What is shadow AI? Unapproved AI tools at work, defined

Shadow AI is the use of AI tools, models or agents by staff outside the systems and processes their organization has approved. The UK National Cyber Security Centre (NCSC) treats it as a form of shadow IT, and the security concern is company or customer data flowing to services nobody has assessed.

Why it matters to a CISO

It is widespread enough to assume it is already present. The NCSC's post on the hidden risks of shadow AI, published on 7 September 2026, cites a study in which 71% of employees said they had used AI tools their employer had not approved. It judges that giving such tools company or customer data likely raises the risk of breaches, intellectual property loss and regulatory failures.

The data question predates the term. In March 2023 guidance on public chatbots, the NCSC warned that queries are stored and will almost certainly be used to develop the service. Agents raise the stakes again: the 2026 post expects attackers to use agents with looser guardrails to exploit weaknesses in the wider corporate network.

How it is governed

The NCSC's goal is reduction, not elimination, and its starting point is visibility, summed up in its line "You cannot manage what you do not know." Its 2023 advice still sets the baseline for staff, which is to keep sensitive information out of public LLMs, while making sure people who want to experiment can do so without putting organizational data at risk.

In the NCSC's view, restriction alone does not solve the problem. It expects that where security policy cannot meet business needs, staff will keep adopting new AI services before the organization has assessed them, and it does not expect shadow AI to disappear completely. Its answer is to give employees AI tools that meet their needs and to encourage open discussion of security, which it says makes people much less likely to turn to shadow services.

Where it appears on cisonews.ai

The explainer what is shadow AI and how should CISOs govern employee use of generative AI gives the full governance approach, and how to set up an AI governance program places it in a wider program. The incident log records Samsung's 2023 restriction on generative AI after sensitive data reached ChatGPT.

Related terms

Agentic AI, AI agent and AI guardrails.