, log AI interactions, and block queries that violate data classification rules. Also, some pre-existing technologies, such as cloud access security brokers (CASBs), have added new features to detect and inventory AI usage.
Gartner predicts that more than half of enterprises will deploy some form of AI security platform by 2028. However, given the ever-accelerating pace of AI adoption, that timeframe may prove too conservative.
Bring AI into the light
AI is evolving more rapidly than we ever imagined, and many are keen to harness its potential faster than governance can adapt. This is the same dynamic that drove shadow IT a decade ago.
Organizations that manage the shadow AI challenge will provide clear policies, controls and, most importantly, pathways to adopt new capabilities quickly and safely. They will make responsible AI adoption easy for everyone—reaping the business benefits, while controlling risk.
AI technologies are changing the way we all work. The key to success is to steer these changes rather than let them happen in the shadows.
Darren Antee is the chief technology officer for security at NETSCOUT.