Shadow AI at work: most UK staff use unapproved AI tools, Microsoft finds
Research from Microsoft shows a large majority of UK workers turn to consumer AI tools their employers have not approved, often weekly. The trend creates security risks but also points to where demand for approved tools already exists.

Key takeaways
- Microsoft research found that 71% of UK employees have used unapproved consumer AI tools at work, with more than half doing so weekly.
- More than a quarter of workers say their employer offers no approved AI alternative, and 41% use tools they already know from their personal lives.
- Unapproved AI creates security risks because organisations lose visibility into how company data is shared and processed.
- Businesses can respond by adopting the tools employees already use, adding shared skills and applying security controls that outlast individual tools.
In this article
Most UK workers use AI their employer did not approve
A large share of UK staff are using consumer AI tools at work without their employer's approval, according to Microsoft research reported by TechRadar on 19 November 2026. The study found that 71% of UK employees have used such tools on the job, and more than half do so every week.
For 41% of these workers, the tools are ones they already rely on in their private lives. Meanwhile, more than a quarter say their employer simply offers no approved alternative.
This pattern has been called the Shadow AI economy. It describes workers who bring their own AI tools into the workplace, in the same way staff once used personal devices or apps without official sanction.
Why employees bypass approved tools
Employees adopt these tools because they find them genuinely useful, often before their employers have caught up with the demand. Consumer AI services set expectations around speed, flexibility and easy experimentation. Workers can try different approaches, switch between tasks and apply the technology to whatever job is in front of them.
Approved enterprise AI has to meet stricter demands. It must protect company data, respect access controls, plug into existing systems and follow governance rules. Staff still expect it to be as easy and effective as the consumer versions, though.
If the sanctioned option is more restrictive or less suited to daily work, employees tend to drift back to the tools that serve them better. Simply blocking unapproved tools may lower the immediate risk, but it does nothing to make the official alternative more useful or to explain why staff sought out other options in the first place.
The security problem with unsanctioned AI

The risks are hard to ignore. When staff use personal accounts or unapproved services, organisations lose sight of what company information is being shared, how it is processed and where it might end up.
TechRadar reports that this lack of visibility is the core security challenge of Shadow AI. Businesses cannot audit or control data flows they cannot see, which leaves sensitive material outside their governance rules.
Yet the same behaviour carries useful information. If workers keep returning to AI to analyse data, draft content or speed up tasks that once took hours, they are showing exactly where AI has value in their daily work.
Turning Shadow AI into approved practice
The article's author describes how their own organisation responded. Staff were already using tools such as ChatGPT and Claude to build presentations, write drafts and produce everyday assets. Rather than banning them, the company examined where time was being saved and moved those use cases into an approved enterprise environment.
The organisation also built shared skills that anyone in the company can use. A shared skill in ChatGPT or Claude can embed enterprise messaging, brand language and the context employees need to represent the company accurately. That way, anyone can produce a presentation that follows standard brand guidelines and style.
The result is a balance: staff keep tools they find useful, now with company context included, while the business gains visibility into usage and data flows, wrapped in enterprise security and access controls.
What businesses should do next
The advice for other companies is to watch what their own employees are already doing. Instead of trying to predict every AI use case centrally, businesses can look at where staff are turning to AI and bring what works into a sanctioned environment.
The tools themselves will change. Workers favour services such as ChatGPT, Claude or Gemini today, but new options will appear and existing ones will improve. Companies cannot rebuild their setup every time that happens.
For that reason, the argument goes, controls should outlast the tools. Access to company data, permitted actions and management visibility should not depend on which model a person uses. Applying those controls across the whole AI environment lets a business swap tools as needs change without rebuilding everything around them. That way, the next tool staff discover does not have to become the next Shadow AI problem.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



