Summary
- The NCSC has issued guidance warning that unapproved AI tools used by staff — so-called shadow AI — pose measurable security risks to organisations.
- Corporate data entered into unsanctioned AI systems may be exposed to third parties or retained by external providers outside organisational control.
- Shadow AI mirrors earlier concerns about shadow IT, but with the added complexity of AI systems that may train on or log submitted data.
- Organisations are encouraged to establish clear AI usage policies and approved toolsets to reduce unsanctioned adoption.
- The risk is not hypothetical — employees are already using publicly available AI tools for work tasks, often without security team awareness.
A familiar problem, a new wrapper
Shadow IT has been a persistent headache for security teams for well over a decade. Employees adopt convenient tools that solve a problem quickly, and governance catches up later — sometimes much later. The NCSC is now drawing attention to the same dynamic playing out with artificial intelligence, and the risks are at least as significant.
What the NCSC is warning about
The UK’s National Cyber Security Centre has cautioned that unapproved AI tools can expose corporate data and introduce new security risks into organisations. The core concern is straightforward: when staff submit work-related information into external AI systems that have not been vetted or approved by their organisation, they lose control of where that data goes and how it may be used or stored.
The data exposure pathway
Unlike many traditional shadow IT tools, AI systems present a distinct data risk. Queries, documents, and prompts submitted to public AI services may be logged, retained, or potentially used to improve the underlying model, depending on the provider’s terms and configuration. An employee pasting a client contract or internal strategy document into an unapproved AI assistant is, in effect, transmitting sensitive information to an external party. Most staff doing this are not acting with malicious intent — they are simply trying to work more efficiently. That does not reduce the exposure.
Why this is harder to govern than shadow IT
Traditional shadow IT — think unauthorised cloud storage or productivity apps — was at least visible at the network level and could, in principle, be blocked or monitored. AI tool usage is harder to detect. Many public AI services run over standard HTTPS, making traffic-level detection difficult without more intrusive inspection. And the tooling is proliferating rapidly: browser extensions, embedded assistants, API-connected plugins, and standalone web applications all provide entry points. A blanket prohibition without an approved alternative is also unlikely to succeed; staff will find workarounds.
The governance gap
Many organisations have not yet established formal AI usage policies, and fewer still have defined an approved set of AI tools for different use cases. Without that framework, employees are essentially making individual risk decisions on behalf of the organisation every time they reach for a convenient AI tool. The NCSC’s warning is a prompt to close that governance gap before it widens further.
Scope of the problem
It is worth noting that the NCSC’s warning is based on the general risk landscape rather than a specific reported incident, and the source material does not cite particular breach cases attributable to shadow AI at the time of publication. What is clear from the advisory is that the behaviour is already occurring and that the potential for harm is real and present.
Why it matters
For CISOs, shadow AI is a data governance and third-party risk problem that sits squarely in the gap between policy and employee behaviour. The challenge is not just technical — it is cultural and organisational. Staff who are not provided with sanctioned, capable AI tools will source their own, and sensitive data will follow. Without visibility into which AI tools are in use across the organisation, security leaders cannot assess exposure, enforce data handling obligations, or respond meaningfully if something goes wrong. The NCSC’s advisory should serve as a prompt to conduct an honest audit of current AI tool usage, formalise an acceptable use policy, and accelerate the procurement or approval of enterprise-grade alternatives that meet security and privacy requirements.
What to do now
- Establish a formal AI acceptable use policy that clearly defines which tools are approved for which categories of work, and communicate it to all staff.
- Conduct an audit of AI tool usage across the organisation, including browser extensions and third-party integrations, to understand current exposure.
- Evaluate and approve enterprise AI tools that offer appropriate data handling guarantees, reducing the incentive for staff to seek unsanctioned alternatives.
- Review data classification practices to ensure staff understand what categories of information should not be submitted to external systems of any kind.
- Engage HR and legal teams to ensure AI usage expectations are reflected in onboarding, training, and acceptable use frameworks.
