May I use ChatGPT as a municipal employee?

Your municipality decides that, not the tool. Some municipalities have allowed Copilot and decided nothing about ChatGPT yet; others have set up a business ChatGPT environment. Check what has been decided about the tool before you use it, and in any case do not use a personal free account for text that contains personal data.

May a citizen service number go into ChatGPT?

No. A citizen service number (BSN) makes someone directly identifiable and the municipality may only use it for the purpose it was issued for. In a prompt it is almost never needed: the letter does not get better from the real number. Take it out or replace it with a placeholder before you paste.

May I have an objection letter summarised?

The summary of the argument often can be, the person behind it does not belong in it. Remove name, address, reference, date of birth and case number and describe the situation generally enough that the objector is not recognisable. If the situation itself stays recognisable, for instance because it concerns the only business of its kind in the village, do not do it in a tool that has not been approved for that.

Is it safe once I remove the name?

Not always. An address, a file number, a date of birth or a combination of details (single parent, three children, street name, on benefits since March) can still make someone recognisable, especially in a small municipality. Look at the whole text, not only the name.

What if I pasted personal data by accident?

Report it to the data protection officer or through the route your municipality has for this. Personal data in a chatbot can be a data breach, and the Dutch Data Protection Authority has received notifications about exactly this. Reporting is not a confession; it is how the municipality can assess what needs to happen.

ChatGPT in municipalities: what may go in the prompt?

Quick answer

In a municipality the task may usually go into ChatGPT, the personal data may not. Making a letter easier to read, summarising an objection or answering a council question works without the citizen service number, name, address, file number and the details that make someone recognisable. What never belongs in a prompt: the citizen service number (BSN), benefits data, anything from youth care or debt counselling, health data and the contents of a case file. Which ChatGPT you use matters too: a free personal account is not an environment the municipality set up, and which tool is allowed is your municipality's decision, not a property of the tool.

Keep the task, take the identifying data out before you paste

BSN, benefits, youth care, debt counselling and health never belong in a prompt

A free personal account and an environment set up by the municipality are two different decisions

Without a name someone can still be recognisable by address, file number or the situation itself

Which tool is allowed is your municipality's decision; when in doubt, ask the DPO

A caseworker in the benefits department has to rewrite a letter about a repayment in plain language. The letter is legally correct and unreadable. She opens ChatGPT, pastes the whole letter in, and within a minute has a version the recipient will understand. At the top of the letter were a name, an address, a citizen service number and the amount.

The task was fine. The personal data did not need to be there. That is the whole answer: in a municipality, the task may usually go into an AI tool, the data about the resident may not. What exactly has to come out depends on what you are doing.

Why does this weigh more heavily in a municipality?

Because the text you handle on an ordinary day is almost always about someone. A benefit, an objection, a report to the neighbourhood team, an application for a parking permit. Part of it is special-category or highly sensitive: health, debts, a family situation in youth care. And the citizen service number (BSN) appears on nearly every document the municipality sends.

The Dutch Data Protection Authority received several breach notifications in which employees had put personal data into an AI chatbot, and advises organisations to make clear agreements on whether it is allowed and with which data. For a municipality there is also the BIO, the Dutch government information security baseline, which asks for classification and careful handling of residents' data. What that means for AI use is in BIO and NEN 7510: AI and privacy in government and healthcare.

This is not about you as a weak spot. You use the tool because the letter gets better. The only question is what goes into the tool with it.

Which ChatGPT are you actually using?

There is no single ChatGPT. A free personal account runs under consumer terms: the input can be used to improve the model unless you switch that off yourself, and the municipality has no data processing agreement with it. OpenAI states on its enterprise privacy page that its business environments do not use input for training by default. That is a different product with different agreements, even though the screen looks the same. The difference is set out in free versus business AI accounts.

Which variant you may use is a decision of your municipality. Many municipalities have allowed Copilot and decided nothing about ChatGPT yet. "No decision yet" does not mean it is allowed; it means nobody has looked at it. As long as that is the case, no text with personal data should go in.

In municipalities that work with BeeSensible, the municipality's decision appears in the browser at the moment you open an AI tool: allowed, limited use with a condition, no decision yet, or not allowed with the alternative named. It does not stop you; it shows what has been decided.

Search tool, vendor or category…

865 tools

ChatGPT🇺🇸

HighAllowed

Claude🇺🇸

MediumAllowed

DeepSeek🇨🇳new

CriticalNot allowed

Perplexity🇺🇸new

MediumNo decision yet

Mistral🇫🇷

LowNo decision yet

Midjourney🇺🇸

MediumNo decision yet

Every tool, scored for risk

865 AI tools, each scored Low to Critical, with nothing pre-approved or pre-blocked until someone decides.

More about AI Tools

What comes out, per task?

The rule of thumb from what you can and cannot share with AI applies here too: if you would not email this to a stranger outside the municipality, it does not belong in a chatbot. In a municipality that looks like this per task.

Task	What you want to paste	What comes out	How you still do the task
Rewrite a letter in plain language	The whole letter	Name, address, BSN, case number, amount, dates	Paste only the paragraphs that read badly, with placeholders like [name] and [amount]
Summarise an objection	The objection as submitted	Name, address, reference, date of birth, and details that make the situation recognisable	Describe the argument, not the person: "the objector argues the deadline was too short"
Answer a council question	The question and the file behind it	Residents' cases used as examples	Work from the public documents and the policy, not the file
Write up a neighbourhood team meeting note	Your notes	Everything: it concerns a family, health or debts	Not in a tool that has not been approved for this; ask the DPO which route exists
Draft an email to a resident	The previous email thread	Name, address, file number, the substance of the case	Describe the kind of message: "a friendly reminder that an application is incomplete"

The three ways to take data out of a prompt, remove, replace and mask, are covered with examples in the prompt redaction guide. Replacing with a placeholder works best for a letter: the tool still understands the structure and the real value stays with the municipality.

What if the name is already gone?

An employee in the permits department has an objection summarised and has neatly removed the name. What remains: the address, the reference of the decision, and the sentence "as the only bakery on the church square, my business is disproportionately affected". In a municipality of twelve thousand residents, that identifies the objector.

Without a name, someone is not automatically anonymous. An address, a file number, a date of birth or the situation itself can together point to one person. How that works, and why it happens faster in a small municipality than in a large city, is in quasi-identifiers explained. So look at the whole text before you paste, not only the name at the top.

If the situation stays recognisable even without the data, the objection does not belong in a tool that has not been approved for case work. That is no loss: summarising five pages yourself takes ten minutes, and the file stays where it belongs.

How do you see it at the moment itself?

A list in your head works until it gets busy. At half past four, with three letters to go, a BSN slips through. That is why it helps if sensitive data is marked while you type or paste, before you send.

BeeSensible marks sensitive data such as a BSN, an IBAN or a name while you write a prompt, in the tools you already work in. You choose what to do with it: mask, remove, or leave it. What you type is not stored and an admin sees only counts, never your text. What it does exactly is on the AI tools page, and the rest of the subject is under AI governance.

Tomorrow the caseworker from the first paragraph pastes only the paragraphs that read badly, with [name] and [amount] in place of the real data. The letter comes out just as well, and the repayment stays where it belongs: with the municipality.

ChatGPT and personal data in municipalities: what may and may not go in the prompt?

Why does this weigh more heavily in a municipality?

Which ChatGPT are you actually using?

Every tool, scored for risk

What comes out, per task?

What if the name is already gone?

How do you see it at the moment itself?

Common questions

Would you rather someone showed you?

Rather talk to someone first? Book a call.

#Why does this weigh more heavily in a municipality?

#Which ChatGPT are you actually using?

Every tool, scored for risk

#What comes out, per task?

#What if the name is already gone?

#How do you see it at the moment itself?

Common questions

Which AI systems are used in our organisation?

What is AI governance, and what does it mean for everyday AI use?

What is shadow AI, and why does a ban not solve it?

Would you rather someone showed you?

Why does this weigh more heavily in a municipality?

Which ChatGPT are you actually using?

What comes out, per task?

What if the name is already gone?

How do you see it at the moment itself?