After someone reports a customer list pasted into a chatbot, the IT team at a consultancy blocks a few public AI tools on the network. There is no approved alternative yet; that is planned for later. A week later, a project manager has until five o'clock to summarise a thirty-page report for the client. She picks up her phone, opens a free chatbot on her personal account and pastes in the report, client name on the first page.
The block is still in place. The work has simply moved somewhere nobody can see it.
There is nothing strange about that. The project manager wanted to do her work well and on time, and AI helps with that. The block took away the tool, not the task. That is what this piece is about: not whether you may restrict tools, because you may, but what it takes for a restriction to do what you want.
Why do people look for a workaround?
Because the task is still there. Summarising a report, rewriting a letter, explaining a piece of code: anyone who has seen how much time AI saves on those tasks will not go back to half an hour of manual work because a domain is blocked on the network. On a phone or a home connection, the same tool works fine.
The pattern is wider than one firm. Microsoft and LinkedIn's 2024 Work Trend Index, based on 31,000 knowledge workers in 31 markets, found that 78 per cent of AI users bring their own AI tools to work. People do not wait for policy when the work has to be done today.
In late 2025, an internal analysis at the Dutch municipality of Eindhoven found that employees had uploaded files containing personal data of residents and colleagues to public AI websites. The municipality then blocked those websites: an understandable first step. The next question is the same as at the consultancy: where do employees now go with the work they used those tools for?
What changes when there is a workable alternative alongside it?
The restriction now points somewhere. Instead of "this is not allowed", the organisation says "use this tool for that". That is a different conversation.
Suppose the consultancy has set up Microsoft Copilot with a business account and a data processing agreement. When the project manager opens DeepSeek, a tool the firm has assessed and does not allow, she immediately sees where she can go instead. The report goes into a tool covered by a data processing agreement, on her laptop, where it is visible. The restriction is then no longer a ban on AI but a choice of tool.
An alternative is workable when it does the same job, about as quickly, and is easy to find. A business tool that takes a two-week request to get hold of is no alternative on a Thursday afternoon. Free versus business AI accounts covers what the difference between the two means in practice.
Which rules belong with a restriction?
Two kinds, and they often get mixed up.
The first is about the tool: which tools are allowed, which are not and why. One sentence per decision is enough, as long as the reason holds and the alternative is named. Four statuses for AI tools explains how to record that for each tool.
The second is about the data: what does not belong in any AI tool, including the approved one. Client files, medical data, performance reviews. Restricting a tool does not solve that, because the data simply moves on to the next tool. It helps to keep that rule separate from the tool list, and to give real examples of what does and does not belong in a prompt.
How do you help people when they open a tool?
By showing the decision where the work happens, rather than on the intranet. For that, you first need to know which tools are in use.
DeepSeek
