There is a good chance that some of your most capable people are using AI tools you do not know about, to do work you would be impressed by, and saying nothing. In Deloitte's UK GenAI Workforce Survey, published 16 September 2026, 63% of 25,000 surveyed UK workers said they knowingly used generative AI for work. Among the 16,749 GenAI users, 31% said they used it without their employer's knowledge. This is a reported survey response, not a count of every worker in the country. Ipsos conducted the online survey of employed and self-employed adults aged 18 to 70 from 7 May to 10 June 2026.
What is shadow AI?
Shadow AI is the use of generative AI tools for work without the employer's knowledge or approval. This can include personal accounts, unapproved apps or AI features embedded in other tools. The term describes an operating and governance gap; it does not tell us whether the person's purpose is harmful.
What the UK numbers say
In the same Deloitte survey, 65% of GenAI users reported a lack of convincing organisational leadership on how to use it, and around half of users reported no formal training in safe and effective workplace use. Among weekly users, 64% worried managers would think AI could do their jobs; the survey also reports 23% perceived a workplace stigma around using AI. These figures have different denominators: neither the 31% hidden-use figure nor the 64% concern figure is a percentage of all UK workers. Separately, in polling of more than 1,000 UK managers released 10 June 2026, the Chartered Management Institute found that 13% strongly agreed senior leaders actively used AI tools themselves, while 12% said they were very confident managing teams using AI. Those are responses to distinct questions, not an observation of all managers' behaviour. The findings show why practical guidance and honest discussion belong alongside security controls; they do not prove why any individual conceals use.
The temptation is to treat this purely as a governance problem: unsanctioned tools, data leaving the building, risk to manage. Those concerns are legitimate. But reading shadow AI only as a security issue misses the more important signal. People hide things at work for a reason, and the reason is usually a lack of safety.
What hiding really tells you
When employees use AI productively but do not disclose it, they may be making a quiet calculation. They may fear that admitting they used AI will make their work seem less valuable, or less their own. They may worry that visible AI use marks them as replaceable. They may simply not know what is approved. These are hypotheses to ask about, not diagnoses of any individual. Psychological safety, the shared belief that you can take an interpersonal risk without punishment, helps make an honest conversation possible.
This is the deeper cost of shadow AI. An organisation where people hide their most effective working methods cannot learn from them. The best techniques stay locked inside individuals instead of spreading. The official picture of how work gets done drifts further from reality. And leaders make decisions about training, tooling and productivity based on information they cannot fully trust.
Build this capability in your organisation
Growth Performance designs bespoke programmes that turn insight into measurable behaviour change.
From permission to transparency: a manager's checklist
- Say what AI is for (Adoption). Name approved tasks, prohibited uses and who owns the output.
- Publish guardrails (Adaptation). Identify approved tools, sensitive data that must not be entered and required verification and escalation routes. Involve security and data protection colleagues.
- Model use in the open (Adoption). Show a real example, including where the output needed correction. Do not demand that others disclose use if you cannot explain how disclosures will be handled.
- Reward safe disclosure (Adoption). Thank people for surfacing useful practices and risks; offer practical training rather than treating every question as a violation.
- Review together (Adaptation and Acceleration). Test one workflow, inspect the quality and risks, then update the shared guidance. Use the AI team charter to record it.
What this means for leaders
Shadow AI can signal a gap between the written policy and the work people actually do. It also creates real security and data risks. Ask about both. The Adoption condition page, Adoption Condition Diagnostic and Manager AI Confidence Assessment offer routes to improve the manager response.
Try this: one week
On Monday, clarify the rules with your information-security lead. On Tuesday, invite examples of helpful AI-assisted work without asking anyone to disclose confidential material in a group. On Wednesday, pick one approved task and agree what must be checked. On Thursday, test it with a named reviewer. On Friday, write down the changes to the team's guidance and a route for private questions. This is a learning exercise, not a survey estimate of hidden use. See AI adoption guidance and the verification habit.
Frequently asked questions
Is shadow AI a security risk? It can be, particularly if sensitive information is entered into an unapproved tool. Classify the data and check the tool before use; do not assume every undisclosed task carries the same risk.
How should a manager respond? Make approved use and safeguards explicit, invite questions without blame, model checking and review the rules with the team.
How much do UK workers spend on their own AI tools? Deloitte's survey announcement estimates around £1 billion spent personally by British workers on GenAI for work. That is a survey-based aggregate estimate, not an individual monthly budget. Ask about cost and procurement issues in your own team.