Shadow AI: how to see which AI tools your teams actually use

Enterprise AI adoption is outpacing governance faster every year. Some organizations now have employees using AI tools that IT never approved, procurement never purchased, and compliance never vetted. The cost is significant: higher breach risk, redundant spending, and no evidence that AI investments deliver value. Many organizations respond by blocking AI tools entirely. That does not solve the problem. What enterprises need is real-time visibility into which AI tools people actually use, and a way to measure whether those tools work.

AI tools arrived faster than governance could follow

The gap between AI deployment and AI governance opened fast. Microsoft’s 2024 Work Trend Index, surveying 31,000 knowledge workers across 31 markets, found that 75% already use AI at work. Use of generative AI nearly doubled during first six months, and 78% of AI users bring their own tools rather than relying on what their employer provides.

That last number defines the problem. When four out of five AI users choose their own tools, the enterprise does not have visibility into a significant part of its technology estate. Not because employees are acting recklessly, but because sanctioned alternatives either do not exist or do not meet the need.

The latest McKinsey’s State of AI report quantified that awareness gap. C-suite leaders estimated that 4% of employees use generative AI for at least 30% of their daily work. Employees self-reported the actual figure at roughly 13%, three times higher. Leadership does not know how much AI their teams use.

What shadow AI actually looks like

Shadow AI is not a single tool or a single behavior, it is a pattern. An analyst pastes customer data into ChatGPT to summarize a quarterly review. A nurse enters clinical notes into a free-tier AI tool to draft discharge instructions. A procurement team adopts a department-level AI contract without informing IT.

The tools include general-purpose assistants (ChatGPT, Claude, Gemini, Microsoft Copilot) and domain-specific AI services (clinical documentation tools like Dragon Copilot and DAX Copilot, coding assistants, design generators). Many run entirely in the browser. Many are free or low-cost. Most leave no procurement trail.

Netskope’s 2025 Cloud and Threat Report, based on telemetry from over 3,500 enterprises, found that 72% of enterprise generative AI use runs through personal accounts rather than corporate-managed ones. That is not a policy problem. It is a visibility problem.

Why traditional IT asset management misses it

Shadow AI does not look like traditional shadow IT. There is no unauthorized hardware. There is no rogue software install. There is no procurement request to intercept.

As Gartner noted in a November 2025 analysis, shadow AI “typically does not require more than visiting a website with a browser.” That single observation explains why conventional controls fail. Network-level monitoring sees traffic categories, not individual AI tool usage. Software asset management inventories installed applications, not browser tabs. CASBs were designed for SaaS platforms with corporate accounts, not free-tier AI tools accessed through personal logins.

The Zylo 2026 SaaS Management Index, analyzing over 40 million licenses, found that IT leaders have visibility into only about 13.5% of the applications their organizations actually use. AI tools expand that blind spot further because they add a category that existing inventories were never designed to track.

The compliance and cost dimensions

The risk is not hypothetical. Samsung’s 2023 experience became the canonical example: semiconductor engineers pasted proprietary source code and test sequences into ChatGPT within weeks of the tool being permitted internally. Samsung banned generative AI on company devices. JPMorgan Chase, Goldman Sachs, Apple, and others followed with their own restrictions.

IBM’s 2025 Cost of a Data Breach Report, covering 600 organizations globally, found that shadow AI was involved in 20% of breaches and added $670,000 to the average breach cost. Among organizations that experienced an AI-related breach, 97% lacked proper AI access controls. The same report found that only 34% of organizations with AI governance policies actually audit for unsanctioned AI use.

For regulated industries, the exposure compounds. Healthcare organizations handling protected health information face HIPAA obligations regardless of which tool processes that data. Financial services firms operate under data residency and audit trail requirements that personal AI accounts cannot satisfy. Manufacturing and pharma organizations subject to GxP or SOX compliance need documented evidence of controlled processes, which disappears when employees use unmanaged tools.

The cost dimension extends beyond risk. When multiple departments independently purchase AI tools that overlap in function, the organization pays for redundancy it cannot see. Without visibility, there is no way to consolidate, negotiate, or measure return.

What visibility actually requires

Blocking does not work as a long-term strategy. Gartner’s own research on AI agent sprawl explicitly states that employees route around restrictions. Organizations that banned ChatGPT in 2023 found usage moving to personal devices and alternative tools.

The NIST AI Risk Management Framework offers a more durable model. Its four functions, govern, map, measure, and manage, assume that AI tools exist across the organization and that the goal is to understand them, not eliminate them. Mapping means maintaining a current inventory of AI systems in use. Measuring means tracking risk and value over time. Managing means prioritizing and treating risk based on evidence.

What this translates to in practice is continuous, browser-level visibility into which AI tools people use, through which accounts, and how often. Not a one-time audit or a survey. Real-time telemetry that updates as new tools appear and usage patterns change.

The market is responding. Gartner projects that spending on AI governance platforms will reach $492 million in 2026 and exceed $1 billion by 2030. Organizations that deploy AI governance tooling are 3.4 times more likely to achieve high effectiveness in their AI governance programs.

But governance that only detects presence is not enough. Detection tells you what exists. Measurement tells you what works. The organizations that close that gap will not just reduce risk. They will know which AI investments to scale, which to retire, and where the next dollar should go.