In our last post, we made the case that shadow AI risk starts with the data, not the tool. Blocking unsanctioned AI apps can’t keep pace with how quickly new tools appear. What reduces risk is protecting sensitive data itself, so it stays safer wherever it ends up, whether that’s an approved AI assistant or a tool nobody signed off on.

That strategy has a prerequisite. You can’t protect data you don’t know you have, or data you don’t know is sensitive. In most organizations, a large share of data falls into exactly that gap: a customer export from a migration that wrapped up long ago, a copy of an HR spreadsheet saved to a team share “just for now,” a project folder from a deal that closed three fiscal years back.

For years, files like these were relatively harmless, largely because they were hard to find. AI changes that. Sanctioned or not, AI can search, summarize, and surface content in seconds. The forgotten files that stayed quiet because nobody went looking for them are starting to wake up.

What is shadow data?

Shadow data is data your organization holds but doesn’t know enough about. It usually isn’t hidden or completely unknown. More often, it sits outside your main, well-managed repositories or in places that are rarely used, so nobody has a clear view of what it contains, how sensitive it is, or who can reach it.

It could include copies of production data in test environments, exports saved to offline or seldom-accessed file shares, data in SaaS aaps adopted without approval, backups and snapshots that outlived their purpose, and records left behind in legacy applications.

Shadow data rarely comes from bad intent. Like dust or technical debt, it builds up through everyday work. Someone needs a copy to troubleshoot an issue, analyze a trend, or share with a partner, and the copy outlives the task.

Much of it overlaps with ROT data: redundant, obsolete, or trivial information that no longer has business value but still carries risk. A duplicate customer list or the database of a retired application may be useless to the business. If it holds personal or regulated information, it can still be very useful to someone who shouldn’t have it.

The evidence: shadow data and shadow AI show up in breaches

This isn’t a theoretical problem. IBM’s 2024 Cost of a Data Breach Report found that more than one-third of breaches involved shadow data, which the report defines as data stored in unmanaged sources. The 2026 edition found that incidents involving shadow AI more than doubled compared with the previous year.

The pattern shows up inside individual organizations too. When Queensland Museum ran automated discovery across its repositories, it surfaced sensitive and long-forgotten files spread across millions of records, along with opportunities to cut storage costs.

Taken together, these findings point to a compounding risk. The data nobody is tracking is meeting the AI use nobody approved.

Finding files like these means looking well beyond the systems you actively manage. The OpenText™ Voltage™ Data Security Platform product overview shows how automated scanning across file shares, email, and cloud repositories brings forgotten sensitive data to light, flagging ROT data along the way.

How AI wakes up zombie files

Forgotten files have always carried some risk. What’s new is how quickly they can be found and moved, and that applies to every kind of AI in your organization, not just shadow AI:

Finding old content takes seconds. AI assistants built into approved platforms can draw on any content a user has permission to reach. A file buried six folders deep with overly broad access can appear in an AI-generated answer, even if no one has opened it in years.
Old data finds new exits. When someone comes across a useful legacy export, uploading it to an unsanctioned AI tool for quick analysis is easy. That’s how AI data leakage often happens: not through a single bad decision, but through ordinary work with data that should have been retired.
Context gets lost. The person who created a file usually knows what’s sensitive in it. Years later, the person who finds it may not, and neither does the AI tool processing it.

The result is a widening gap. AI adoption is moving faster than most organizations’ understanding of their own data, and every file you don’t know about is a file you can’t govern or protect.

Why you can’t discover shadow AI risk without discovering shadow data

Many organizations start by trying to discover shadow AI: building an inventory of the AI tools and features people use. That visibility is valuable. But knowing a tool is in use tells you only half the story. To understand your exposure, you also need to know what sensitive data that tool could reach or receive.

Shadow data discovery answers the other half. When you know where sensitive data lives, including the copies nobody remembers making, you can see which forgotten files pose the greatest risk and act on them before an AI assistant or an unsanctioned tool does.

What this looks like in practice

You don’t need to review every file by hand. Automated discovery across file shares, databases, cloud storage, and SaaS repositories can locate shadow data at scale and flag ROT data along the way. Tools such as OpenText™ Voltage™ Data Security Platform automate this discovery, so your team can spend its time deciding what to keep rather than searching for it.

From there, some of the most effective risk reduction is also the simplest: an AI tool can’t surface a file that no longer exists. Defensibly deleting or archiving ROT data under policy shrinks what AI can reach. The OpenText™ Voltage™ Structured Data Manager product overview explains how organizations retire inactive data from application databases. Clear retention policies then help stop new copies from piling up.

Know your data before AI does

Zombie files aren’t a sign that your team has fallen short. They’re a natural byproduct of years of people getting work done. What has changed is how easily AI can find them and how quickly they can travel.

The organizations that adopt AI with confidence will be the ones that know what’s in their repositories before AI does. Shadow data discovery, paired with steady reduction of ROT data, turns forgotten files back into known, governed information and leaves far less for shadow AI to wake up.

See how Voltage Data Security Platform helps you find shadow data and ROT data across your environment, so you can put zombie files to rest before AI finds them.

Coming up next in this series: once you know where your sensitive data is, how do you keep it safe wherever it goes, including into AI tools you didn’t approve? We’ll look at data-centric security for shadow AI.