Shadow AI is the use of AI tools, such as chatbots, writing assistants, and browser-based AI apps, that employees adopt without IT's knowledge or approval. The fastest way to detect it is to look at what employees actually sign in to, not at what your identity system already knows about. Then decide which tools to allow, which to warn about, and which to block.
If you run IT for a small or mid-sized company, that sounds like a lot of work. It does not have to be. This guide walks through a practical process you can finish without a dedicated security team or a heavy security stack.
What is shadow AI, and why does it happen?
Shadow AI is the AI version of shadow IT. An employee finds a tool that saves them an hour, signs up with a work or personal email, and starts using it. Nobody in IT approved it, so nobody is tracking it.
It happens for ordinary reasons:
- AI tools are free or cheap to try, and sign-up takes a minute.
- Teams are under pressure to move faster, and AI helps.
- New tools appear weekly, faster than any approval process.
- Most of this work happens in the browser, outside the systems IT manages.
The risk is practical. You cannot set a policy for tools you do not know exist, and you cannot answer an auditor, a customer questionnaire, or your own leadership about which AI tools touch company work.
Step 1: Find out which AI tools employees are using
Start with discovery. You want a list of the AI tools in use, who is using them, and how often.
Look for these signals:
- Sign-ins and login methods. Which AI tools are employees logging in to, and how: with a company email and SSO, with a company email and a password, with a personal email, or with a social login?
- Usage frequency. A tool used daily by one team matters more than a tool someone tried once.
- Personal-email sign-ups. Employees who sign up for AI tools with a personal address are easy to miss in most reviews.
- Where people reach AI. Browser, desktop app, and phone usage each leave different traces.
- Redundant tools. Teams often pay for or use three AI writing tools when one would do.
Common sources include expense reports, SSO logs, browser-level monitoring, and surveys. Each one shows part of the picture. Expense reports miss free tools. SSO logs miss anything that never went through SSO. Surveys depend on people remembering and reporting.
Step 2: Check what your current tools miss
A common assumption is "We have Microsoft 365, so we already have SaaS visibility covered." Usually that is only partly true.
Microsoft 365 and identity providers such as Okta or Entra ID see the apps that are connected to them. They do not automatically see a tool an employee signed up for directly with a username and password. SSO and identity systems cover the apps you have already integrated, which leaves gaps for the ones you have not.
That is why discovery needs a source that sees sign-in activity where it happens, in the browser. A browser-based approach can surface the tools that SSO and identity systems miss, along with how people log in to them.
Step 3: Decide what to allow, warn, or block
Once you have a list, sort every tool into one of three groups.
|
Policy |
When to use it |
Example |
|
Allow |
The tool is vetted and fits your data rules |
An approved enterprise AI assistant |
|
Warn |
The tool is acceptable for some uses but carries risk |
A general AI chatbot where employees should not paste customer data |
|
Block |
The tool fails your security, privacy, or compliance review |
An unvetted tool with unclear data handling |
A warning is often more useful than a block. A blunt ban pushes people to use tools on their personal devices, where you have even less visibility. A short message at the moment of use, reminding people what not to share, guides behavior without stopping work.
A simple AI acceptable use policy
Your policy does not need to be long. Cover these points on one page:
- Which AI tools are approved, and how to request a new one.
- What types of data must never be entered into AI tools (customer data, credentials, source code, financial or HR records).
- Which accounts to use (company accounts, not personal ones).
- What happens when someone finds or uses an unapproved tool.
- Who owns the list and how often it is reviewed.
Step 4: What to do when you find unauthorized AI tools
Finding unapproved tools is the point of the exercise, so plan the response before you start.
- Do not start with discipline. Most shadow AI comes from people trying to do their job well.
- Triage by risk. Review the tools used most, and the ones likely to handle sensitive data, first.
- Decide allow, warn, or block for each, using the framework above.
- Offer an approved alternative wherever you block something. A block with no alternative drives the behavior underground.
- Communicate the policy in plain language, and tell people how to request a new tool.
- Review on a schedule. New tools appear constantly, so treat this as a recurring check, not a one-time project.
Do you need a CASB or SSPM, or is there something simpler?
Cloud access security brokers (CASBs), SaaS security posture management (SSPM) platforms, and tools like Microsoft Defender for Cloud Apps can all help with SaaS and AI visibility. They also tend to bring setup effort, integrations, and security expertise that a small IT team may not have.
|
Option |
Strengths |
Tradeoffs for a lean team |
|
CASB or SSPM |
Broad coverage and deep controls |
Longer setup, more integrations, often needs security staff to run |
|
Microsoft Defender for Cloud Apps |
Strong fit if you are already deep in the Microsoft stack |
Works best with the surrounding Microsoft licensing and configuration |
|
Browser-extension-based discovery and control |
Quick to deploy, sees sign-in activity in the browser, no agents or proxies |
Focused on browser activity |
If you have a dedicated security team and a complex environment, a larger platform may be worth it. If you are a lean team that needs to see which SaaS and AI tools are in use and set simple rules, a lighter approach may be enough to start.
How LastPass helps with SaaS and AI discovery and control
LastPass approaches this from the browser. With the LastPass browser extension, admins can use SaaS Monitoring and AI Monitoring to discover which SaaS apps and AI tools employees use, including tools that SSO and identity systems miss. Admins can see details such as login methods, usage frequency, and redundant or over-provisioned apps. Admins can also monitor logins from corporate and personal email domains to help tell approved usage from unapproved usage.
With SaaS Protect and AI Protect, admins can allow, warn, or block SaaS apps and AI tools based on the risk and credential signals they see. Both capabilities are available in LastPass Business Max, and they work from the same browser extension employees already use for their logins, with no agents, proxies, or new hardware.
No tool gives you a complete picture of every app and AI tool in use, and no tool can guarantee that a breach or data leak will not happen. What discovery and control give you is a clear starting point, a policy you can enforce, and a way to keep that list current as new tools appear.
Learn more about LastPass Business Max
