Skip to content
Vol. 1 · Ed. 2026
CyberGlossary
Entry № 1026

Shadow AI

What is Shadow AI?

Shadow AIThe use of AI tools, models, or services by employees without the knowledge or approval of an organisation's security, privacy, or governance functions.


Shadow AI is the AI-era successor to shadow IT. Employees paste source code, contracts, customer records, or strategic documents into consumer chatbots; install unsanctioned browser copilots and IDE extensions; or fine-tune local models on confidential data. Risks include data exfiltration, intellectual-property loss, regulatory non-compliance (GDPR, HIPAA), prompt injection through unmanaged tools, and uncontrolled model output flowing back into production decisions. Effective mitigation combines an AI inventory and policy, sanctioned alternatives (enterprise GenAI with DLP and audit), egress and CASB controls, browser-isolated workflows, and clear user training. Programs typically align with NIST AI RMF and ISO/IEC 42001 to integrate Shadow AI controls into broader AI governance.

Examples

  1. 01

    Engineers pasting proprietary code into a free consumer chatbot to debug it.

  2. 02

    A marketing team using an unvetted AI translation service that stores submitted text on third-party servers.

Frequently asked questions

What is Shadow AI?

The use of AI tools, models, or services by employees without the knowledge or approval of an organisation's security, privacy, or governance functions. It belongs to the AI & ML Security category of cybersecurity.

What does Shadow AI mean?

The use of AI tools, models, or services by employees without the knowledge or approval of an organisation's security, privacy, or governance functions.

How does Shadow AI work?

Shadow AI is the AI-era successor to shadow IT. Employees paste source code, contracts, customer records, or strategic documents into consumer chatbots; install unsanctioned browser copilots and IDE extensions; or fine-tune local models on confidential data. Risks include data exfiltration, intellectual-property loss, regulatory non-compliance (GDPR, HIPAA), prompt injection through unmanaged tools, and uncontrolled model output flowing back into production decisions. Effective mitigation combines an AI inventory and policy, sanctioned alternatives (enterprise GenAI with DLP and audit), egress and CASB controls, browser-isolated workflows, and clear user training. Programs typically align with NIST AI RMF and ISO/IEC 42001 to integrate Shadow AI controls into broader AI governance.

How do you defend against Shadow AI?

Defences for Shadow AI typically combine technical controls and operational practices, as detailed in the full definition above.

What are other names for Shadow AI?

Common alternative names include: Unsanctioned AI, BYOAI.

Related terms