Written by: Mio Kanzaki(Security Analyst (AI))Reviewed by: 千葉圭介Updated
Shadow AI is generative AI that employees use without the company knowing. An example would be use through a personal account. The AI Guidelines for Business (Ministry of Internal Affairs and Communications and Ministry of Economy, Trade and Industry) list, as a matter for businesses that use AI to work on, paying attention not to enter confidential information and the like inappropriately.
How does shadow AI work?
Picture an employee who, to work more efficiently, pastes company documents into an outside advice service they signed up for personally, and asks for an opinion. The company cannot know which documents went where.
We could not confirm a common definition of this term by a public body. In this glossary, it means generative AI used for work without the company having permitted it or grasped it. It is the generative AI version of shadow IT, which refers to information systems in general that are used without the company knowing.
The core of the problem is that you cannot see where the information entered goes. As countermeasures, OWASP lists educating users not to enter confidential information, and stating clearly the policy on data retention and the choice not to let your data be used for training. The routes are organized under LLM data leak, and the encyclopedia side is on the shadow IT vulnerability page.
Why does shadow AI matter?
In IPA's "Information Security 10 Major Threats 2026", "cyber risks surrounding the use of AI" was chosen for the first time, in third place in the organization edition. This item, newly chosen by IPA, is material for thinking about the need for a company to decide how AI is used.
However, if you only ban it, the convenience on the floor is lost, and unseen use may increase instead. It lasts better if you first put in writing the line between the generative AI that may be used and the information that may be entered.
As a basic rule for that line, customers' personal information, documents received from business partners, and unpublished information go on the side that is not entered. When you make an exception, also check the terms and settings of the service concerned. The free URL check shows only the outside view of your site, not which AI services your staff use.
Related terms
- Shadow IT: information systems and services that are used without the company knowing.
- LLM data leak: the problem of confidential information escaping through what is typed into an AI or what it answers.
Next step
Find out which generative AI is in use inside the company by asking around. Then decide which services may be used and the range of information that may be entered. The encyclopedia entries are on the shadow IT vulnerability page and the LLM data leak vulnerability page. To draw up the rules, contact us. The free URL check is about what is visible from outside.
Frequently asked questions
Can we prevent shadow AI by banning its use?
A ban alone may leave it being used where you cannot see. A realistic approach is to decide which services may be used and the range of information that may be entered, and to tell employees. OWASP also lists educating users not to enter confidential information as a countermeasure.
Does it mean the same as shadow IT?
The ideas are close. Shadow IT refers to information systems in general that are used without the company knowing. Shadow AI is used as the name for the case limited to generative AI.
Frequently asked questions
Can we prevent shadow AI by banning its use?
A ban alone may leave it being used where you cannot see. A realistic approach is to decide which services may be used and the range of information that may be entered, and to tell employees. OWASP also lists educating users not to enter confidential information as a countermeasure.
Does it mean the same as shadow IT?
The ideas are close. Shadow IT refers to information systems in general that are used without the company knowing. Shadow AI is used as the name for the case limited to generative AI.