.png)
In most large companies, some employees already use AI without saying so, and others quietly refuse to use it. Shadow AI covers both.
The pages that rank for the topic treat it as a security problem: data leaks, compliance, detection tools. That risk is real, but it skips the question an executive team needs answered: why do capable people choose to hide what they're doing?
This article comes from the AI Corner podcast episode in which Thomas Spitz, CEO of AI Partners, talks with Quentin Amaudry, co-founder and CEO of Mendo, and from what we see with our clients. It explains what shadow AI covers, why it spreads, and how to bring usage into the open without banning everything.
Shadow AI is employees' use of AI outside the boundaries their company has set: tools the company doesn't provide, personal accounts, usage nobody declares. It has a less visible side too: employees who say they've adopted AI and don't actually use it.
The first side is widespread. In France, an OpinionWay survey for Cegid published in September 2026 found that 67% of employees who use AI rely on tools their company doesn't provide. A Microsoft France study run by YouGov in January 2026 with 657 managers and executives found that 61% of employees who use AI at work go through personal accounts at least once a week, 38% of them daily.
The second side doesn't show up in any detection tool. Quentin Amaudry sums it up in the episode: if a company doesn't address its teams' fears, it ends up with one problem of people who don't use AI and another problem of people who do. Both have the same cause, which is why they need to be handled together.
Because admitting a productivity gain feels riskier than keeping it quiet. Until the company says what it will do with the time saved, hiding AI use is a rational calculation.
Quentin Amaudry sees this at Mendo, often with high performers who have automated a large part of their work. They stay quiet because they'll be asked to do more, and because their job might be taken away. He calls those fears entirely legitimate, all the more since some companies that promised no layoffs have gone back on their word.
In PagerDuty's April 2026 survey of 1,250 office professionals at large companies, 39% would rather use AI without telling anyone. A Personio survey of 6,000 European employees, published in October 2026, found that these hidden users are also far more likely to be looking for a new job than non-users (67% versus 42%).
Because learning it means losing time today to save time tomorrow, and that bet is frightening when nobody protects the time. The refusal rarely looks like open opposition: people skip the workshops and don't build the agent they were asked for.
According to Microsoft's 2026 Work Trend Index, 45% of AI users say it feels safer to focus on current goals than to redesign their work with AI. Quentin Amaudry sees the same pattern in Mendo's data: typically 70% of a company uses the platform and 30% don't.
This almost never reaches senior management, but we see the effect with our clients. A team moves at the pace of its least advanced member, and once one person has automated their analysis while the person next to them has never opened the tool, they stop working the same way.
| Hidden use | Silent boycott | |
|---|---|---|
| What the employee does | Uses tools the company doesn't provide, or doesn't say they use AI | Doesn't use the approved tools, skips the workshops |
| What they fear | Being asked to do more, or their job disappearing | Losing time, not knowing how, seeing their expertise devalued |
| What leadership sees | Nothing, or gains it can't explain | Adoption that stalls for no visible reason |
| Main risk | Company data in tools nobody controls | Skill gaps inside teams, no return on investment |
| First response | Tools that are good enough and a written rule on time saved | Manager backing and protected time to experiment |
No. A ban doesn't remove the usage, it makes it invisible. Blocking tools without providing an equivalent pushes employees onto personal accounts, where the company sees nothing at all.
The data risk is real: in the PagerDuty survey, 43% of respondents had entered work correspondence and 34% customer data into public AI tools. The 2026 AI barometer by Forvis Mazars and ESSCA, covering 826 IT professionals in France, the UK, Germany and Spain, found that two-thirds see shadow AI in their organisation, and in 12% of companies it persists despite a clear ban.
Shadow AI is often a response to an approved tool that's too limited. In Deloitte's September 2026 survey of 25,000 UK workers, one generative AI user in six (17%) pays for at least one tool themselves for work. We see it with clients too: entry-level licences without access to AI agents, which cap usage as soon as teams want to go further. Gartner makes the same point about agents, as we explained in our piece on AI agent sprawl: rules that only restrict push people toward shadow AI.
Measure real usage of the tools you provide instead of counting licences. A low weekly adoption rate in a productive team signals usage happening elsewhere, or a refusal nobody voices.
Quentin Amaudry puts average weekly adoption at between 20% and 40%, and warns that improving one email a week is enough to count as weekly. The gap between that figure and the usage people report in surveys is a first measure of what the company can't see.
Network monitoring tools show which AI services people use, not why. An anonymous survey, conversations with managers and usage analysis by team tell you more: who uses what, for which tasks, and what the approved tools are missing. That's also the starting point for AI agent governance, which depends on knowing what is already running.
Make disclosure more rewarding than silence: tools that are good enough, a written framework and a clear rule on what happens to the time saved. It starts with senior management, because employees only invest time when their manager asks them to.
For Quentin Amaudry, employees feel entitled to spend time on AI when their manager asks them at the end of the month whether they did, and that manager asks because their own manager does. According to Microsoft, the most advanced employees are far more likely to say their manager openly uses AI (85% versus 64%).
The rule on time saved has to be written down before any initiative starts, because it's the first question employees ask before showing what they've automated. Usage then has to become visible: our team at AI Partners meets every week for a sharing session. Giving the people who built agents responsibility for running them, as in the AI operator role, also gives them a reason to show their work instead of hiding it.
Thomas Spitz, CEO of AI Partners, talks with Quentin Amaudry, co-founder and CEO of Mendo, about AI adoption when the technology changes every week, the role of managers, the usage teams keep hidden and the know-how they hand over to agents. The episode is in French.
Should employees have to disclose their AI use?
For work that leaves the team, such as a client deliverable or a decision, yes, as long as it's framed as a quality rule rather than a confession. Only require it once you've said what happens to the time saved, or the rule will push usage back underground.
Who should own shadow AI: IT, HR or the business?
IT chooses the tools and protects the data, but the causes of shadow AI are managerial, and IT on its own usually ends up blocking. It needs to be owned by a member of the leadership team, with HR setting the framework and managers making it work day to day.
How many employees use AI without telling anyone?
Between a quarter and a third of users, according to surveys published in 2026: 31% in the UK according to Deloitte, 23% across four European countries according to Personio. In France, Microsoft found that 61% of employees who use AI go through personal accounts every week.
Where do you start reducing shadow AI?
Start by measuring what your teams actually use and what the approved tools lack, before deciding to block anything. Then write the rule on time saved, have managers carry it and involve teams before you deploy.
Measuring real adoption, setting the framework and bringing managers on board is the work of our AI First Organization engagement, and what we build into teams through AI Academy. Book a 30-minute call and we'll look at what your teams already do with AI, including what they don't tell you.
© 2026 AI Partners. All rights reserved. Designed & Developed by DivMonk.
