Shadow AI.

What is shadow AI?

The UK National Cyber Security Centre (NCSC) describes shadow AI as the use of AI technology which is not captured in an organization’s approved systems and processes. NCSC says it is a form of shadow IT, which means technology used outside what the organization has approved.

A simple illustration: an employee pastes work text into a public AI chatbot that the organization has not approved. The organization may not know it happened.

Why do staff use unapproved AI tools?

NCSC says organizations’ policies and guidance have not always developed at the same pace as AI adoption. When approved options do not meet what the work needs, staff can turn to other tools.

What risks does shadow AI create?

NCSC names three main risks.

• Data exposure. NCSC says giving shadow AI access to company or customer data likely increases the risk of data breaches and intellectual property loss.
• Loss of control. NCSC says data sent to consumer services may be stored, retained or used to improve the service, and the organization loses visibility of it.
• Attacker opportunities. NCSC says AI systems contain complex vulnerabilities that attackers can exploit to reach the data, services and privileges the tool has.

OWASP, which publishes a top-ten list of risks for large language model applications, explains what kinds of data are at stake. It says sensitive information can include personal identifiable information, financial details, health records, confidential business data, security credentials and legal documents.

How can an organization reduce shadow AI?

NCSC recommends a transparent cyber security culture, understanding why staff adopt unapproved tools, and providing secure approved alternatives. NCSC puts the core problem plainly: you cannot manage what you do not know.

OWASP gives related advice for users of language model tools. It says to provide guidance on avoiding the input of sensitive information, and to offer training on best practices for interacting with these tools securely.

Do official bodies define shadow AI?

NCSC does, in its blog on the hidden risks of shadow AI. The OWASP page on sensitive information disclosure does not use the term. This page relies on NCSC for the definition.

Is shadow AI the same as shadow IT?

Not exactly. NCSC calls shadow AI a form of shadow IT. Shadow IT is the wider idea of unapproved technology. Shadow AI is the part that involves AI tools.

Does shadow AI always mean data theft?

No. Shadow AI describes the lack of approval and visibility. NCSC links it to higher risk of data exposure, but it does not say every case leads to harm.

Related terms: generative AI, AI agents and data exfiltration.