The National Cyber Security Centre (NCSC) has published advice on the cyber security dangers of shadow AI, warning that employees who turn to unapproved AI tools for work tasks can expose sensitive data, remove it from organisational control and create new routes for attackers.
In a blog published on 7 September, the NCSC warned that organisational policies and guidance have not kept pace with the rapid growth of AI use in workplaces, and that where cyber security policies cannot meet business needs, staff are likely to adopt new AI services before their employer has assessed them or provided approved alternatives.
It cited Microsoft research findings that 71% of employees report using AI tools not approved by their employer, and said that the trend is likely to strengthen as AI capabilities become cheaper and more readily available.
The NSCC defines shadow AI as the use of AI technology not captured in an organisation's approved systems and processes, a form of shadow IT and the NCSC identified three main risks:
- First, giving unapproved AI tools access to company or customer data increases the likelihood of data breaches, loss of intellectual property and regulatory failures.
- Second, employees who transfer sensitive or proprietary information to consumer AI services reduce the organisation's visibility and control over it, because the information may be stored, retained or used to improve the service outside established security and governance arrangements unless specific privacy controls are in place.
- Third, AI agents are complex software that can contain critical vulnerabilities, and an attacker who exploits one gains the same access to data, services and privileges the agent holds. The NCSC says attackers are highly likely to use agents with looser guardrails to exploit vulnerabilities or misconfigurations in the wider corporate IT estate.
Under the UK GDPR. personal data entered into a consumer AI service is a disclosure to a third party that requires a lawful basis and, where the provider processes it for its own purposes such as model training, makes that provider a controller rather than a processor. Article 32 requires controllers to implement appropriate technical and organisational measures to secure personal data, and Article 5(1)(f) makes security a principle of processing. Many consumer AI services also process data outside the UK, engaging the international transfer provisions in Chapter V.
The NCSC did not recommend that individuals stop using AI but advises staff to think carefully about which apps and services they use before sharing data, and warns that sticking with a familiar personal AI service for work tasks can cause real problems for an employer.
For organisations, the NCSC recommended adopting a positive cyber security culture in which open communication makes staff less likely to resort to shadow tools, and understanding why people are using shadow AI so that secure alternatives can be provided.
It also highlighted its own guidance with international partners on the careful adoption of agentic AI services, and to its 20 August blog on managing the cyber risk of agentic AI.
