Call Contact Us

What Is Shadow AI? And How It Differs From Shadow IT

Shadow AI is artificial intelligence used for work without the organization's formal review and approval, so it sits outside what IT and security normally see and manage. Typical forms are a free chatbot on a personal login, an AI feature that appeared inside software the company already pays for, and an AI agent connected to company email or files. It is a form of shadow IT, with a difference that matters: the risk travels in what people type or upload, and in what the tool is allowed to do.

Our shadow AI guide covers detection and governance in full.

In short

Why it is not just shadow IT

Sensitive information can travel out in a prompt, the answer flows back into the work, and some AI tools can take actions once they are connected to company systems.

Is it always dangerous?

No. The risk depends on the data involved and on what the tool can reach.

What counts as shadow AI?

A quick test settles most cases. Is it AI? Is it being used for work? Has the organization approved that tool for that use? If the answers are yes, yes, and no or nobody knows, it is shadow AI. Review means someone has checked what the tool does with your data, the contract terms and what it can connect to.

It reaches well past the free chatbot: the writing assistant in the email program, the page-summarizing extension, the meeting note-taker, the coding assistant and the inbox automation. The guide maps where shadow AI hides.

An approved AI tool used within its rules is not shadow AI. Shadow AI is a visibility problem first: an organization cannot manage what it does not know is there.

Shadow AI examples at work

Shadow AI does not have to look like misconduct. It can look like someone trying to finish the day's work. These are illustrations, not reported cases.

The nurse

At the end of a shift, a nurse pastes notes on several patients into a free chatbot on a personal phone to write a quick handoff for the next shift. Those details now sit with a service the hospital never vetted.

The loan officer

To clear a backlog, a loan officer uploads a member's tax returns and bank statements to a free AI tool for a cash-flow summary. The credit union has no contract with that provider and has not reviewed what it keeps.

The plant engineer

To get a stalled line running before the next shift, an engineer pastes machine logs and part specifications into a chatbot. The company's engineering data has now left the building.

The city clerk

Wanting accurate minutes, a clerk connects an AI note-taker to a work calendar so it joins meetings and emails summaries. An outside company now holds recordings of internal discussions, and nobody at the city has decided how long they are kept.

The developer

To ship faster, a developer uses a coding assistant on a personal plan with company code, and gives an AI agent an access key to run tests unattended. The code goes to an outside service, and the agent acts with the developer's access.

The finance director

A finance director tries the new AI assistant in spreadsheet software the company has used for years and asks it to analyze results that are not yet public. The director took an approved product to mean an approved feature, but nobody has reviewed what the new assistant does with the data.

None of them meant harm. But nobody reviewed any of it, so the organization cannot tell the harmless uses from the risky ones. The guide's section on shadow AI by industry explains how health, banking and government rules come into it.

Shadow AI vs. shadow IT: what is the difference?

Shadow IT is the older idea. A NIST guide on mobile device security, SP 800-124 Revision 2, says shadow IT "typically denotes staff members' work-related use of IT-related hardware, software, or cloud services without the knowledge of the IT organization." Shadow AI is the part of that problem that involves AI. It differs in where the risk sits: what people type or upload, the answers that come back into the work, and what an AI agent can do with someone's access.

Both can start when the approved route is slow or missing, and both can leave the organization blind to where its information goes. What AI adds to the old problem:

  • The data can leave in the prompt. Any question can carry sensitive information, because people paste in the very thing they need help with. With an outside tool, that information leaves the organization.
  • The provider may keep it, or use it to improve its models. For example, OpenAI says that on its individual plans (Free, Plus and Pro) it may use your content to train its models unless you opt out. By default, OpenAI says, it does not train on inputs or outputs from ChatGPT Business, ChatGPT Enterprise or its API. Personal and company accounts of the same tool can follow different rules, and a personal paid plan is still a personal account.
  • The answer comes back into the work. An AI tool returns something new, such as a summary, a figure or code, and it lands in a patient letter, a loan file or a product. If it is wrong, the mistake travels with it.
  • Agents may choose their own steps. A conventional script runs the logic written into it. AI agents connected to mail, files or other systems may decide which tools to use as they go, with a person's or a service account's access.
QuestionShadow ITShadow AI
Where the risk sitsIn technology IT does not manage: where data sits, who can reach it and how it is securedWhat people type or upload, what the tool keeps, and what it sends back
How it arrivesA new sign-up, download or purchase, or an add-on to something already in useThe same routes, and often as a new feature inside software already approved
What it does once it is runningWhat it was set up to do: an app stores or syncs data, a script runs the logic written into itAn AI agent given access to systems may choose its own steps, and what it reads can change what it does
How it is foundApp and device inventories, network logs, connected-app records, expense reportsThe same sources, plus the AI settings inside software already approved

Put simply, shadow IT is technology the organization does not know it is running. Shadow AI adds what your people tell a machine, what it tells them back, and what it is allowed to do.

Is all unsanctioned AI use dangerous?

No. Asking a free chatbot to tighten a job posting that is about to go public puts nothing confidential into the tool. The risk climbs with the data, the decision and the access, and the response should climb with it:

Shadow AI can be a sign of unmet demand. The aim is to move it onto tools the organization has reviewed, not to treat users as suspects. The guide explains why a ban alone falls short.

Why do employees use shadow AI?

Like shadow IT, shadow AI can grow where the approved way of doing a job is slow, or where there is none. Other causes are simpler:

  • Nobody explained the rules, or the policy went unread.
  • The AI came to them. A vendor switched it on, or a meeting invite brought a note-taker along.

AI use, approved or not, can also stay hidden. In Microsoft and LinkedIn's 2024 Work Trend Index, a survey of 31,000 people in 31 countries, 52% of people who used AI at work were reluctant to admit using it for their most important tasks. When people are reluctant to admit they use AI, that use is harder to review.

What to do first

These steps make shadow AI visible and cut the worst risk quickly.

01

Find out what is already in use: network and DNS logs, connected apps, browser extensions, expense reports, and a no-blame survey of staff. See how to detect shadow AI on your network.

02

Decide which data never goes into an unapproved tool, such as patient records, customer financial data, source code and designs, then sort what you found with an AI inventory and risk tiers.

03

Give people an approved option for the common jobs, so the approved route is also the convenient one. An approved-tools list is part of the AI governance basics.

04

Write the rules down and teach them, starting with what an AI acceptable use policy should contain.

For a phased plan sized for a small team, see your first 90 days of AI governance.

Shadow AI and shadow IT: common questions

Shadow AI is AI used for work without the organization's formal review and approval. A free chatbot used with a personal login, an AI feature turned on inside approved software, and an AI agent given access to company systems are all examples. It can stay invisible to IT until someone looks.

Shadow IT is any hardware, software or cloud service used for work without the IT team's knowledge or approval. Shadow AI is the AI part of that problem, and it adds risks of its own: sensitive data can leave in what people type, the answers flow back into real work, and AI agents can act with a person's access.

One example of shadow AI is an employee pasting a customer contract into a free chatbot on a personal account to get a quick summary. Others include an AI note-taker that joins meetings without IT review and a browser extension that can read what is on screen. A new AI assistant switched on inside approved software, before anyone checks what it does with the data, counts too.

Yes. When a vendor adds an AI assistant to a product approved before the feature existed, the original review may not have covered it. Until someone checks what the feature does with your data and whether your contract covers it, using it for work is shadow AI. Where a platform has AI settings in its admin console, that is a good place to start.

See what is running in your environment

For detection, policy, frameworks and a 90-day plan, read the full guide to shadow AI.

InfoSight's AI Governance & Exposure Assessment includes Shadow AI Discovery, which looks at what is actually running. The service page lists it as SaaS/AI app inventory, browser extension analysis, DNS/proxy logs, CASB & Defender data and an OAuth app audit.

Ready to Slash Your Cyber Risk in 30 Days?

Schedule a consultation from one of our expert advisors.

No spam—one security expert follow‑up, that’s it.

Want to Receive our Newsletter?

Stay informed of the latest cyber trends.