Arsenii Palivoda // Shutterstock
Workers are quietly using unapproved AI tools at work. Security teams can’t keep up.
If you’ve signed up for ChatGPT, Claude, or another AI tool on a work laptop without looping in your IT department, you’re not alone. Across industries, employees are under pressure to work faster with AI, and many are finding and installing tools on their own to keep up. New data suggests this is happening at a scale most companies aren’t equipped to manage.
Vanta, a security compliance company, examined anonymized data from more than 15,000 businesses using its Third-Party Risk Management platform to see how workplace AI use is reshaping the software running inside companies. The findings, the first in a new data series called Trust Signals, point to a rapid rise in what researchers call “shadow AI“: AI tools employees adopt on their own, without their company’s IT or security team knowing they exist.
Shadow AI has outgrown shadow IT
“Shadow IT” is a decades-old term for any software running inside a company that was never approved through official channels — it’s always been a headache for security teams, and AI is making it worse. According to Vanta’s data, 70% of companies now have shadow AI somewhere in their environment: AI tools with access to company data that were never vetted or approved.
Overall, shadow IT grew 36% year over year, and companies discovered an average of around 140 unapproved tools accessing their systems within 90 days of connecting to Vanta’s platform. Employees under pressure to move fast, the data suggests, often can’t or won’t wait for procurement to catch up. When approval is too slow, many skip it.
Banned tools keep coming back
Blocking a tool doesn’t appear to stop much. Vanta’s analysis found that for the average customer, employees reinstall revoked tools more than 100 times within a 30-day period, and roughly 1,000 times over a year. These aren’t different tools cycling through — it’s largely the same ones, removed and reinstalled again and again, a pattern researchers describe as a kind of security whac-a-mole.
How often this happens depends on what controls a company has in place, but the pattern holds: If there’s a way back in, employees tend to find it. That suggests shadow AI isn’t a problem security teams are failing to spot but a demand problem outpacing whatever controls already exist.
The tools workers fight hardest to keep
The tools that show up most in this remove-and-reinstall cycle aren’t obscure. Anthropic, OpenAI, and Cursor — some of the most talked-about names in AI — top the list, according to Vanta’s data. That tracks: When someone’s daily workflow depends on a specific AI tool, losing access creates an immediate gap, and they tend to close it themselves.
The stakes are also higher than with an ordinary, unapproved app. AI tools typically have deep access to company data, which is part of why Vanta’s data shows LLM vendors are 52% more likely to be flagged as “high risk” than traditional software vendors. The most common reasons cited: These tools touch sensitive company data, connect to source code, and are treated as essential to how the business runs. Without clear rules for how AI tools get vetted, that mix of broad access, high stakes, and low oversight raises the risk of a data leak or unauthorized access.
Almost none of these tools get reviewed
Vanta’s data shows that only 2% of shadow IT vendors ever go through a security review. The other 98% remain essentially unvetted, even as employees keep using them daily. More than half — 55% — of the average organization’s entire vendor footprint now qualifies as shadow IT, and that share is climbing as AI adoption pressure builds.
What’s replacing the ban-it-all approach
Security teams have tried the traditional playbook — block the tool, revoke access — and it isn’t holding up against how much employees want to use these apps. There’s precedent for this kind of standoff: iPhones, Dropbox, and more recently, the AI note-taking tool Granola all spread inside companies well ahead of IT approval. In each case, the tools that prevailed were the ones eventually brought inside the security perimeter, not the ones that stayed blocked.
That history is shaping how some companies respond now. Instead of only blocking harder, they’re building faster review processes that still weigh risk seriously — checking a tool as soon as it shows up, rather than trying to keep it out indefinitely. Some use third-party risk management tools for this, which connect to a company’s identity provider to spot which vendors employees are actually using and score them against a risk framework automatically, before a person has to step in.
The bet is that speed, not just strictness, is what keeps security oversight relevant. A tool banned on a Friday has a good chance of being back in use by Monday, so treating review speed itself as a security capability — rather than relying on blocking alone — looks like the more realistic path forward.
About the data
The findings come from anonymized usage data across Vanta’s Third-Party Risk Management customer base — thousands of businesses — collected from February 2024 through April 2026. Vanta compared discovered vendor counts, security review rates, and risk levels across industries and company sizes by connecting to customers’ identity providers (the software that manages employee logins). Year-over-year comparisons used point-in-time snapshots from January 2025 and January 2026 within the same group of customers.
This story was produced by Vanta and reviewed and distributed by Stacker.
