Almost half of Dutch employees use a private AI account for work: three steps for a firm
· By Bee Flow · 4 min read
The Dutch Alert Online survey of 1 October found that 49 percent of employees sometimes use a private AI account for work, and that 39 percent know which tools are allowed. Three steps for a firm, without blame.

On 1 October 2026 Alert Online, the Dutch cybersecurity awareness campaign, published its yearly survey of how people in the Netherlands behave online. This year's edition was made by ECP, the Dutch platform for the information society, and Behavior Change Group. The headlines were about phishing. Further on in the report (in Dutch) is a figure that matters to any firm with client files: 49 percent of the employees surveyed sometimes, often or always use a private AI account or app for work. And 35 percent use AI tools without the permission or knowledge of their organisation.
Chances are this is happening in your office too, even if nobody has ever mentioned it.
What the survey does and does not say
It is a questionnaire among 609 consumers and 378 employees. The questions about AI at work were answered by 322 employees. People report their own behaviour, and the researchers call the AI questions exploratory. Of the employees at companies with up to 50 people, 71 took part. In that small group the share that uses a private account is about the same, though we think 71 people are too few to draw firm conclusions from.
The researchers add a note of their own: sharing information with AI is not unwanted in itself, and they did not ask in detail how people use it. Using a private account for work and using AI without permission are the two things they do call clearly unwanted.
For an accounting or tax firm it is easy to see why. With a private account the firm has no contract with the supplier, so no data processing agreement either, and nobody can find out what went in. In the same survey 21 percent of employees say they sometimes share customer data with AI, and 11 percent say they often do. What happens to such a message is described in is ChatGPT safe to use at work?.
Why people do it
The report gives two explanations that a firm can do something about.
The first is not knowing. 39 percent of employees know which AI tools they may use for work, and 35 percent know which guidelines their organisation has. Employees who know which tools are allowed use a private account less often.
The second is time. According to the researchers the urge for speed beats an abstract risk as soon as the approved alternative is slower, can do less or does not exist. One in five employees finds the risks of AI acceptable.
There is little ill will in any of this. People want to get their work done.
Step 1: ask, without blame
Put it on the agenda of the next office meeting, and say beforehand that nobody will be called to account for what they have done so far. Ask per task: what do you use AI for, with which account, and what can that account do that our own tools cannot? You get a list like this: rewriting a client e-mail, summarising a long document, looking up a scheme, having an Excel formula written. That list is your set of requirements.
The report recommends an easy-to-reach desk where staff can say which functions they miss. In a firm of twenty people that desk is one colleague with a name.
After that, ask everyone to delete conversations with client data from their private account, and look with whoever handles privacy in your firm at whether anything has to be reported. This post is not legal advice.
Step 2: offer an approved place that is just as easy
The researchers advise an approved AI environment in which data tracking and training are switched off, and which is as fast and as easy to reach as a private tool. That last part weighs heavily. Go through the list from step 1 and check whether the approved place can handle each task. Where it cannot, you know what people will keep opening their private account for.
Look at the threshold as well. A separate login, a slow request procedure or a limit that runs out halfway through the month, and the old habit is back.
Step 3: write down what is allowed
The report calls it the green list: an overview of the tools and situations that have a green light, in place of a list of bans only. One page is enough. How to set it up and introduce it is in a one-page AI policy for a small company, and this page has a policy you can copy.
We would add one thing from the report: the example from the top. Have a partner show in the meeting how he or she uses the approved tool. Ask again after a month who uses what. The list will be shorter, and what is still on it tells you what the approved place lacks.
Where Bee Flow fits
Bee Flow is an AI workspace for the whole firm that you can set up as that approved place: on your own server or on Bee Flow Cloud in the EU, with model providers chosen by the administrator, including EU-hosted and local models. The report recommends software that filters personal data automatically before a message goes to the AI server. In Bee Flow that is Privacy Shield: a message with personal data is held back before it reaches the model, in every licence. How that works is in what happens when staff paste customer data into an AI chat. For the looking-up on the list from step 1 there are knowledge bases, which answer from your own documents and show the source with each answer.
The page for accounting and tax firms shows what this looks like in practice, privacy and security describes the set-up, and the plans have their own page. To put your own firm's list next to it, plan a call.