Why Employees Still Use Shadow AI Despite Your AI Policy

Employees use shadow AI tools at work because they perceive these tools to be faster, more familiar, and often better for productivity than the approved option. And they persist because most of them see little personal risk in doing so. As many as 71 per cent of workers have used unapproved AI tools at work, with 51 per cent confessing to doing so every week. In the UK, research estimates that workers are spending around £958 million of their own money on GenAI tools for work every year, instead of, or in addition to, the approved tools they’ve been given.

Security leaders have no choice but to accept that shadow AI (that is, AI use that’s happening without visibility and approval) exists within their organisation. The question is what to do about it.

Why do employees use shadow AI when they know it goes against policy?

Approved options are missing, slower, or weaker

Microsoft’s UK research found 41 per cent of shadow AI users rely on tools they already use in their personal lives, and 28 per cent said their company offers no work-approved option. In Ireland, only 44 per cent of workers say their employer provides the AI tools and resources they need.

They do not see the risk

Only 32 per cent of UK workers are concerned about the privacy of company or customer data going into consumer AI tools, and just 29 per cent worry about the security of their organisation’s IT systems. Six in ten employees say speed is worth the security risk, and a third admit to sharing enterprise research or datasets with unsanctioned tools. In Ireland, one in ten office workers have entered confidential company data into an AI tool their employer hadn’t approved.

They fear being judged, so they hide it

This is the one discovery tooling alone can’t fix. Almost a quarter of UK workers believe there’s a stigma attached to using GenAI at work, and 64 per cent of weekly users worry their managers will think GenAI can do their jobs. People who feel judged are more likely to hide their use, which makes shadow AI harder to see and thus harder to measure.

What does shadow AI cost when it goes wrong?

Shadow AI’s involvement in security incidents has more than doubled in a year, from 20 per cent to 43 per cent. Breaches involving it now average $5.39 million, up from $4.63 million. Two-thirds of breached organisations had no AI governance policy, and about one in five shadow AI incidents ended with the organisation paying a fine.

Why bans alone do not work

Largely, because they treat a design problem as a discipline problem. The UK’s National Cyber Security Centre isn’t recommending that people stop using AI. Its September 2026 guidance accepts that shadow AI is unlikely to disappear completely, and says the goal should be to reduce risk rather than assume it can be eliminated. Its advice emphasises the importance of understanding why staff use unapproved tools: “You cannot manage what you do not know.”

Blanket bans push AI use further into the shadows, and long-winded approval processes simply get bypassed. Among employees who admit to using unapproved tools, around half have been told informally to stop and a similar share have faced formal consequences. Meanwhile, 77 per cent of workers say AI restrictions are limiting their professional development, and 75 per cent say they’d be likely to move to an employer offering better AI skills development.

Five ways to reduce shadow AI risk

1. Measure real usage, not assumed usage

Use discovery tooling such as Microsoft Defender for Cloud Apps and Microsoft Purview’s capabilities to see which AI services are actually in use.

2. Provide a sanctioned tool people genuinely prefer

If the approved option is slower or weaker than a free chatbot, it will lose. If that option is Microsoft Copilot, fix permissions first, because Copilot oversharing can turn a sanctioned tool into its own data exposure.

3. Write a policy people can follow quickly and easily

Classify tools as sanctioned, limited-use or prohibited. Spell out in plain English what data can and cannot be entered. Publish a fast approval route for new tools, with a stated turnaround time.

4. Train people and make it safe to disclose AI use

Address the stigma head on. Ask employees which tools they already rely on, without blame, so that discovery starts from honesty rather than detection.

5. Give AI an owner and a metric

Assign board-level accountability, for example through a cross-functional AI council, and report on a small set of measures: sanctioned AI adoption, unsanctioned tools discovered, and policy exceptions requested. Treat it as an ongoing exposure, in the same way as continuous threat exposure management treats the wider attack surface.

Three questions every security team should ask

Do we know which AI tools our people are using today, and how do we know?
Do our employees have an approved alternative that they would choose over a free one?
Could any employee tell us, in plain English, what they may and may not paste into an AI tool?

If the answer to any of these is “not sure”, the policy is probably not the control you think it is.

Find out what's really happening

Shadow AI is rarely a single decision. It’s hundreds of small ones, made by people trying to get their work done. Threatscape’s Microsoft Security Practice helps organisations see what is in use, secure the tools they approve, and build governance that employees will follow. Our free eBook, Securing AI with Microsoft 365, written by Threatscape’s own Microsoft MVPs, gives you a practical, step-by-step security baseline before AI sprawl becomes a governance problem.